devflow_core/agent_result.rs
1//! Agent completion detection — parses DEVFLOW_RESULT markers and evaluates
2//! exit codes to determine whether a coding agent succeeded or failed.
3//!
4//! Four-layer decision engine:
5//! 0. Run operator-authored external post-condition probes (authoritative failure)
6//! 1. Parse DEVFLOW_RESULT from agent stdout (authoritative for ordinary plans)
7//! 2. Exit code + commit count gate (reliable fallback)
8//! 3. Process gone + commits exist (last resort warning)
9
10use crate::config::GitFlowConfig;
11use crate::git::git_command;
12use crate::stage::Stage;
13use crate::state::State;
14use std::path::{Path, PathBuf};
15
16/// Parsed agent completion result.
17#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
18pub struct AgentResult {
19 pub status: AgentStatus,
20 pub exit_code: Option<i32>,
21 pub reason: Option<String>,
22 pub commits: Option<u32>,
23 pub summary: Option<String>,
24 /// The Validate stage's self-reported verdict — distinct from `status`.
25 /// `status` reports whether the stage's task (running `/gsd-validate-phase`)
26 /// completed; `verdict` reports whether validation ITSELF passed. Only
27 /// `Some(Verdict::Pass)` should advance Validate to Ship; `Some(Verdict::Gaps)`
28 /// and `None` both gate/loop back to Code (see `advance()`'s Validate arm).
29 /// Ignored entirely for non-Validate stages.
30 ///
31 /// Deserialized leniently via [`deserialize_verdict_lenient`]: an absent,
32 /// unknown, or mis-cased value becomes `None` rather than failing the
33 /// whole `AgentResult` parse (T-13-14) — a malformed verdict must never
34 /// silently drop a valid `status` to Layer 2.
35 #[serde(default, deserialize_with = "deserialize_verdict_lenient")]
36 pub verdict: Option<Verdict>,
37 /// Which evaluation layer (0-3) produced this result (D-10, 17-01). Set by
38 /// every constructor in this module; `None` is reserved for test-only
39 /// fixture literals that don't route through the real cascade.
40 #[serde(default)]
41 pub decided_by_layer: Option<u8>,
42}
43
44/// Agent completion status determined by DevFlow.
45#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
46#[serde(rename_all = "lowercase")]
47pub enum AgentStatus {
48 /// Agent self-reported success via DEVFLOW_RESULT.
49 Success,
50 /// Agent self-reported failure, or exit code + commit gate indicated failure.
51 Failed,
52 /// Agent stopped because an upstream API or usage quota rate-limited it.
53 RateLimited,
54 /// No signal received — fallback to exit code / commit heuristic.
55 Unknown,
56 /// Layer 2 classified the process as killed for resource exhaustion
57 /// (exit code 137, typically SIGKILL from an OOM killer) (D-07, 17b).
58 #[serde(rename = "resource_killed")]
59 ResourceKilled,
60 /// Layer 2 classified the process as unable to start (exit code 127,
61 /// typically "command not found") (D-07, 17b).
62 #[serde(rename = "agent_unavailable")]
63 AgentUnavailable,
64 /// The pipe-owning monitor gave up waiting: the child's stream went silent
65 /// for longer than the idle window and DevFlow terminated it (D-06, 31-02).
66 ///
67 /// Deliberately distinct from BOTH neighbours it would otherwise collapse
68 /// into. Against `Failed`: nothing reported a failure — the agent simply
69 /// stopped talking, and a graceful close would fall through to Layer 2,
70 /// which scores partial commits as `Success` (999.64 reborn inside its own
71 /// fix). Against `ResourceKilled`: the box did not run out of memory;
72 /// DevFlow itself did the killing. Only a third variant lets the completion
73 /// oracle tell "we gave up waiting" from either.
74 ///
75 /// The explicit `#[serde(rename)]` is required, not stylistic: the
76 /// enum-level `rename_all = "lowercase"` would collapse the two words into
77 /// `idletimeout`. The two existing two-word variants above carry the same
78 /// rename for the same reason.
79 #[serde(rename = "idle_timeout")]
80 IdleTimeout,
81}
82
83impl AgentStatus {
84 /// The wire-format name for this variant, pinned equal to
85 /// `serde_json::to_string(&self)` with the surrounding quotes stripped
86 /// (see the `as_wire_str_matches_serde_form` test). Exhaustive match with
87 /// NO wildcard arm — adding a variant without updating this is a compile
88 /// error. This is the sanctioned replacement for
89 /// `format!("{:?}", status).to_ascii_lowercase()`, which collapses word
90 /// boundaries on multi-word variants (review consensus #1).
91 pub fn as_wire_str(&self) -> &'static str {
92 match self {
93 AgentStatus::Success => "success",
94 AgentStatus::Failed => "failed",
95 AgentStatus::RateLimited => "ratelimited",
96 AgentStatus::Unknown => "unknown",
97 AgentStatus::ResourceKilled => "resource_killed",
98 AgentStatus::AgentUnavailable => "agent_unavailable",
99 AgentStatus::IdleTimeout => "idle_timeout",
100 }
101 }
102}
103
104/// The Validate stage's self-reported verdict (13b verdict-vs-ran split).
105#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
106#[serde(rename_all = "lowercase")]
107pub enum Verdict {
108 /// Validation found no gaps — ready to advance to Ship.
109 Pass,
110 /// Validation found gaps that still need fixing — must loop back to Code
111 /// (or gate, depending on the consecutive-failure threshold).
112 Gaps,
113}
114
115/// Deserialize `verdict` leniently: an absent, unknown, or mis-cased value
116/// (e.g. `"wat"`, `"Pass"`) becomes `Ok(None)` rather than an error, so a
117/// malformed verdict never fails the whole `from_str::<AgentResult>` parse
118/// and silently drops a valid `status` to Layer 2 (T-13-14, consensus #5).
119///
120/// Matching is intentionally exact-case (only the wire-format lowercase
121/// strings `"pass"`/`"gaps"` are accepted) — a mis-cased value like `"Pass"`
122/// is NOT case-folded into a match; it is treated the same as an unknown
123/// value and maps to `None`, so a subtly wrong-case verdict fails safe
124/// (gate/loop) instead of silently passing.
125///
126/// WR-09 (13-REVIEW.md): decodes as `serde_json::Value` first, then only
127/// pattern-matches the string case — a non-string JSON type (`true`, `123`,
128/// an object) is a wrong *type*, not a malformed string value, and must
129/// still fall through to `None` rather than erroring out the entire
130/// `AgentResult` parse (the same guarantee this deserializer already gives
131/// mis-cased/unknown string values).
132fn deserialize_verdict_lenient<'de, D>(deserializer: D) -> Result<Option<Verdict>, D::Error>
133where
134 D: serde::Deserializer<'de>,
135{
136 let raw = <Option<serde_json::Value> as serde::Deserialize>::deserialize(deserializer)?;
137 Ok(raw.and_then(|v| {
138 v.as_str().and_then(|s| match s {
139 "pass" => Some(Verdict::Pass),
140 "gaps" => Some(Verdict::Gaps),
141 _ => None,
142 })
143 }))
144}
145
146/// Errors produced by agent result evaluation.
147#[derive(Debug, thiserror::Error)]
148pub enum ResultError {
149 #[error("I/O error reading agent output: {0}")]
150 Io(#[from] std::io::Error),
151 #[error("phase directory not found")]
152 NoPhaseDir,
153}
154
155/// Search stdout for a DEVFLOW_RESULT marker.
156///
157/// The marker is a single line starting with `DEVFLOW_RESULT:` followed by
158/// a JSON object with at minimum a `status` field. Matching is case-insensitive.
159///
160/// When an agent is run with `--output-format json` (e.g. Claude), its final
161/// message is wrapped in a JSON result envelope with the text — and its
162/// embedded newlines — escaped inside a `result` field. In that case the
163/// marker never appears at the start of a line, so we first unwrap the
164/// envelope and search the inner text.
165pub fn parse_devflow_result(stdout: &str) -> Option<AgentResult> {
166 // normalise_stream_marker_provenance on BOTH arms: parse_marker_lines
167 // deserializes the agent's own JSON, so without the overwrite an agent
168 // writing `"decided_by_layer":0` into its marker forges Layer-0
169 // external-verification provenance, which `classify_validate_outcome`
170 // (pipeline_outcomes.rs) trusts when classifying a Validate stage. The
171 // stream path has normalised since 30-01; this generic path — the one
172 // production hits today — did not (fourth adversarial pass, Medium 1;
173 // the class 999.67 tracks).
174 if let Some(inner) = extract_json_result_text(stdout)
175 && let Some(result) = parse_marker_lines(&inner)
176 {
177 return Some(normalise_stream_marker_provenance(result));
178 }
179 parse_marker_lines(stdout).map(normalise_stream_marker_provenance)
180}
181
182/// Detect agent-specific rate-limit output and return the retry description.
183///
184/// Claude can emit a JSON result envelope when run with `--output-format json`;
185/// Codex commonly emits plain text such as "Try again at ...". This function is
186/// intentionally conservative so ordinary progress text does not become a
187/// false positive.
188pub fn detect_rate_limit(stdout: &str) -> Option<String> {
189 detect_claude_rate_limit(stdout).or_else(|| detect_codex_rate_limit(stdout))
190}
191
192fn detect_claude_rate_limit(stdout: &str) -> Option<String> {
193 // strip_corruption_padding, not trim(): this detector OUTRANKS the generic
194 // envelope-failure detector, and rate-limit envelopes carry `is_error:
195 // true`. When only the lower-precedence detector stripped edge corruption,
196 // one stray byte inverted the precedence — a RateLimited envelope (routes
197 // to auto-resume) decayed into a generic Failed (routes to review/gating).
198 // Fifth adversarial pass, Medium 1.
199 let value: serde_json::Value = serde_json::from_str(strip_corruption_padding(stdout)).ok()?;
200 let rate_limited = json_has_str(&value, "subtype", "error_rate_limit")
201 || json_has_i64(&value, "api_error_status", 429)
202 || json_has_i64(&value, "status", 429)
203 || json_has_i64(&value, "status_code", 429);
204 if !rate_limited {
205 return None;
206 }
207 json_find_key(&value, "retry_after")
208 .and_then(json_scalar_to_string)
209 .or_else(|| json_find_key(&value, "message").and_then(json_scalar_to_string))
210 .or_else(|| json_find_key(&value, "error").and_then(json_scalar_to_string))
211 .or_else(|| Some("usage limit".to_string()))
212}
213
214fn detect_codex_rate_limit(stdout: &str) -> Option<String> {
215 // This heuristic exists for Codex's PLAIN-TEXT output. JSONL event lines
216 // are authoritative and handled by parse_codex_event_result — scanning
217 // them here false-positives on document content echoed into events
218 // (13-06 dogfood finding: GSD reference tables mentioning "rate limiting"
219 // were read by the agent, echoed into an `item.completed` payload, and
220 // this scan returned that entire multi-KB line as the "retry time").
221 // The JSON-line exclusion applies the SAME edge-strip policy as
222 // ParsedCapture::parse (sixth-pass Medium 4): an event line whose leading
223 // byte was corrupted to U+FFFD failed the bare parse here and was treated
224 // as prose — re-admitting the exact multi-KB echoed-document false
225 // positive this filter exists to exclude, after ParsedCapture had already
226 // correctly recovered the line as an event.
227 let stdout: String = stdout
228 .lines()
229 .filter(|line| {
230 serde_json::from_str::<serde_json::Value>(strip_corruption_padding(line))
231 .map(|v| !v.is_object())
232 .unwrap_or(true)
233 })
234 .collect::<Vec<_>>()
235 .join("\n");
236 let stdout = stdout.as_str();
237 let lower = stdout.to_ascii_lowercase();
238 if let Some(idx) = lower.find("try again at ") {
239 let start = idx + "try again at ".len();
240 let retry = stdout[start..]
241 .lines()
242 .next()
243 .unwrap_or_default()
244 .trim()
245 .trim_end_matches(['.', ',', ';'])
246 .trim();
247 if !retry.is_empty() {
248 return Some(retry.to_string());
249 }
250 }
251
252 // "429" counts as rate-limit evidence only as a STANDALONE token
253 // (sixth-pass Medium 5): a bare substring check fired on "processed issue
254 // #429 successfully" and any number containing 429, routing a healthy run
255 // into auto-resume. A neighbor that is alphanumeric or '#' means the
256 // digits belong to something else.
257 fn standalone_429(line: &str) -> bool {
258 let bytes = line.as_bytes();
259 line.match_indices("429").any(|(i, _)| {
260 let before_ok = i == 0 || {
261 let b = bytes[i - 1];
262 !b.is_ascii_alphanumeric() && b != b'#'
263 };
264 let after_ok = i + 3 >= bytes.len() || !bytes[i + 3].is_ascii_alphanumeric();
265 before_ok && after_ok
266 })
267 }
268
269 if lower.contains("usage limit") || lower.contains("rate limit") || standalone_429(&lower) {
270 stdout
271 .lines()
272 .find(|line| {
273 let line = line.to_ascii_lowercase();
274 line.contains("usage limit") || line.contains("rate limit") || standalone_429(&line)
275 })
276 .map(str::trim)
277 .filter(|line| !line.is_empty())
278 .map(str::to_string)
279 .or_else(|| Some("usage limit".to_string()))
280 } else {
281 None
282 }
283}
284
285/// If `stdout` is a JSON result envelope, return the decoded `result` text
286/// field (with escapes such as `\n` resolved). Returns `None` for plain text.
287fn extract_json_result_text(stdout: &str) -> Option<String> {
288 // strip_corruption_padding, not trim(): a stray invalid byte decoded to
289 // U+FFFD at either EDGE of the envelope must not defeat the `{` guard
290 // (third-pass High). Interior corruption still fails the parse, by design.
291 let trimmed = strip_corruption_padding(stdout);
292 if !trimmed.starts_with('{') {
293 return None;
294 }
295 let value: serde_json::Value = serde_json::from_str(trimmed).ok()?;
296 value.get("result")?.as_str().map(str::to_string)
297}
298
299/// Read the top-level `session_id` string from a Claude JSON result envelope
300/// (`--output-format json`). Returns `None` for plain-text stdout, a
301/// non-JSON-object envelope, an envelope with no `session_id` key, or a
302/// `session_id` of a non-string JSON type — never panics.
303///
304/// D-04 / T-28-04 (this plan's `<threat_model>`): deliberately reads ONLY the
305/// envelope's TOP-LEVEL `session_id` key via a direct [`serde_json::Value::get`],
306/// never the module's [`json_find_key`]/[`json_scan`] traversal helpers. Those
307/// helpers descend into nested objects, and the agent-authored `DEVFLOW_RESULT`
308/// marker payload — embedded inside this same envelope's `result` text and
309/// deserialized by [`parse_marker_lines`] directly into [`AgentResult`] — is
310/// reachable that way. A top-level `get` makes it true BY CONSTRUCTION that an
311/// agent cannot redirect the session DevFlow later resumes into by planting a
312/// different `session_id` key inside its own self-authored marker JSON.
313/// Regression test: `session_id_in_devflow_result_marker_is_not_returned`.
314///
315/// Deliberate deviation from RESEARCH.md § "Discretion Resolutions" item 5,
316/// which suggested adding a `session_id` field directly to [`AgentResult`].
317/// NOT done: `parse_marker_lines` deserializes the agent's own
318/// `DEVFLOW_RESULT` JSON straight into `AgentResult` via `serde_json::from_str`,
319/// so a `#[serde(default)]` field there would be agent-settable — the agent
320/// could name the session DevFlow resumes into (T-28-04). A standalone reader
321/// over the top-level envelope key carries no such surface and is equally
322/// available to every caller; D-04's persistence target (`State::session_id`)
323/// is unchanged, only the carrier differs.
324pub fn claude_session_id(stdout: &str) -> Option<String> {
325 // strip_corruption_padding, not trim(): a stray invalid byte decoded to
326 // U+FFFD at either EDGE of the envelope must not defeat the `{` guard
327 // (third-pass High). Interior corruption still fails the parse, by design.
328 let trimmed = strip_corruption_padding(stdout);
329 if !trimmed.starts_with('{') {
330 return None;
331 }
332 let value: serde_json::Value = serde_json::from_str(trimmed).ok()?;
333 value.get("session_id")?.as_str().map(str::to_string)
334}
335
336/// Read the CLI-emitted `session_id` from a Claude `--output-format
337/// stream-json` JSONL capture: the top-level `session_id` of the LAST
338/// `system`/`init` event. `None` for any other capture shape.
339///
340/// The stream sibling of [`claude_session_id`], and it carries that function's
341/// D-04 / T-28-04 discipline **for the same reason** — read its doc comment
342/// before changing anything here. Only the event's TOP-LEVEL `session_id` is
343/// read, via a direct [`serde_json::Value::get`]; the
344/// [`json_find_key`]/[`json_scan`] traversal helpers are NOT to be used. They
345/// descend into nested objects, and a stream carries agent-authored text in
346/// every `result` event — including the `DEVFLOW_RESULT` marker JSON that
347/// [`parse_marker_lines`] deserializes. A traversal would make a `session_id`
348/// the agent planted in its own marker reachable, handing it the ability to
349/// name the session DevFlow later resumes into (T-30-11). Regression test:
350/// `claude_stream_session_id_ignores_agent_planted_value`.
351///
352/// The LAST `init` event wins, consistent with the last-`result`-wins
353/// convention. Verified against the archived capture: its three `init` events
354/// (lines 5, 32 and 47) all carry the same `session_id`, so last-wins and
355/// first-wins agree on today's evidence — but only last-wins stays correct if a
356/// future capture rotates the value mid-stream. Three `init` events do NOT mean
357/// three sessions: session continuity must never be keyed off "have I seen an
358/// `init` event".
359///
360/// No `session_id` field is added to [`AgentResult`] — see
361/// [`claude_session_id`]'s doc comment for why that design stays rejected.
362pub fn claude_stream_session_id(stdout: &str) -> Option<String> {
363 let capture = ParsedCapture::parse(stdout);
364 if classify(&capture) != CaptureKind::ClaudeStream {
365 return None;
366 }
367
368 // A session can rotate mid-capture: each turn opens with its own `init`, and
369 // the LAST one carries the id a resume must target. A torn later `init` is
370 // invisible to the scan below, which would silently return an EARLIER
371 // session's id — resuming the wrong session with a token that looks
372 // perfectly valid. Fail closed on any TORN JSON line: it could have been a
373 // newer `init`. `None` costs a resume; the wrong id corrupts one. (Third
374 // adversarial pass, 2026-08-02.)
375 //
376 // Prose noise lines do NOT block recovery — an `init` is a JSON line, so a
377 // non-`{` line can never be a torn one. The first version of this guard
378 // failed closed on ANY unparsed line and rejected captures with benign
379 // interleaved progress output (fourth adversarial pass, Medium 3).
380 if capture.torn_json_line_present() {
381 return None;
382 }
383
384 capture
385 .events
386 .iter()
387 .rev()
388 .find(|v| {
389 v.get("type").and_then(serde_json::Value::as_str) == Some("system")
390 && v.get("subtype").and_then(serde_json::Value::as_str) == Some("init")
391 })?
392 .get("session_id")?
393 .as_str()
394 .map(str::to_string)
395}
396
397/// Thin file-reading wrapper over the two session-id readers: reads the phase's
398/// captured stdout file (via [`stdout_path`]) and delegates. `None` for a
399/// missing capture file, never an `Err` — mirrors [`evaluate_layer1`]'s
400/// lossy-read convention (CR-01: one invalid UTF-8 byte from raw `sh`
401/// redirection must not silently disable this reader).
402///
403/// [`claude_stream_session_id`] is tried FIRST, then [`claude_session_id`].
404/// Stream-first is safe and behavior-preserving: the stream gate
405/// ([`is_claude_event_stream`]) declines a single-document envelope, so every
406/// capture shape that ships today still resolves through `claude_session_id`
407/// bit-for-bit. Without this chain the Phase 28 checkpoint-resume path — whose
408/// whole delivery is reconstructing a session via `claude --resume` — returns
409/// `None` for every `stream-json` capture.
410pub fn session_id_from_capture(project_root: &Path, phase: u32) -> Option<String> {
411 let stdout = read_capture(&stdout_path(project_root, phase))?;
412 claude_stream_session_id(&stdout).or_else(|| claude_session_id(&stdout))
413}
414
415/// The ONE decode policy for capture files: read the bytes and replace invalid
416/// UTF-8 with U+FFFD. Every capture-file consumer (`evaluate_layer1`,
417/// `checkpoint_reported_in_capture`, `session_id_from_capture`) reads through
418/// here, so the policy cannot silently diverge per call site again.
419///
420/// REPLACE, never drop. A drop-based decode was tried (third adversarial pass
421/// remediation) and refuted by the fourth pass: deleting invalid bytes JOINS
422/// the tokens on either side, and `DEVFLOW_RESULT: {"status":"suc<FF>cess"}`
423/// decoded to a fabricated, VALID success marker that short-circuited a
424/// nonzero exit code. Replacement keeps corruption visible: the marker parser
425/// sees `suc\u{FFFD}cess`, which is not a recognized status, and correctly
426/// refuses to trust it. Consumers that need to tolerate corruption at the
427/// EDGES of a single-document capture strip it explicitly via
428/// [`strip_corruption_padding`] — bounded, and incapable of joining tokens.
429fn read_capture(path: &Path) -> Option<String> {
430 let bytes = std::fs::read(path).ok()?;
431 Some(String::from_utf8_lossy(&bytes).into_owned())
432}
433
434/// Trim whitespace and U+FFFD replacement characters from both ends of a
435/// single-document capture.
436///
437/// U+FFFD is what [`read_capture`] substitutes for invalid bytes, and it is a
438/// printing, non-whitespace character — so a stray byte written before or after
439/// the JSON envelope survives `trim()` and defeats every `starts_with('{')`
440/// guard. That was the third pass's High: Layer 1 abstained on an authoritative
441/// `is_error: true` and the exit-code fallback turned a reported failure into a
442/// Ship-gate success. Stripping only the EDGES is deliberate: corruption inside
443/// the envelope must stay visible and fail the parse, because "repairing" it is
444/// how the fourth pass's marker-fabrication High happened.
445fn strip_corruption_padding(s: &str) -> &str {
446 s.trim_matches(|c: char| c.is_whitespace() || c == '\u{FFFD}')
447}
448
449// WR-12 (13-REVIEW.md), revised: these traversal helpers run on the coding
450// agent's raw stdout (via detect_claude_rate_limit, which every `devflow
451// advance` invocation runs through evaluate_layer1), so deeply nested JSON —
452// accidental or adversarial — must not stack-overflow the process. The
453// traversal is iterative (an explicit worklist), so nesting depth never
454// consumes call stack and no depth cap is needed. The first WR-12 fix capped
455// recursion at 64, which silently missed keys at depths 64–128 — nesting
456// serde_json's default 128-level parse recursion limit (the only producer of
457// these `Value`s) accepts just fine.
458
459/// Depth-first pre-order scan over every JSON object in `value`, returning
460/// the first `Some` produced by `visit` on an object's map.
461fn json_scan<'a, T>(
462 value: &'a serde_json::Value,
463 visit: impl Fn(&'a serde_json::Map<String, serde_json::Value>) -> Option<T>,
464) -> Option<T> {
465 let mut stack = vec![value];
466 while let Some(current) = stack.pop() {
467 match current {
468 serde_json::Value::Object(map) => {
469 if let Some(found) = visit(map) {
470 return Some(found);
471 }
472 // Push in reverse so pop order preserves document order.
473 for child in map.values().rev() {
474 stack.push(child);
475 }
476 }
477 serde_json::Value::Array(values) => {
478 for child in values.iter().rev() {
479 stack.push(child);
480 }
481 }
482 _ => {}
483 }
484 }
485 None
486}
487
488fn json_has_str(value: &serde_json::Value, key: &str, expected: &str) -> bool {
489 json_scan(value, |map| {
490 (map.get(key)?.as_str()? == expected).then_some(())
491 })
492 .is_some()
493}
494
495fn json_has_i64(value: &serde_json::Value, key: &str, expected: i64) -> bool {
496 json_scan(value, |map| {
497 (map.get(key)?.as_i64()? == expected).then_some(())
498 })
499 .is_some()
500}
501
502fn json_find_key<'a>(value: &'a serde_json::Value, key: &str) -> Option<&'a serde_json::Value> {
503 json_scan(value, |map| map.get(key))
504}
505
506fn json_scalar_to_string(value: &serde_json::Value) -> Option<String> {
507 match value {
508 serde_json::Value::String(s) => Some(s.clone()),
509 serde_json::Value::Number(n) => Some(n.to_string()),
510 _ => None,
511 }
512}
513
514/// Read the top-level `is_error` boolean (and, if present, `num_turns`) from
515/// a Claude JSON result envelope (`--output-format json`) and treat
516/// `is_error: true` as an authoritative Layer-1 failure.
517///
518/// This is checked BEFORE the `DEVFLOW_RESULT` marker path in
519/// [`evaluate_layer1`], so `is_error: true` OVERRIDES a stale/echoed success
520/// marker embedded in the same envelope's `result` text — the envelope is
521/// authoritative for errors. `is_error` absent or `false` returns `None`,
522/// deferring to the marker path and, ultimately, Layer 2. It runs AFTER
523/// `detect_claude_rate_limit`, though: rate-limit envelopes also carry
524/// `is_error: true`, and the specific `RateLimited` classification (which
525/// drives the primary rate-limit resume cron) must win over this
526/// generic `Failed`.
527///
528/// Per RESEARCH Pitfall 5, `is_error` (not specific `subtype` strings) is
529/// the documented, stable signal — this does not special-case non-success
530/// subtype values beyond what already exists in `detect_claude_rate_limit`.
531fn detect_claude_envelope_failure(stdout: &str) -> Option<AgentResult> {
532 // strip_corruption_padding, not trim(): a stray invalid byte decoded to
533 // U+FFFD at either EDGE of the envelope must not defeat the `{` guard
534 // (third-pass High). Interior corruption still fails the parse, by design.
535 let trimmed = strip_corruption_padding(stdout);
536 if !trimmed.starts_with('{') {
537 return None;
538 }
539 let value: serde_json::Value = serde_json::from_str(trimmed).ok()?;
540 let is_error = value.get("is_error")?.as_bool()?;
541 if !is_error {
542 return None;
543 }
544
545 let num_turns = value.get("num_turns").and_then(serde_json::Value::as_u64);
546 let base_reason = value
547 .get("result")
548 .and_then(serde_json::Value::as_str)
549 .map(str::to_string)
550 .or_else(|| {
551 value
552 .get("subtype")
553 .and_then(serde_json::Value::as_str)
554 .map(str::to_string)
555 })
556 .unwrap_or_else(|| "agent reported is_error".to_string());
557 let reason = match num_turns {
558 Some(n) => format!("{base_reason} (num_turns: {n})"),
559 None => base_reason,
560 };
561
562 Some(AgentResult {
563 status: AgentStatus::Failed,
564 exit_code: None,
565 reason: Some(reason),
566 commits: None,
567 summary: None,
568 verdict: None,
569 decided_by_layer: Some(1),
570 })
571}
572
573/// The rendered VALUE of a human-blocking checkpoint's `**Gate:**` line.
574///
575/// **CONFIRMED against a live end-to-end run (2026-07-31).** Assumption A1 is
576/// closed. A real `devflow start` run drove a synthetic phase declaring a
577/// `gate="blocking-human"` task through DevFlow's own monitor process (not a
578/// Claude Code agent session, which is what blocked `28-PROBE.md`'s original
579/// attempt at the Bash-tool permission classifier). The checkpoint fired and
580/// `.devflow/phase-NN-stdout` captured it inside the JSON envelope's `result`
581/// text as:
582///
583/// ```text
584/// **Gate:** `blocking-human`
585/// ```
586///
587/// The VALUE is what this constant holds. The surrounding markdown — bold
588/// label, and a **code span around the value** — is handled by
589/// [`text_reports_human_gate`]'s trim set, not by this constant.
590///
591/// The code span is the part RESEARCH.md did not predict. Its § "Architecture
592/// Patterns / Pattern 2" derived the literal by reading the *emitting* source
593/// (`gsd-executor.md:356`, `execute-phase.md:1053`) and predicted a bare
594/// `**Gate:** blocking-human`. The real relay renders the value as a code
595/// span, which defeated the original matcher entirely — see
596/// [`text_reports_human_gate`] for that failure and its fix. Lesson worth
597/// keeping: the emitting source told us the value, not the rendering.
598const HUMAN_GATE_VALUE: &str = "blocking-human";
599
600/// Confirm whether captured stdout reports a human-blocking checkpoint, by
601/// searching for a `**Gate:**`-labeled line whose VALUE is exactly
602/// [`HUMAN_GATE_VALUE`] — see that constant's doc comment for the live
603/// observation (2026-07-31) the matched rendering is built from.
604///
605/// This is the CONFIRMATION half of D-01: it is only ever consulted AFTER
606/// [`crate::verify::phase_has_blocking_human_checkpoint`] has already
607/// returned `true` for the stage's plan(s) (D-01's static half, plan 28-01).
608/// A false negative here is the SAFE direction — it falls back to today's
609/// never-silent generic gate, losing nothing. A false positive is bounded by
610/// the resume ceiling (`mode::MAX_CHECKPOINT_RESUMES`, plan 28-03) and
611/// unconditionally recorded by the `checkpoint_auto_decided` audit event
612/// (plan 28-03) — it can never silently authorize anything.
613///
614/// Searches BOTH the raw stdout text and — when the stdout is a Claude JSON
615/// result envelope — the unescaped inner `result` text obtained via
616/// [`extract_json_result_text`], because the `Gate:` line typically crosses
617/// into the capture escaped inside that envelope (RESEARCH § "Common
618/// Pitfalls / Pitfall 2": two indirections, subagent emission → orchestrator
619/// relay → DevFlow's captured top-level stdout). Matching is
620/// case-insensitive on the `Gate` LABEL and tolerates surrounding markdown
621/// emphasis (`*`) and whitespace, but the VALUE comparison is exact — this
622/// deliberately does NOT widen into a general "does this look like a
623/// checkpoint" heuristic (D-02 rejected that class of predicate); the scope
624/// is one declared field label with one enumerated value.
625///
626/// **A Claude `stream-json` capture takes a separate branch** and is answered
627/// by [`claude_stream_reports_human_gate`] ALONE — it never consults raw stdout.
628/// That is not an oversight to be "completed" later: under a stream capture the
629/// raw stdout contains the operator's prompt echoed back as a `user` event, so
630/// also scanning it would reinstate the exact false positive the branch exists
631/// to remove (review constraint 3 — the unbounded raw scan is the reader that
632/// "survives by accident" once the single-document invariant is gone). See that
633/// function for which events are eligible and why.
634///
635/// The branch is taken when [`classify`] says [`CaptureKind::ClaudeStream`], so
636/// a single-document envelope, plain text and a Codex stream all fall through to
637/// the two-target logic below, unchanged (T-30-25). Classification is
638/// deliberately weaker than [`is_claude_event_stream`]: requiring a parsed
639/// `system`/`init` here made a single torn line fail OPEN back to the raw scan,
640/// reinstating the echoed-prompt false positive this branch exists to remove.
641/// See [`classify`] for the full rule set and the defects each rule encodes;
642/// see [`is_claude_event_stream`] for why the verdict path keeps its stricter
643/// init-only gate.
644pub fn blocking_human_checkpoint_reported(stdout: &str) -> bool {
645 let capture = ParsedCapture::parse(stdout);
646 if classify(&capture) == CaptureKind::ClaudeStream {
647 return claude_stream_reports_human_gate(&capture.events);
648 }
649 if text_reports_human_gate(stdout) {
650 return true;
651 }
652 extract_json_result_text(stdout)
653 .as_deref()
654 .is_some_and(text_reports_human_gate)
655}
656
657/// Core matcher shared by both search targets (raw stdout and the unescaped
658/// inner envelope text) in [`blocking_human_checkpoint_reported`]. Scans for
659/// a case-insensitive `gate` label, tolerating surrounding markdown emphasis
660/// (`*`), code-span backticks (`` ` ``), and whitespace up to the following
661/// `:`, then compares the VALUE token immediately after the colon exactly
662/// against [`HUMAN_GATE_VALUE`].
663///
664/// The backtick tolerance is not speculative — it is the single reason this
665/// matcher failed against the first real checkpoint ever observed. The live
666/// A1 run (2026-07-31) captured the value as a markdown code span,
667/// ``**Gate:** `blocking-human` ``, and the original trim set (`*` and space
668/// only) left the leading backtick in place, so the `take_while` below
669/// terminated immediately and produced an EMPTY value token. The reader
670/// returned `false` and a genuine checkpoint fell through to the generic
671/// gate. Trimming the backtick is what makes the observed rendering match;
672/// do not narrow this set back without re-running that live probe.
673///
674/// Note the closing backtick needs no handling: `take_while` already stops
675/// at it, since a backtick is neither alphanumeric nor `-`.
676fn text_reports_human_gate(text: &str) -> bool {
677 let lower = text.to_ascii_lowercase();
678 let mut search_from = 0;
679 while let Some(rel_idx) = lower[search_from..].find("gate") {
680 let idx = search_from + rel_idx;
681 let after_label = &lower[idx + "gate".len()..];
682 let after_label = after_label.trim_start_matches(['*', ' ', '`']);
683 if let Some(rest) = after_label.strip_prefix(':') {
684 let value_region = rest.trim_start_matches(['*', ' ', '`']);
685 let value_token: String = value_region
686 .chars()
687 .take_while(|c| c.is_ascii_alphanumeric() || *c == '-')
688 .collect();
689 if value_token == HUMAN_GATE_VALUE {
690 return true;
691 }
692 }
693 search_from = idx + "gate".len();
694 }
695 false
696}
697
698/// Thin file-reading wrapper over [`blocking_human_checkpoint_reported`]:
699/// reads the phase's captured stdout file (via [`stdout_path`]) and
700/// delegates. `false` for a missing capture file, never an error.
701pub fn checkpoint_reported_in_capture(project_root: &Path, phase: u32) -> bool {
702 let Some(stdout) = read_capture(&stdout_path(project_root, phase)) else {
703 return false;
704 };
705 blocking_human_checkpoint_reported(&stdout)
706}
707
708/// Determine whether a set of parsed JSONL lines look like a Codex `--json`
709/// event stream (as opposed to a single-document Claude envelope or plain
710/// text) — i.e. at least one line is a `thread.started` or `turn.*` event.
711fn is_codex_event_stream(events: &[serde_json::Value]) -> bool {
712 events.iter().any(|v| {
713 v.get("type")
714 .and_then(serde_json::Value::as_str)
715 .is_some_and(|t| t == "thread.started" || t.starts_with("turn."))
716 })
717}
718
719/// Parse a Codex `--json` JSONL event stream (one JSON object per line) and
720/// look at the LAST terminal event (`turn.completed` / `turn.failed`).
721///
722/// Only decisive when the captured stdout is actually a Codex event stream
723/// (per [`is_codex_event_stream`]) — a single-document Claude envelope
724/// (`type: "result"`, no `turn.*` lines) is not consumed here and returns
725/// `None`, so the Claude envelope/marker paths handle it instead.
726///
727/// `turn.failed` is decisive: returns `AgentStatus::Failed` with `reason`
728/// from `error.message`. A final `turn.completed` with no `DEVFLOW_RESULT`
729/// marker returns `None` (defers to Layer 2) rather than an unconditional
730/// Success — a marker-less turn must not silently advance a stage (this is
731/// the composition fix that keeps a marker-less Validate run from
732/// false-passing to Ship).
733///
734/// NOTE: written against the documented `--json` event schema (thread.started
735/// / turn.started / item.* / turn.completed with usage / turn.failed with
736/// error.message) but not yet verified against the installed Codex CLI
737/// version — the 13-06 dogfood run captures real output and reconciles any
738/// delta, the same empirical practice 12-12-SUMMARY.md used for Claude.
739fn parse_codex_event_result(stdout: &str) -> Option<AgentResult> {
740 let capture = ParsedCapture::parse(stdout);
741 let events = &capture.events;
742
743 if !is_codex_event_stream(events) {
744 return None;
745 }
746
747 // Same trailing-torn rule as the Claude stream parser, same R1 root cause:
748 // a torn JSON line after the last parsed event means the capture's tail —
749 // where `turn.failed` would be — may be among the casualties. An earlier
750 // `agent_message` success marker must not decide the stage over a tail we
751 // provably failed to read. The Codex adapter is live in production, so
752 // this is not a Phase-31 deferral.
753 if capture.torn_json_after_last_matching(|_| true) {
754 return Some(indeterminate_capture_failure());
755 }
756
757 // Codex delivers the agent's DEVFLOW_RESULT self-report inside an
758 // `agent_message` item's `text` — never as a raw stdout line — so the
759 // top-level marker scan cannot see it (13-06 dogfood finding: a Codex
760 // `DEVFLOW_RESULT: failed` was invisible and the run fell through to
761 // heuristics). The decoded `text` is a plain marker line; reuse the
762 // marker parser on it. Last marker wins, matching parse_marker_lines.
763 let marker = events.iter().rev().find_map(|v| {
764 if v.get("type").and_then(serde_json::Value::as_str) != Some("item.completed") {
765 return None;
766 }
767 let item = v.get("item")?;
768 if item.get("type").and_then(serde_json::Value::as_str) != Some("agent_message") {
769 return None;
770 }
771 let text = item.get("text").and_then(serde_json::Value::as_str)?;
772 parse_marker_lines(text)
773 });
774 if let Some(result) = marker {
775 // Same provenance overwrite as parse_devflow_result and the Claude
776 // stream path (T-30-26): this AgentResult was deserialized from the
777 // agent's own marker JSON, so a planted `"decided_by_layer":0` would
778 // otherwise forge Layer-0 external-verification provenance. Found by
779 // reading, while closing the identical hole one function over.
780 return Some(normalise_stream_marker_provenance(result));
781 }
782
783 let terminal = events.iter().rev().find(|v| {
784 matches!(
785 v.get("type").and_then(serde_json::Value::as_str),
786 Some("turn.completed") | Some("turn.failed")
787 )
788 })?;
789
790 if terminal.get("type").and_then(serde_json::Value::as_str) != Some("turn.failed") {
791 // turn.completed (or any other terminal we don't recognize) defers
792 // to Layer 2 rather than an unconditional Success.
793 return None;
794 }
795
796 let reason = terminal
797 .get("error")
798 .and_then(|e| e.get("message"))
799 .and_then(serde_json::Value::as_str)
800 .map(str::to_string)
801 .unwrap_or_else(|| "codex turn failed".to_string());
802
803 Some(AgentResult {
804 status: AgentStatus::Failed,
805 exit_code: None,
806 reason: Some(reason),
807 commits: None,
808 summary: None,
809 verdict: None,
810 decided_by_layer: Some(1),
811 })
812}
813
814/// Parse a captured stdout as JSONL: one `serde_json::Value` per non-blank,
815/// parseable line. Lines that are not valid JSON are dropped, so a stream
816/// interleaved with plain-text progress noise still yields its events.
817///
818/// Shared by [`is_claude_event_stream`] and [`last_top_level_result`], which
819/// both need the same parsed vector. Deliberately NOT retrofitted into
820/// [`parse_codex_event_result`], which open-codes the identical idiom: that
821/// parser is correct and shipping, and rewriting it would put an unrelated
822/// adapter's behavior at risk for a cosmetic dedupe.
823/// Determine whether parsed JSONL lines are a Claude `--output-format
824/// stream-json` event stream, as opposed to a single-document Claude envelope,
825/// a Codex `--json` stream, or plain text.
826///
827/// **Gates on `type: "system"` + `subtype: "init"` and NOTHING ELSE.**
828/// 30-RESEARCH.md offered an alternative — also gate on `type: "result"`
829/// carrying a `session_id` — and that alternative is WRONG; do not "restore"
830/// it. The single-document envelope that ships today is literally
831/// `{"type":"result",...,"session_id":"abc"}`, so a `result`-keyed gate would
832/// swallow every production capture in use and silently displace
833/// [`parse_devflow_result`] in the [`evaluate_layer1`] cascade — a change to
834/// the shipped Layer-1 verdict path, disguised as adding stream support
835/// (T-30-02). The `init` event is both stronger and earlier: it opens the
836/// stream and is present in all three archived captures
837/// (`30a-evidence/raw_output_v3.jsonl` lines 5, 32 and 47).
838///
839/// `single_doc_envelope_not_consumed_by_claude_stream_parser` is the test that
840/// fails if this gate is widened.
841fn is_claude_event_stream(events: &[serde_json::Value]) -> bool {
842 events.iter().any(|v| {
843 v.get("type").and_then(serde_json::Value::as_str) == Some("system")
844 && v.get("subtype").and_then(serde_json::Value::as_str) == Some("init")
845 })
846}
847
848/// The shape of one non-empty capture line after a parse attempt.
849///
850/// `TornJson` vs `Noise` is the load-bearing distinction everywhere below: a
851/// line that failed to parse but still opens with `{` could be a torn event —
852/// a truncated write, or a read of a capture still being appended to — while a
853/// prose line cannot be (every stream event line opens with `{`). Conflating
854/// the two produced both prior misclassification defects: requiring ALL lines
855/// to parse sent torn streams back to the raw scan (second-pass fail-open),
856/// and counting any malformed line as suspicious rejected benign interleaved
857/// progress noise (fourth-pass Medium 3).
858#[derive(Clone, Copy, PartialEq, Eq)]
859enum LineShape {
860 /// Parsed as JSON; the value lives at the same index in
861 /// [`ParsedCapture::events`]' insertion order.
862 Event,
863 /// Failed to parse but opens with `{` — potentially a torn event.
864 TornJson,
865 /// Failed to parse and does not open with `{` — cannot be a torn event.
866 Noise,
867}
868
869/// A capture parsed ONCE, keeping both the surviving events and the shape of
870/// every non-empty line — including the ones that did not parse.
871///
872/// This is the R1 root-cause fix from the phase-30 adversarial series: the old
873/// `claude_stream_events` returned a bare `Vec<Value>`, so "I dropped
874/// something" was unrepresentable and every consumer silently assumed the
875/// survivors were complete. Four separate defects came from that assumption
876/// (torn-init gate fail-open, stale-success verdict resurrection, stale
877/// session-id resurrection, torn-user gate reopening). Consumers now see the
878/// full line record and must decide explicitly what a torn line means for them.
879struct ParsedCapture {
880 events: Vec<serde_json::Value>,
881 line_shapes: Vec<LineShape>,
882}
883
884impl ParsedCapture {
885 fn parse(stdout: &str) -> Self {
886 let mut events = Vec::new();
887 let mut line_shapes = Vec::new();
888 for line in stdout.lines() {
889 let trimmed = line.trim();
890 if trimmed.is_empty() {
891 continue;
892 }
893 match serde_json::from_str::<serde_json::Value>(trimmed) {
894 Ok(v) => {
895 events.push(v);
896 line_shapes.push(LineShape::Event);
897 }
898 Err(_) => {
899 // Apply the SAME edge-corruption policy per line that
900 // strip_corruption_padding applies per capture. Without
901 // this, `read_capture`'s U+FFFD replacement in front of an
902 // otherwise-intact line made it classify as Noise — not
903 // `{`-prefixed — so the torn-tail guard could not see a
904 // corrupt superseding event and an earlier success marker
905 // decided the stage (fifth adversarial pass, High 1).
906 //
907 // Retry the parse on the stripped line first: edge
908 // corruption around an intact event RECOVERS the event and
909 // its true verdict. Stripping edges cannot join tokens —
910 // the fabrication hazard was DROPPING bytes inside content
911 // (fourth pass) — and interior corruption still fails the
912 // parse. A line that strips to empty was pure corruption:
913 // torn, fail closed.
914 let stripped = strip_corruption_padding(trimmed);
915 if stripped != trimmed
916 && let Ok(v) = serde_json::from_str::<serde_json::Value>(stripped)
917 {
918 events.push(v);
919 line_shapes.push(LineShape::Event);
920 } else {
921 line_shapes.push(if stripped.starts_with('{') || stripped.is_empty() {
922 LineShape::TornJson
923 } else {
924 LineShape::Noise
925 });
926 }
927 }
928 }
929 }
930 Self {
931 events,
932 line_shapes,
933 }
934 }
935
936 fn torn_json_line_present(&self) -> bool {
937 self.line_shapes.contains(&LineShape::TornJson)
938 }
939
940 /// Whether a torn JSON line sits AFTER the last parsed event matching
941 /// `pred` — or anywhere at all, when no event matches.
942 ///
943 /// This is the question behind constraint 9 item 1: the capture's REAL
944 /// final verdict may be among the casualties, so nothing that survives
945 /// before the tear is allowed to stand in for it. Prose noise lines are
946 /// not counted — they cannot be a torn event (events open with `{`).
947 fn torn_json_after_last_matching(&self, pred: impl Fn(&serde_json::Value) -> bool) -> bool {
948 let mut last_match_line = None;
949 let mut event_idx = 0usize;
950 for (line_idx, shape) in self.line_shapes.iter().enumerate() {
951 if *shape == LineShape::Event {
952 if pred(&self.events[event_idx]) {
953 last_match_line = Some(line_idx);
954 }
955 event_idx += 1;
956 }
957 }
958 self.line_shapes
959 .iter()
960 .enumerate()
961 .any(|(line_idx, shape)| {
962 *shape == LineShape::TornJson && last_match_line.is_none_or(|last| line_idx > last)
963 })
964 }
965}
966
967/// What kind of capture this is — decided ONCE, here, instead of re-derived by
968/// per-call-site heuristics.
969///
970/// This is the R2 root-cause fix from the phase-30 adversarial series. Four
971/// generations of ad-hoc shape checks (`starts_with('{')` guards, "any event of
972/// type X", all-lines-JSON, line counts) each got one case wrong: a torn `init`
973/// un-recognised a stream (fail-open), one stray JSON line hijacked plain text
974/// (V-01, fail-closed), a torn gate-bearing `user` event un-recognised a stream
975/// again, and an interleaved prose line was treated as tearing. One classifier
976/// carries all of those lessons in one place.
977#[derive(Clone, Copy, PartialEq, Eq)]
978enum CaptureKind {
979 /// Not JSONL-shaped in the majority — the raw-scan paths own it.
980 PlainText,
981 /// Exactly one parsed `{"type":"result",…}` line: the envelope the shipped
982 /// `--output-format json` adapter emits (T-30-25). Raw-scan paths own it.
983 SingleDocEnvelope,
984 /// A Claude `stream-json` capture — possibly torn, possibly noisy.
985 ClaudeStream,
986 /// A Codex `--json` capture: dotted top-level types (`thread.started`,
987 /// `item.completed`, `turn.*`). Raw-scan paths own it, as before.
988 CodexStream,
989}
990
991/// Classification rules, in order — each carries the defect that forced it:
992///
993/// 1. **Majority of non-empty lines must be JSON-shaped** (parsed OR torn-`{`),
994/// else `PlainText`. Counting only PARSED lines fails: truncating a real
995/// stream drops its parsed count below any threshold while every surviving
996/// line is still `{`-shaped (the truncation sweep caught exactly that). One
997/// stray JSON line in prose stays under the majority (V-01).
998/// 2. **Any parsed `system`/`user`/`assistant` event → `ClaudeStream`.** Claude
999/// types win over dotted deterministically — the old event loop returned
1000/// whichever it happened to iterate first. Real Codex captures never carry
1001/// these types, and on a corrupt mixed capture the scoped path is the
1002/// fail-closed direction for the gate.
1003/// 3. **Any parsed dotted type → `CodexStream`.**
1004/// 4. **A single parsed `result` line → `SingleDocEnvelope`** — today's shipped
1005/// format, which must keep the raw-scan path (T-30-02 / T-30-25).
1006/// 5. **Multi-line with a `result` event or a torn JSON line → `ClaudeStream`.**
1007/// A stream whose gate-bearing `user` event tore, leaving only a later
1008/// `result`, is still a stream (fourth-pass Low / third-pass Medium shape).
1009/// 6. Everything else → `PlainText`.
1010///
1011/// A LONE torn JSON line is deliberately `PlainText`, not `ClaudeStream`: under
1012/// today's format that shape is a torn single-document envelope, and raw-scanning
1013/// it preserves detection of a REAL gate declaration inside (dropping one is the
1014/// T-30-24 harm — worse than the echo false positive). The residual — a stream
1015/// that died with only its echoed-prompt line, torn, and nothing else — requires
1016/// the `init` line to have never flushed while the echo line partially did.
1017/// Accepted and recorded rather than silently traded away.
1018fn classify(capture: &ParsedCapture) -> CaptureKind {
1019 let total = capture.line_shapes.len();
1020 if total == 0 {
1021 return CaptureKind::PlainText;
1022 }
1023 let noise = capture
1024 .line_shapes
1025 .iter()
1026 .filter(|s| **s == LineShape::Noise)
1027 .count();
1028 if (total - noise) * 2 <= total {
1029 return CaptureKind::PlainText;
1030 }
1031
1032 if capture.events.iter().any(|v| {
1033 matches!(
1034 v.get("type").and_then(serde_json::Value::as_str),
1035 Some("system" | "user" | "assistant")
1036 )
1037 }) {
1038 return CaptureKind::ClaudeStream;
1039 }
1040 if capture.events.iter().any(|v| {
1041 v.get("type")
1042 .and_then(serde_json::Value::as_str)
1043 .is_some_and(|t| t.contains('.'))
1044 }) {
1045 return CaptureKind::CodexStream;
1046 }
1047
1048 let result_events = capture
1049 .events
1050 .iter()
1051 .filter(|v| v.get("type").and_then(serde_json::Value::as_str) == Some("result"))
1052 .count();
1053 if total == 1 {
1054 return if result_events == 1 {
1055 CaptureKind::SingleDocEnvelope
1056 } else {
1057 CaptureKind::PlainText
1058 };
1059 }
1060 if result_events > 0 || capture.torn_json_line_present() {
1061 CaptureKind::ClaudeStream
1062 } else {
1063 CaptureKind::PlainText
1064 }
1065}
1066
1067/// Test-only accessor: does [`classify`] call this capture text a Claude
1068/// `stream-json` capture?
1069///
1070/// Exists so `monitor.rs`'s end-to-end tracer test can assert on the REAL
1071/// classifier rather than re-deriving "looks like a stream" with its own
1072/// heuristic — which is precisely the per-call-site divergence [`classify`]
1073/// was introduced to end. `classify`/`CaptureKind`/`ParsedCapture` stay
1074/// private; only this yes/no question crosses the module boundary, and only
1075/// under `cfg(test)`.
1076#[cfg(test)]
1077pub(crate) fn capture_is_claude_stream(capture: &str) -> bool {
1078 classify(&ParsedCapture::parse(capture)) == CaptureKind::ClaudeStream
1079}
1080
1081/// Whether an event is TOP-LEVEL — authored by the orchestrator session, not
1082/// forwarded from a subagent. `parent_tool_use_id` JSON-null or absent.
1083///
1084/// The ONE provenance predicate, shared by gate scanning and verdict selection
1085/// (constraint 9 item 2 / code-review M2: the two paths previously held
1086/// different notions — gate scanning enforced provenance while
1087/// [`last_top_level_result`] silently did not, despite its name and doc).
1088///
1089/// The absent case must stay top-level: `result` events carry no such key at
1090/// all in any archived capture. Treating absence as positive provenance remains
1091/// NECESSARY for today's captures and UNPROVEN safe — no archived capture
1092/// contains a subagent-origin `result`, so if one can omit the key it would be
1093/// admitted. Recorded, not solved; the type filter is the second, independent
1094/// guard on the gate path.
1095fn is_top_level(event: &serde_json::Value) -> bool {
1096 matches!(
1097 event.get("parent_tool_use_id"),
1098 None | Some(serde_json::Value::Null)
1099 )
1100}
1101
1102/// The LAST top-level `type: "result"` event in a Claude stream capture.
1103///
1104/// One capture can hold several: a session kept alive across turns emits one
1105/// terminal `result` per turn (the archived v3 stream carries three, at lines
1106/// 19, 37 and 54, produced across task-notification wake-ups). The last is the
1107/// session's final verdict, so an earlier turn must never decide the stage.
1108///
1109/// T-30-01: selection runs over TOP-LEVEL objects only — each value here is one
1110/// whole JSONL line. A `result`-shaped structure the agent writes inside its own
1111/// message text is inert string content and structurally unreachable from this
1112/// scan. Never route this through [`json_scan`]/[`json_find_key`], which descend
1113/// into nested objects; that is the same protection class as D-04/T-28-04's
1114/// top-level-only `session_id` read.
1115///
1116/// Provenance is ENFORCED via [`is_top_level`], not merely documented — the
1117/// first version of this function selected on `type == "result"` alone, so a
1118/// subagent-origin `result` event would have decided the stage (code-review
1119/// M2, constraint 9 item 2).
1120fn last_top_level_result(events: &[serde_json::Value]) -> Option<&serde_json::Value> {
1121 events.iter().rev().find(|v| {
1122 v.get("type").and_then(serde_json::Value::as_str) == Some("result") && is_top_level(v)
1123 })
1124}
1125
1126/// Whether a declared canary `token` came back inside a TOP-LEVEL `result`
1127/// event of this capture (D-13).
1128///
1129/// **Why this takes capture TEXT rather than a project root and phase**, unlike
1130/// its siblings [`checkpoint_reported_in_capture`] and
1131/// [`session_id_from_capture`]: the delivery canary runs against its own
1132/// throwaway capture file, not the phase capture. A canary that read (and
1133/// therefore implied writing) `stdout_path(project_root, phase)` would clobber
1134/// the stage's own capture — the one artifact the entire Layer 1 cascade
1135/// decides on.
1136///
1137/// **D-13 trap 1 — this may not be a NEW trust path.** The CLI echoes the
1138/// operator's prompt back into the same stdout as a `user` event, so the
1139/// planted token *will* appear in the stream regardless of whether anything was
1140/// delivered. That echo is exactly what produced the checkpoint false positive
1141/// 30-05 fixed. Matching is therefore confined to events that are both
1142/// `type: "result"` and [`is_top_level`] — the same provenance predicate
1143/// [`last_top_level_result`] enforces, reused rather than reinvented.
1144///
1145/// **D-13 trap 2 — a match proves DELIVERY, never WORK.** The agent can see the
1146/// token in its own prompt and emit it without doing anything (999.67's shape).
1147/// A hit means "the task-notification path is alive"; it never means the
1148/// dispatched work happened. Summaries and merges remain the evidence of work
1149/// (D-16/D-18).
1150///
1151/// Scans EVERY top-level `result`, not just the last one, which is the one
1152/// place this deliberately differs from [`last_top_level_result`]. That
1153/// function selects the session's final *verdict*, so later turns must
1154/// supersede earlier ones. The canary asks a different question — "did the
1155/// token ever come back?" — and a token returned on an earlier
1156/// task-notification turn is a complete answer to it.
1157pub fn token_reported_in_capture(capture: &str, token: &str) -> bool {
1158 ParsedCapture::parse(capture)
1159 .events
1160 .iter()
1161 .filter(|v| {
1162 v.get("type").and_then(serde_json::Value::as_str) == Some("result") && is_top_level(v)
1163 })
1164 .any(|v| {
1165 v.get("result")
1166 .and_then(serde_json::Value::as_str)
1167 .is_some_and(|text| text.contains(token))
1168 })
1169}
1170
1171/// Whether ONE parsed stream event is a top-level `result` carrying a
1172/// `DEVFLOW_RESULT` marker in its `result` text.
1173///
1174/// Exposed for the pipe-owning monitor's close rule (Phase 31, constraint 4),
1175/// which must decide line-by-line and in real time whether the marker arm is
1176/// satisfied — it cannot wait for a whole capture and re-parse it.
1177///
1178/// This is a COMPOSITION of the two existing predicates, deliberately not a
1179/// second implementation of either. T-31-01: the CLI echoes the operator's
1180/// prompt back into the same stdout as a `user` event — that echo is what
1181/// produced the checkpoint false positive 30-05 fixed — so a marker seen
1182/// anywhere but inside an event that is BOTH `type: "result"` AND
1183/// [`is_top_level`] must not close the stream. Reusing [`parse_marker_lines`]
1184/// keeps the marker grammar (case-insensitive prefix, edge-corruption
1185/// stripping, JSON body) in one place rather than letting the monitor grow a
1186/// looser `contains("DEVFLOW_RESULT")` of its own.
1187pub(crate) fn event_is_top_level_result_marker(event: &serde_json::Value) -> bool {
1188 event.get("type").and_then(serde_json::Value::as_str) == Some("result")
1189 && is_top_level(event)
1190 && event
1191 .get("result")
1192 .and_then(serde_json::Value::as_str)
1193 .and_then(parse_marker_lines)
1194 .is_some()
1195}
1196
1197/// Whether any AGENT-AUTHORED text in a Claude stream capture declares a
1198/// human-blocking gate. The stream-capture half of
1199/// [`blocking_human_checkpoint_reported`]; the pure matcher it delegates to,
1200/// [`text_reports_human_gate`], is unchanged.
1201///
1202/// **Why this exists (review constraint 3).** Scanning raw stdout is safe under
1203/// the single-document envelope, because the only place gate text can appear
1204/// there is the one `result` field the agent authored. A stream capture breaks
1205/// that invariant: the operator's prompt is echoed back into the same stdout as
1206/// a `user` event, so a prompt that merely DOCUMENTS a checkpoint gate
1207/// rendering becomes textually indistinguishable from a live declaration. The
1208/// failure is silent — a checkpoint auto-decide fires, or the resume ceiling is
1209/// consumed, on a stage whose prompt only discussed checkpoints. DevFlow's own
1210/// planning documents are exactly that kind of prompt content.
1211///
1212/// Two independent filters, both required, neither a substitute for the other:
1213///
1214/// 1. **Type — keep ONLY `result` events.** `user` events are always either the
1215/// echoed prompt or a `task_notification` summary re-injected as user-role
1216/// content; neither is the agent declaring anything. `system` events carry
1217/// the `init` tool and agent inventory, inert text with no business in a gate
1218/// scan. `assistant` events are excluded too, and that exclusion is
1219/// deliberate — do NOT "restore" it for completeness. Turn-FINAL assistant
1220/// text is duplicated verbatim into the `result` event that follows it
1221/// (`30a-evidence/raw_output_v3.jsonl` lines 17→19, 36→37, 53→54), so
1222/// admitting the class buys no detection the `result` events do not already
1223/// give. What it buys is a new false-positive surface: v3 line 6's top-level
1224/// assistant narration ("I'll spawn both subagents in the background now.")
1225/// reaches no `result` event at all, so an agent narrating "next I'll handle
1226/// the task whose gate the plan declares" would recreate the prompt-echo
1227/// false positive one layer inward.
1228/// 2. **Provenance — keep only top-level events.** An event is top-level when
1229/// `parent_tool_use_id` is JSON null OR the key is absent entirely. The
1230/// absent case is load-bearing: `result` events carry no such key at all
1231/// (confirmed across all three archived captures), so a naive presence check
1232/// would drop exactly the events that matter most. Mistaking
1233/// subagent-forwarded narration for orchestrator output is the error that
1234/// invalidated the v1 experiment outright. Kept even though filter 1 already
1235/// makes it redundant for today's captures — the two guards are meant to
1236/// fail independently, so a future widening of the type filter cannot
1237/// silently inherit subagent content.
1238///
1239/// **ALL eligible `result` events are scanned, not only the last.** This
1240/// deliberately diverges from [`last_top_level_result`]'s last-result-wins
1241/// verdict semantics, and the two conventions must not be "harmonised": a
1242/// verdict is a single final answer, whereas this asks whether a gate was
1243/// reported ANYWHERE in the stage's output. A gate declared in turn N followed
1244/// by task-notification wake-up turns N+1/N+2 — the exact turn shape the v3
1245/// capture archives — would be silently dropped by last-result-only, losing a
1246/// human authorization request to the generic gate. That is the
1247/// opposite-direction harm, and the worse of the two.
1248///
1249/// Text is read with a direct [`serde_json::Value::get`] chain. Never route
1250/// this through [`json_scan`]/[`json_find_key`]: a recursive traversal descends
1251/// straight back into the nested message content both filters just excluded,
1252/// silently undoing the fix while the tests on the outer shape still pass
1253/// (T-30-23).
1254///
1255/// Returns `bool` and short-circuits on the first match rather than collecting
1256/// the eligible text: this runs on every `devflow advance` over a capture that
1257/// grows for the whole stage, and there is no reason to allocate a copy of it.
1258fn claude_stream_reports_human_gate(events: &[serde_json::Value]) -> bool {
1259 events
1260 .iter()
1261 .filter(|event| event.get("type").and_then(serde_json::Value::as_str) == Some("result"))
1262 .filter(|event| is_top_level(event))
1263 .filter_map(|event| event.get("result").and_then(serde_json::Value::as_str))
1264 .any(text_reports_human_gate)
1265}
1266
1267/// The `rate_limit_info.status` values that mean the CLI DENIED the request.
1268///
1269/// Provenance, per entry — required reading before adding one:
1270///
1271/// - `rejected` — drawn from the observed vocabulary of this schema: it is the
1272/// value the CLI writes for `overageStatus` in the only archived
1273/// `rate_limit_event`
1274/// (`.planning/phases/30-keep-the-session-alive-past-turn-end/30a-evidence/raw_output_v3.jsonl`
1275/// line 15), so it is the denial token this schema actually speaks. It has
1276/// NOT been observed as a `status` value — no archived capture is of a
1277/// blocked stream, and every capture DevFlow has taken carries
1278/// `status: "allowed"`.
1279///
1280/// Nothing else is listed, deliberately. Speculatively adding tokens is how the
1281/// false positive this list exists to prevent comes back: an unrecognised
1282/// status must DEFER (see [`detect_claude_stream_rate_limit`]), never classify.
1283/// Correct this list the first time a real blocked capture is archived — that
1284/// is the only evidence that settles the vocabulary.
1285const CLAUDE_STREAM_RATE_LIMIT_DENIAL_STATUSES: &[&str] = &["rejected"];
1286
1287/// Detect an explicit quota DENIAL in a Claude `stream-json` capture and return
1288/// the retry description, mirroring what [`detect_claude_rate_limit`] returns
1289/// for the single-document envelope.
1290///
1291/// **A `rate_limit_event` is not a rate limit.** The CLI emits these routinely
1292/// as quota telemetry on healthy streams: the only archived one
1293/// (`raw_output_v3.jsonl` line 15) says `rate_limit_info.status: "allowed"` and
1294/// sits in a stream that then completed three turns successfully. Classifying
1295/// on the event's PRESENCE would mark every healthy Claude stream stage
1296/// `RateLimited`, and `outcome_policy.rs` maps that to `Action::AutoResume` —
1297/// so every stage would be auto-resumed against a fabricated retry time
1298/// instead of advancing (T-30-26). Note the second trap in the same object:
1299/// `overageStatus` is `rejected` one level below `status: "allowed"`, so any
1300/// nested search for the token also false-positives. Hence every field here is
1301/// read with a direct [`serde_json::Value::get`] on the top-level event and its
1302/// `rate_limit_info` child — never [`json_find_key`]/[`json_scan`], which
1303/// descend into nested (and, elsewhere in the stream, agent-authored) content
1304/// and would let the agent supply the retry hint that drives the resume cron's
1305/// scheduling (T-30-12).
1306///
1307/// Two independent guards, both required, neither a substitute for the other:
1308///
1309/// 1. **Positional** — only events after the SECOND-TO-LAST `result` event are
1310/// eligible, i.e. the final turn. A session kept alive across turns emits one
1311/// `result` per turn, and rate-limit chatter from an earlier turn must never
1312/// outrank the outcome of a turn that finished later. (In the archived
1313/// capture the rate event is at line 15 and the results at 19/37/54, so it is
1314/// excluded on position alone.) With fewer than two `result` events the whole
1315/// stream IS the final turn.
1316/// 2. **Semantic** — only a `status` in
1317/// [`CLAUDE_STREAM_RATE_LIMIT_DENIAL_STATUSES`] classifies. A missing
1318/// `rate_limit_info`, a missing or non-string `status`, or any unrecognised
1319/// value returns `None`.
1320///
1321/// **Deferring is the deliberately safe direction, not an oversight.**
1322/// Under-classifying means an unknown denial status falls through to the
1323/// envelope-failure path and is reported `Failed` — a real degradation (the
1324/// operator loses automatic resume) but a never-silent one that still gates.
1325/// Over-classifying means a healthy stream is auto-resumed against a retry time
1326/// the parser invented. The asymmetry is the whole reason this function reads
1327/// one field instead of matching a shape.
1328fn detect_claude_stream_rate_limit(events: &[serde_json::Value]) -> Option<String> {
1329 // Index of the second-to-last `result` event: everything at or before it is
1330 // previous-turn history. `None` (fewer than two results) means the whole
1331 // stream is the final turn.
1332 let boundary = events
1333 .iter()
1334 .enumerate()
1335 .filter(|(_, v)| v.get("type").and_then(serde_json::Value::as_str) == Some("result"))
1336 .map(|(idx, _)| idx)
1337 .rev()
1338 .nth(1);
1339 let eligible = match boundary {
1340 Some(idx) => &events[idx + 1..],
1341 None => events,
1342 };
1343
1344 // Last eligible event wins, matching the last-`result`-wins convention.
1345 let event = eligible
1346 .iter()
1347 .rev()
1348 .find(|v| v.get("type").and_then(serde_json::Value::as_str) == Some("rate_limit_event"))?;
1349
1350 let info = event.get("rate_limit_info")?;
1351 let status = info.get("status")?.as_str()?;
1352 if !CLAUDE_STREAM_RATE_LIMIT_DENIAL_STATUSES.contains(&status) {
1353 return None;
1354 }
1355
1356 // `resetsAt` is epoch seconds, rendered from the JSON number as-is: nothing
1357 // parses this string. `outcome_policy.rs` routes on the
1358 // `AgentStatus::RateLimited` variant alone and the `reason` text is
1359 // operator-facing. Mirrors `detect_claude_rate_limit`'s `retry_after` →
1360 // `message` → `error` chain; its final `"usage limit"` default has no
1361 // counterpart here because a matched `status` is by construction one of the
1362 // non-empty enumerated strings above, so a third rung would be unreachable.
1363 Some(
1364 info.get("resetsAt")
1365 .and_then(json_scalar_to_string)
1366 .unwrap_or_else(|| status.to_string()),
1367 )
1368}
1369
1370/// The stream-path counterpart of [`detect_claude_envelope_failure`]: treat
1371/// `is_error: true` on a stream's last `result` event as an authoritative
1372/// Layer-1 failure.
1373///
1374/// The `reason` shape is reproduced deliberately rather than shared — `result`
1375/// text, else `subtype`, else `agent reported is_error`, with a
1376/// ` (num_turns: {n})` suffix when present. This phase's scope fence keeps the
1377/// four shipped single-document parsers unmodified, so factoring the common
1378/// body out of `detect_claude_envelope_failure` is out of bounds here; the two
1379/// must be kept in step by hand. `is_error` absent, non-bool, or `false`
1380/// returns `None`, deferring exactly as the single-document path does.
1381fn claude_stream_envelope_failure(result_event: &serde_json::Value) -> Option<AgentResult> {
1382 if !result_event.get("is_error")?.as_bool()? {
1383 return None;
1384 }
1385
1386 let num_turns = result_event
1387 .get("num_turns")
1388 .and_then(serde_json::Value::as_u64);
1389 let base_reason = result_event
1390 .get("result")
1391 .and_then(serde_json::Value::as_str)
1392 .map(str::to_string)
1393 .or_else(|| {
1394 result_event
1395 .get("subtype")
1396 .and_then(serde_json::Value::as_str)
1397 .map(str::to_string)
1398 })
1399 .unwrap_or_else(|| "agent reported is_error".to_string());
1400 let reason = match num_turns {
1401 Some(n) => format!("{base_reason} (num_turns: {n})"),
1402 None => base_reason,
1403 };
1404
1405 Some(AgentResult {
1406 status: AgentStatus::Failed,
1407 exit_code: None,
1408 reason: Some(reason),
1409 commits: None,
1410 summary: None,
1411 verdict: None,
1412 decided_by_layer: Some(1),
1413 })
1414}
1415
1416/// Parse a Claude `--output-format stream-json` JSONL capture and read the
1417/// `DEVFLOW_RESULT` marker out of its LAST `result` event.
1418///
1419/// The new sibling of [`parse_codex_event_result`], mirroring its shape. Only
1420/// decisive when the capture is actually a Claude event stream (per
1421/// [`is_claude_event_stream`]); every other shape returns `None` and falls
1422/// through to the parser that owns it. Before this existed, a JSONL capture
1423/// returned `None` from all four single-document parsers —
1424/// `serde_json::from_str` on the whole multi-line document is a hard "trailing
1425/// characters" error — so every Claude-driven stage fell through to Layer 2's
1426/// coarse exit-code+commit heuristic.
1427///
1428/// **Precedence, mirroring [`evaluate_layer1`]'s single-document ordering
1429/// rather than inventing a new one** — do not reshuffle without reading the
1430/// reasons:
1431///
1432/// 1. Format gate ([`is_claude_event_stream`]); every other shape declines here.
1433/// 2. [`detect_claude_stream_rate_limit`] — a final-turn explicit quota denial
1434/// wins over EVERYTHING below it, for the same reason `evaluate_layer1`
1435/// already puts `detect_claude_rate_limit` ahead of the generic failure
1436/// check: a rate-limited run classified as plain `Failed` kills the primary
1437/// rate-limit resume cron, the one automated path that exists to recover
1438/// from it (T-30-13). The precedence is narrow, not broad — the detector
1439/// only fires on an explicit denial inside the final turn, so it cannot
1440/// shadow the outcome of a stream that completed.
1441/// 3. The `DEVFLOW_RESULT` marker in the last `result` event. A non-success
1442/// marker is decisive and returns immediately; a success marker is HELD, not
1443/// returned, because step 4 may override it.
1444/// 4. [`claude_stream_envelope_failure`] — `is_error: true` on that same event
1445/// overrides a held success marker, matching the single-document rule that
1446/// the envelope is authoritative for errors and a stale or echoed success
1447/// marker must not win (T-30-15).
1448/// 5. The held success marker, else `None`.
1449///
1450/// A last `result` event with no marker and no `is_error` returns `None`
1451/// (defer to Layer 2) rather than an unconditional Success, matching the
1452/// `turn.completed` convention: a marker-less turn must never silently advance
1453/// a stage.
1454///
1455/// Passing the isolated `result` text to [`parse_marker_lines`] is the correct
1456/// scoping, not a workaround. The marker is JSON-escaped inside a
1457/// `"result":"..."` string value, so it can never appear as a line starting
1458/// with `DEVFLOW_RESULT:` in the raw capture, and that parser's 4000-character
1459/// tail window is smaller than a single stream `result` line. Once serde
1460/// decodes the field the escaped newlines become real newlines and the existing
1461/// tail scan works on it as designed.
1462fn parse_claude_event_result(stdout: &str) -> Option<AgentResult> {
1463 let capture = ParsedCapture::parse(stdout);
1464 if !is_claude_event_stream(&capture.events) {
1465 return None;
1466 }
1467
1468 // Constraint 9 item 1 (code-review H1): a torn JSON line at or after the
1469 // last surviving top-level result means the session's REAL final verdict
1470 // may be among the casualties — a capture read while the CLI was still
1471 // appending, or a truncated write. Nothing that survives before the tear
1472 // is allowed to stand in for it; in particular an earlier turn's SUCCESS
1473 // must never advance the stage. Returning a Failed verdict rather than
1474 // None is deliberate: None would fall through to `parse_devflow_result`'s
1475 // raw tail scan, which can find the stale marker TEXT inside the surviving
1476 // JSON lines and resurrect it through the back door. The cost is a false
1477 // failure when the torn trailing line was a quiet task-notification turn;
1478 // that reads as loop-back noise, not a silent wrong advance.
1479 if capture.torn_json_after_last_matching(|v| {
1480 v.get("type").and_then(serde_json::Value::as_str) == Some("result") && is_top_level(v)
1481 }) {
1482 return Some(indeterminate_capture_failure());
1483 }
1484
1485 if let Some(retry) = detect_claude_stream_rate_limit(&capture.events) {
1486 return Some(rate_limited_result(retry));
1487 }
1488
1489 let last_result = last_top_level_result(&capture.events)?;
1490
1491 let marker = last_result
1492 .get("result")
1493 .and_then(serde_json::Value::as_str)
1494 .and_then(parse_marker_lines)
1495 .map(normalise_stream_marker_provenance);
1496
1497 let held_success = match marker {
1498 // A non-success marker is the agent's own final word and nothing below
1499 // can improve on it.
1500 //
1501 // 31-02 audit (non-exhaustive equality site 1 of 3). This `!= Success`
1502 // is CORRECT AS-IS for `AgentStatus::IdleTimeout` and is deliberately
1503 // left unchanged. The compiler cannot flag this site — an equality test
1504 // compiles fine against a new variant — so it is audited by hand here
1505 // rather than left to the wildcard-free-match mechanism, which does not
1506 // reach it.
1507 //
1508 // The only way `IdleTimeout` arrives here is an agent writing
1509 // `DEVFLOW_RESULT: {"status":"idle_timeout"}` into its own output,
1510 // claiming a verdict only DevFlow's monitor is supposed to produce.
1511 // The predicate handles that in the fail-safe direction: it is not
1512 // `Success`, so it returns immediately as decisive non-success and
1513 // `decide_action` gates it for review. A forged idle timeout can
1514 // therefore only make a run gate, never advance. The REAL
1515 // monitor-produced verdict does not travel this path at all — it is
1516 // read from its own side-channel file at the top of `evaluate_layer1`,
1517 // before this parser ever runs.
1518 Some(result) if result.status != AgentStatus::Success => return Some(result),
1519 other => other,
1520 };
1521
1522 if let Some(failure) = claude_stream_envelope_failure(last_result) {
1523 return Some(failure);
1524 }
1525
1526 held_success
1527}
1528
1529/// The Layer-1 verdict for a stream capture whose TAIL is provably unreadable:
1530/// a torn JSON line after the last surviving result (constraint 9 item 1).
1531///
1532/// Failed, not `None`, and not the pre-tear result. `None` hands the same
1533/// stdout to `parse_devflow_result`'s raw tail scan, which can resurrect the
1534/// stale marker text out of the surviving JSON lines; the pre-tear result is
1535/// exactly the stale-success defect this exists to close. A false failure on a
1536/// torn-but-benign tail surfaces as a retried stage, never as a silent wrong
1537/// advance — the asymmetry this whole module is built around.
1538fn indeterminate_capture_failure() -> AgentResult {
1539 AgentResult {
1540 status: AgentStatus::Failed,
1541 exit_code: None,
1542 reason: Some(
1543 "stream capture ends in an unparseable line; the final verdict is indeterminate"
1544 .to_string(),
1545 ),
1546 commits: None,
1547 summary: None,
1548 verdict: None,
1549 decided_by_layer: Some(1),
1550 }
1551}
1552
1553/// T-30-26: overwrite the agent-supplied `decided_by_layer` unconditionally.
1554///
1555/// [`parse_marker_lines`] deserializes the agent's own marker JSON straight
1556/// into [`AgentResult`], and the field is `#[serde(default)]`, so an ordinary
1557/// `{"status":"success"}` marker leaves it `None` while a hostile
1558/// `{"status":"success","decided_by_layer":0}` leaves it `Some(0)`. Neither is
1559/// acceptable: every other Layer-1 constructor in this module sets `Some(1)`
1560/// explicitly, and `Some(0)` is a Layer-0 external-probe provenance that
1561/// `classify_validate_outcome` (devflow-cli's `pipeline_outcomes.rs`) reads as
1562/// `external` when classifying a Validate stage. An agent must not be able to
1563/// claim a probe verdict it did not earn, so the value is derived here rather
1564/// than trusted.
1565fn normalise_stream_marker_provenance(mut result: AgentResult) -> AgentResult {
1566 result.decided_by_layer = Some(1);
1567 result
1568}
1569
1570/// Scan a bounded tail of `stdout` in reverse line order for the last
1571/// `DEVFLOW_RESULT` marker.
1572///
1573/// `DEVFLOW_RESULT` markers are ASCII. Searching the bounded tail and returning
1574/// the last valid marker ensures the agent's final status wins over an earlier
1575/// prompt echo without requiring the surrounding output to be ASCII.
1576///
1577/// Three sixth-pass corrections, each with a paired regression:
1578/// - The tail budget counts WHOLE LINES, never bisecting one (High 2): the old
1579/// fixed 4000-char window could cut through the final marker line itself
1580/// when it carried a long `reason`, silently dropping the authoritative
1581/// failure and handing the verdict to the exit code.
1582/// - Each line is edge-stripped before prefix matching (High 1): the capture
1583/// is read lossily, so one stray byte became U+FFFD glued to the prefix or
1584/// the JSON and the marker vanished. Same policy as every other reader:
1585/// edges stripped, interior corruption stays visible and untrusted.
1586/// - The prefix match is genuinely case-insensitive (High 3), as this
1587/// parser's contract has promised all along — the old strip_prefix chain
1588/// accepted only ALL-upper or ALL-lower.
1589fn parse_marker_lines(stdout: &str) -> Option<AgentResult> {
1590 const TAIL_BUDGET_CHARS: usize = 4000;
1591 const PREFIX: &str = "DEVFLOW_RESULT:";
1592
1593 let mut budget_used = 0usize;
1594 for line in stdout.lines().rev() {
1595 // The line that crosses the budget is still scanned whole; only the
1596 // NEXT one stops the walk. The last line is always scanned, however
1597 // long — that is the line the fixed window used to bisect.
1598 if budget_used > TAIL_BUDGET_CHARS {
1599 break;
1600 }
1601 budget_used += line.chars().count() + 1;
1602
1603 let line = strip_corruption_padding(line);
1604 let Some(head) = line.get(..PREFIX.len()) else {
1605 continue;
1606 };
1607 if !head.eq_ignore_ascii_case(PREFIX) {
1608 continue;
1609 }
1610
1611 let json_str = line[PREFIX.len()..].trim();
1612 if let Ok(result) = serde_json::from_str::<AgentResult>(json_str) {
1613 return Some(result);
1614 }
1615 }
1616 None
1617}
1618
1619/// One commit the agent made before its stream went silent (D-07, 31-02).
1620///
1621/// The subject is carried alongside the sha because a bare sha list is not
1622/// operator-actionable — D-07's requirement is that the commits be *named*, so
1623/// that a silent miscount becomes something a human can act on.
1624#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
1625pub struct IdleTimeoutCommit {
1626 /// Full commit sha, as `git log --format=%H` emits it.
1627 pub sha: String,
1628 /// Commit subject line (`%s`).
1629 pub subject: String,
1630}
1631
1632/// The pipe-owning monitor's authoritative idle-timeout verdict, as written to
1633/// [`idle_timeout_path`] BEFORE the child is terminated (D-05, 31-02).
1634///
1635/// This is a SIDE CHANNEL, deliberately not the stdout capture. See
1636/// [`parse_idle_timeout_side_channel`] for why that distinction is a
1637/// correctness requirement rather than a filing preference.
1638#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
1639pub struct IdleTimeoutRecord {
1640 /// Always [`AgentStatus::IdleTimeout`]'s wire string. Recorded so the file
1641 /// is self-describing to a human reading `.devflow/` by hand.
1642 pub status: String,
1643 /// The idle window that elapsed with no line on the child's stdout.
1644 pub idle_secs: u64,
1645 /// The supervised child's pid, from the in-memory `Child` handle — never
1646 /// re-read from the on-disk pid file, which is exposed to pid reuse
1647 /// (T-31-07).
1648 pub agent_pid: u32,
1649 /// Unix seconds at which the monitor wrote this record.
1650 pub written_at: u64,
1651 /// Every commit on the phase branch when the timeout fired. NONE of these
1652 /// is rolled back — see [`parse_idle_timeout_side_channel`].
1653 pub commits: Vec<IdleTimeoutCommit>,
1654}
1655
1656/// Read the monitor's own idle-timeout verdict, if it wrote one.
1657///
1658/// **This is consulted as the FIRST statement of [`evaluate_layer1`], before
1659/// `read_capture` and before every marker parser. That placement is
1660/// load-bearing and must not be "tidied" into the `.or_else` chain below it.**
1661///
1662/// The obvious-looking alternative — appending the verdict to the stdout
1663/// capture — is a real correctness bug, not a style choice.
1664/// `evaluate_layer1`'s chain reaches `parse_devflow_result`'s tail scan only
1665/// when `parse_claude_event_result` returns `None`, and that parser resolves to
1666/// the LAST top-level `result` event regardless of what text follows it. On any
1667/// stream that already completed one successful turn — the normal shape of a
1668/// run long enough to idle out at all — an appended verdict is therefore never
1669/// reached, and a stale success stands as the recorded outcome of a run DevFlow
1670/// itself killed (T-31-06, 31-RESEARCH Pitfall 3).
1671///
1672/// Reading before `read_capture` matters for a second reason: that call is an
1673/// early `return None` when the capture is missing, so a timeout that fired
1674/// before the child emitted anything at all would otherwise be discarded
1675/// entirely.
1676///
1677/// `decided_by_layer` stays `1`. This is a Layer-1-CLASS authoritative verdict
1678/// — it just comes from the monitor that supervised the run rather than from
1679/// parsing what the agent said about itself. It is emphatically not `0`, which
1680/// is reserved for operator-authored external probe provenance that
1681/// `classify_validate_outcome` reads as `external`.
1682///
1683/// **The file's PRESENCE is the signal; its contents are enrichment.** A record
1684/// that exists but cannot be read still returns an `IdleTimeout` verdict,
1685/// carrying a reason that says the details were lost. Returning `None` there
1686/// would drop the verdict back into the cascade and let precisely the stale
1687/// success above win — turning a corrupt file into a silent wrong advance,
1688/// which is the exact failure this function exists to prevent. The asymmetry is
1689/// the one this whole module is built around: a false failure surfaces as a
1690/// gate, never as a wrong advance.
1691///
1692/// **Nothing here rolls anything back** (D-07, T-31-09). The commits are read
1693/// and named, never reverted: an idle timeout may be a false positive, and
1694/// destroying real work on a false positive is unrecoverable.
1695fn parse_idle_timeout_side_channel(project_root: &Path, phase: u32) -> Option<AgentResult> {
1696 let path = idle_timeout_path(project_root, phase);
1697 let raw = read_capture(&path)?;
1698
1699 let Ok(record) = serde_json::from_str::<IdleTimeoutRecord>(&raw) else {
1700 return Some(idle_timeout_result(
1701 format!(
1702 "idle timeout: DevFlow's monitor recorded a timeout verdict at {} but the \
1703 record itself is unreadable, so the commit list and idle duration are lost. \
1704 The timeout stands regardless — the file's presence is the authoritative \
1705 signal. Inspect the phase branch by hand; nothing was rolled back.",
1706 path.display()
1707 ),
1708 None,
1709 ));
1710 };
1711
1712 let named: Vec<String> = record
1713 .commits
1714 .iter()
1715 .map(|commit| {
1716 let short: String = commit.sha.chars().take(7).collect();
1717 format!("{short} {}", commit.subject)
1718 })
1719 .collect();
1720
1721 let commit_phrase = if named.is_empty() {
1722 "No commits were found on the phase branch.".to_string()
1723 } else {
1724 format!(
1725 "The agent made {} commit(s) before going quiet and NONE of them were rolled \
1726 back: {}.",
1727 named.len(),
1728 named.join("; ")
1729 )
1730 };
1731
1732 Some(idle_timeout_result(
1733 format!(
1734 "idle timeout: the agent's output stream was silent for {}s, so DevFlow \
1735 terminated it (agent pid {}). {commit_phrase} Review the branch before deciding \
1736 what to keep — this run is TERMINAL and is not retried automatically.",
1737 record.idle_secs, record.agent_pid
1738 ),
1739 Some(record.commits.len() as u32),
1740 ))
1741}
1742
1743/// Build the `IdleTimeout` verdict Layer 1 reports for a monitor-recorded
1744/// timeout.
1745///
1746/// `verdict` stays `None` deliberately: at `Stage::Validate`,
1747/// `classify_validate_outcome` matches `Some(Verdict::Pass)` FIRST and would
1748/// classify the stage as passed on the strength of that field alone, whatever
1749/// the status says. A timeout has no verdict to offer, and inventing one here
1750/// would advance a run that never reported.
1751fn idle_timeout_result(reason: String, commits: Option<u32>) -> AgentResult {
1752 AgentResult {
1753 status: AgentStatus::IdleTimeout,
1754 exit_code: None,
1755 reason: Some(reason),
1756 commits,
1757 summary: None,
1758 verdict: None,
1759 decided_by_layer: Some(1),
1760 }
1761}
1762
1763/// Layer 1: Try to detect agent result from the native per-adapter envelope
1764/// or the DEVFLOW_RESULT marker in stdout.
1765///
1766/// The monitor's own idle-timeout side channel is consulted FIRST, ahead of
1767/// everything below including `read_capture` itself — see
1768/// [`parse_idle_timeout_side_channel`], where that ordering is a correctness
1769/// requirement rather than a preference.
1770///
1771/// Precedence: Claude rate-limit envelope (a SPECIFIC failure that must
1772/// outrank the generic `is_error` check — rate-limit envelopes carry
1773/// `is_error: true`, and classifying them `Failed` would kill the primary
1774/// rate-limit resume cron path) → Claude envelope `is_error: true` (authoritative,
1775/// overrides a success marker) → Claude `stream-json` JSONL event stream (the
1776/// last `result` event's marker decides; a marker-less last turn defers) →
1777/// DEVFLOW_RESULT marker (portable; works for plain text and a Claude
1778/// envelope's unwrapped `result` text) → Codex JSONL event stream
1779/// (`turn.failed` decisive; `turn.completed` defers) → Codex plain-text
1780/// rate-limit heuristic (least authoritative, stays last).
1781///
1782/// The Claude stream parser's position is load-bearing in BOTH directions
1783/// (T-30-03). The two single-document detectors stay ahead of it because they
1784/// remain authoritative for the `--output-format json` envelope that ships
1785/// today. It goes ahead of `parse_devflow_result` so that an adapter-specific
1786/// stream capture is owned whole by the parser that understands its framing,
1787/// rather than letting the generic 4000-character tail scan take a bite of a
1788/// mid-line window of JSONL first.
1789pub fn evaluate_layer1(project_root: &Path, phase: u32) -> Option<AgentResult> {
1790 // FIRST STATEMENT, before `read_capture` and before every parser below.
1791 // Do not move this into the `.or_else` chain: `parse_claude_event_result`
1792 // resolves the LAST top-level `result` event and would shadow it on any
1793 // stream that already had one successful turn. See
1794 // `parse_idle_timeout_side_channel`'s doc comment (T-31-06).
1795 if let Some(timed_out) = parse_idle_timeout_side_channel(project_root, phase) {
1796 return Some(timed_out);
1797 }
1798
1799 let stdout = read_capture(&stdout_path(project_root, phase))?;
1800 detect_claude_rate_limit(&stdout)
1801 .map(rate_limited_result)
1802 .or_else(|| detect_claude_envelope_failure(&stdout))
1803 .or_else(|| parse_claude_event_result(&stdout))
1804 .or_else(|| parse_devflow_result(&stdout))
1805 .or_else(|| parse_codex_event_result(&stdout))
1806 .or_else(|| detect_codex_rate_limit(&stdout).map(rate_limited_result))
1807}
1808
1809/// Build the `RateLimited` result Layer 1 reports for a detected retry hint.
1810fn rate_limited_result(retry: String) -> AgentResult {
1811 AgentResult {
1812 status: AgentStatus::RateLimited,
1813 exit_code: None,
1814 reason: Some(format!("rate limited until {retry}")),
1815 commits: None,
1816 summary: None,
1817 verdict: None,
1818 decided_by_layer: Some(1),
1819 }
1820}
1821
1822/// Commits on the phase's feature branch that are not on `develop`.
1823///
1824/// Derives the branch name from `git_flow.feature_prefix` and the zero-padded
1825/// `phase`, verifies the branch exists with `rev-parse --verify`, and on
1826/// success counts `{git_flow.develop}..{branch}` with `rev-list --count`.
1827/// This is the single implementation of that count — [`evaluate_layer2`] and
1828/// `pipeline_outcomes::handle_validate_outcome`'s forward-progress check both
1829/// call it rather than each re-deriving the branch name and re-running the
1830/// same two git commands, which is what made the two counts able to silently
1831/// diverge before this extraction.
1832///
1833/// Must be called with the main `project_root`, never a worktree path — git
1834/// worktrees share refs and the object database, so a commit made inside a
1835/// linked worktree is immediately visible to a count run from the main
1836/// checkout, which is the property every caller already relies on.
1837///
1838/// A `0` return is deliberately indistinguishable across three causes:
1839/// genuinely no commits, the branch does not exist, or `git` could not be
1840/// run. Every consumer treats all three the same way.
1841pub fn phase_commit_count(project_root: &Path, git_flow: &GitFlowConfig, phase: u32) -> u32 {
1842 let branch = format!("{}phase-{:02}", git_flow.feature_prefix, phase);
1843
1844 let branch_exists = git_command(project_root)
1845 .args(["rev-parse", "--verify", &branch])
1846 .output()
1847 .map(|o| o.status.success())
1848 .unwrap_or(false);
1849
1850 if !branch_exists {
1851 return 0;
1852 }
1853
1854 let range = format!("{}..{branch}", git_flow.develop);
1855 git_command(project_root)
1856 .args(["rev-list", "--count", &range])
1857 .output()
1858 .ok()
1859 .and_then(|o| String::from_utf8_lossy(&o.stdout).trim().parse().ok())
1860 .unwrap_or(0)
1861}
1862
1863/// Layer 2: Use exit code + commit count to determine result.
1864///
1865/// Reads exit code from `.devflow/phase-NN-exit` file.
1866/// Counts commits in `feature/phase-NN` branch (if it exists), via
1867/// [`phase_commit_count`].
1868///
1869/// The commit-count gate ("no commits → failed") is scoped to `stage` — it
1870/// only applies to `Stage::Plan`/`Stage::Code` (checked via an explicit
1871/// `matches!`, NOT `Stage::is_agent_stage()`, since that also includes
1872/// `Define`, which legitimately produces zero commits). `exit≠0` is ALWAYS
1873/// `Failed`, for every stage — only the `exit=0`/zero-commits branch is
1874/// stage-scoped.
1875///
1876/// Decision matrix:
1877/// exit=137 → ResourceKilled (ALL stages, D-07)
1878/// exit=127 → AgentUnavailable (ALL stages, D-07)
1879/// exit≠0 (excluding 137/127) → Failed (ALL stages)
1880/// exit=0, stage in {Plan, Code}, commits=0 → Failed ("no work done")
1881/// exit=0, stage in {Plan, Code}, commits>0 → Success
1882/// exit=0, stage NOT in {Plan, Code} (Define/Validate/Ship), commits=0 → Success
1883/// (not commit-gated; Validate's real pass signal is its verdict,
1884/// not a bare zero-commit — see Task 2's turn.completed deferral)
1885/// exit unknown → fall to Layer 3 (return None)
1886///
1887/// WR-06 (13-REVIEW.md): takes only the explicit `project_root` parameter
1888/// for both the `.devflow/` file paths and the git subprocess `current_dir`
1889/// — previously it also accepted `state: &State` and used `state.project_root`
1890/// for the git calls, which every caller happened to pass consistently with
1891/// `project_root` but which the function itself had no way to enforce.
1892pub fn evaluate_layer2(
1893 project_root: &Path,
1894 phase: u32,
1895 git_flow: &GitFlowConfig,
1896 stage: Stage,
1897) -> Result<Option<AgentResult>, ResultError> {
1898 let exit_path = devflow_dir(project_root).join(format!("phase-{:02}-exit", phase));
1899 let exit_code: i32 = match std::fs::read_to_string(&exit_path) {
1900 Ok(s) => s.trim().parse().unwrap_or(-1),
1901 Err(_) => return Ok(None), // fall to Layer 3
1902 };
1903
1904 let branch = format!("{}phase-{:02}", git_flow.feature_prefix, phase);
1905 let commits: u32 = phase_commit_count(project_root, git_flow, phase);
1906
1907 let commit_gated = matches!(stage, Stage::Plan | Stage::Code);
1908 let no_work_done = commit_gated && commits == 0;
1909
1910 // 137 (SIGKILL, typically OOM) and 127 (command not found) are classified
1911 // BEFORE the generic `exit_code != 0 -> Failed` catch-all, using the same
1912 // trusted plain-i32 already parsed above from the monitor-written exit
1913 // file (D-07, 17b — no ExitStatusExt/signal API per Pitfall 1a).
1914 let status = if exit_code == 137 {
1915 AgentStatus::ResourceKilled
1916 } else if exit_code == 127 {
1917 AgentStatus::AgentUnavailable
1918 } else if exit_code != 0 || no_work_done {
1919 AgentStatus::Failed
1920 } else {
1921 AgentStatus::Success
1922 };
1923
1924 Ok(Some(AgentResult {
1925 status,
1926 exit_code: Some(exit_code),
1927 reason: if exit_code == 137 {
1928 Some(format!(
1929 "agent process was killed (exit code 137, likely OOM) ({} commits on {})",
1930 commits, branch
1931 ))
1932 } else if exit_code == 127 {
1933 Some(format!(
1934 "agent command was unavailable (exit code 127, command not found) ({} commits on {})",
1935 commits, branch
1936 ))
1937 } else if exit_code != 0 {
1938 Some(format!(
1939 "agent exited with code {} ({} commits on {})",
1940 exit_code, commits, branch
1941 ))
1942 } else if no_work_done {
1943 Some(format!(
1944 "no commits found on {} (agent exit code was {})",
1945 branch, exit_code
1946 ))
1947 } else {
1948 Some(format!(
1949 "{} commits on {} (agent exit code was {})",
1950 commits, branch, exit_code
1951 ))
1952 },
1953 commits: Some(commits),
1954 summary: None,
1955 verdict: None,
1956 decided_by_layer: Some(2),
1957 }))
1958}
1959
1960/// Layer 3: Last resort — agent process is gone.
1961///
1962/// Split per D-02/D-03 case 3 (17-03): "process gone, commits exist" stays
1963/// `Unknown` — unverified but there is SOMETHING to account for, and Plan
1964/// 04's never-advance dispatch gates it downstream (D-04) rather than
1965/// reclassifying it here. "Process gone, zero commits, nothing declared" is
1966/// no longer a blanket advanceable `Unknown` — it is reclassified to
1967/// `Failed` so a vanished agent that produced and declared nothing cannot
1968/// masquerade as ambiguous-but-fine; the reason flags that human review is
1969/// needed. This only fires when neither Layer 1 nor Layer 2 produced a
1970/// definitive result.
1971pub fn evaluate_layer3(
1972 project_root: &Path,
1973 phase: u32,
1974 git_flow: &GitFlowConfig,
1975) -> Result<AgentResult, ResultError> {
1976 let branch = format!("{}phase-{:02}", git_flow.feature_prefix, phase);
1977 let commits = git_command(project_root)
1978 .args([
1979 "rev-list",
1980 "--count",
1981 &format!("{}..{branch}", git_flow.develop),
1982 ])
1983 .output()
1984 .ok()
1985 .and_then(|o| String::from_utf8_lossy(&o.stdout).trim().parse().ok())
1986 .unwrap_or(0);
1987
1988 let (status, reason) = if commits > 0 {
1989 (
1990 AgentStatus::Unknown,
1991 format!(
1992 "unverified — agent process is gone but {} commits exist on {}",
1993 commits, branch
1994 ),
1995 )
1996 } else {
1997 (
1998 AgentStatus::Failed,
1999 "no work accounted for — agent process is gone with no commits and no declared \
2000 external post-condition; human review needed"
2001 .to_string(),
2002 )
2003 };
2004
2005 Ok(AgentResult {
2006 status,
2007 exit_code: None,
2008 reason: Some(reason),
2009 commits: Some(commits),
2010 summary: None,
2011 verdict: None,
2012 decided_by_layer: Some(3),
2013 })
2014}
2015
2016/// Layer 0: run explicitly operator-approved external post-condition probes.
2017///
2018/// A failed probe outranks every agent-controlled signal. An approved,
2019/// all-passing set of declared probes is itself affirmative completion
2020/// evidence — `Success` — so a legitimately external-only stage with zero
2021/// commits can still complete cleanly (D-05 gap 2). Evaluated for EVERY
2022/// stage, not only Code (D-05 gap 1 / D-06). With no declarations (or when
2023/// disabled), behavior is byte-for-byte the pre-Phase-16 cascade.
2024///
2025/// Both DISCOVERY and probe EXECUTION read `execution_root` — the worktree
2026/// when one is set, `project_root` otherwise (999.76, ROADMAP criterion 6).
2027///
2028/// This knowingly OVERTURNS a recorded prior peer-review decision
2029/// (review Plan 03 MEDIUM, OpenCode). That decision held the two roots must
2030/// stay distinct, discovery reading `project_root` because
2031/// `.planning/phases/` "lives there, not in a worktree checkout". **The
2032/// premise has the direction backwards.** `.planning/` is TRACKED content,
2033/// so an in-flight phase's `{N}-PLAN.md` is committed on `feature/phase-{N}`
2034/// and therefore exists INSIDE the worktree while absent from the main checkout for
2035/// the phase's whole duration. Discovering from `project_root` meant a
2036/// correctly-declared probe set silently never ran in worktree mode —
2037/// DevFlow's default operating shape — with no error and no log, and the
2038/// "PLAN removed" veto below fired in its place. Recorded as an overturn
2039/// rather than patched quietly, so a later reader can see the direction was
2040/// reconsidered on evidence rather than overlooked.
2041///
2042/// Three sibling reads deliberately KEEP `project_root` and must not be
2043/// "corrected" to match: [`phase_commit_count`] (git worktrees share refs and
2044/// the object database, so counting from the main checkout is right), and
2045/// [`checkpoint_reported_in_capture`] and [`evaluate_layer1`] (both read the
2046/// stdout capture under `.devflow/`, which lives in the project root).
2047fn evaluate_layer0(
2048 project_root: &Path,
2049 state: &State,
2050 approved_commands: Option<&[String]>,
2051) -> Option<AgentResult> {
2052 if !crate::config::external_verify_enabled(project_root) {
2053 return None;
2054 }
2055
2056 let execution_root = state.worktree_path.as_deref().unwrap_or(project_root);
2057 let commands = crate::verify::external_verify_commands(execution_root, state.phase);
2058 if commands.is_empty() {
2059 return approved_commands.map(|_| AgentResult {
2060 status: AgentStatus::Failed,
2061 exit_code: None,
2062 reason: Some(
2063 "external verification approval mismatch; PLAN declaration was removed".into(),
2064 ),
2065 commits: None,
2066 summary: None,
2067 verdict: None,
2068 decided_by_layer: Some(0),
2069 });
2070 }
2071 let Some(approved_commands) = approved_commands else {
2072 return Some(AgentResult {
2073 status: AgentStatus::Failed,
2074 exit_code: None,
2075 reason: Some(format!(
2076 "external verification is not approved; set {} to the reviewed JSON command array",
2077 crate::verify::TRUST_EXTERNAL_VERIFY_ENV
2078 )),
2079 commits: None,
2080 summary: None,
2081 verdict: None,
2082 decided_by_layer: Some(0),
2083 });
2084 };
2085 if commands != approved_commands {
2086 return Some(AgentResult {
2087 status: AgentStatus::Failed,
2088 exit_code: None,
2089 reason: Some("external verification approval mismatch; PLAN commands changed".into()),
2090 commits: None,
2091 summary: None,
2092 verdict: None,
2093 decided_by_layer: Some(0),
2094 });
2095 }
2096 match commands
2097 .into_iter()
2098 .find(|command| !crate::verify::run_external_verification(command, execution_root))
2099 {
2100 Some(command) => Some(AgentResult {
2101 status: AgentStatus::Failed,
2102 exit_code: None,
2103 reason: Some(format!("external verification failed: {command}")),
2104 commits: None,
2105 summary: None,
2106 verdict: None,
2107 decided_by_layer: Some(0),
2108 }),
2109 // Every declared, approved probe passed — affirmative completion
2110 // evidence on its own (D-05 gap 2), even with zero commits.
2111 None => Some(AgentResult {
2112 status: AgentStatus::Success,
2113 exit_code: None,
2114 reason: Some(
2115 "external verification passed — all declared, approved probes succeeded".into(),
2116 ),
2117 commits: None,
2118 summary: None,
2119 verdict: None,
2120 decided_by_layer: Some(0),
2121 }),
2122 }
2123}
2124
2125/// Reconciles Layer 0's affirmative-success result with Layer 1's
2126/// self-reported verdict at `Stage::Validate` (18e).
2127///
2128/// Layer 0's affirmative-success arm above short-circuits the cascade before
2129/// Layer 1 ever runs (`evaluate_agent_result_inner` returns immediately on
2130/// any `Some(..)` from Layer 0), but Layer 1 is the ONLY carrier of a
2131/// `verdict` — `status` reports whether the stage's task ran; `verdict`
2132/// reports whether validation itself passed (see `AgentResult::verdict`'s
2133/// doc comment). At `Stage::Validate` that meant an agent's explicit
2134/// `verdict: pass` was silently discarded and `advance()` computed a failure
2135/// from it — a regression introduced by this project's own 17-03, fixed
2136/// here.
2137///
2138/// `decided_by_layer` deliberately stays `Some(0)` — Layer 0 still DECIDED
2139/// the `status`; Layer 1 only supplies the `verdict`. The CLI relies on that
2140/// value to tell an `external_verify` Validate apart from an ordinary one
2141/// (`classify_validate_outcome`, 18e).
2142///
2143/// Scoped to `Stage::Validate` only (flagged assumption in 18-05-PLAN.md): at
2144/// every other stage an affirmative Layer 0 success keeps `verdict: None`,
2145/// unchanged from current behavior. A Layer 0 FAILURE is never passed here —
2146/// only its affirmative-success arm is, so a failed probe still outranks
2147/// every agent-controlled signal.
2148///
2149/// 31-02 audit (non-exhaustive equality site 2 of 3). The `!= Success` guard
2150/// below is CORRECT AS-IS for `AgentStatus::IdleTimeout` and is left unchanged.
2151/// The compiler cannot flag an equality test against a new variant, so this is
2152/// audited by hand. An idle-timeout result is rejected here by BOTH independent
2153/// guards, not just one: its status is not `Success`, and its
2154/// `decided_by_layer` is `Some(1)` (the monitor's side-channel verdict is a
2155/// Layer-1-class fact), never `Some(0)`. It returns unchanged, which is right —
2156/// this function exists only to graft Layer 1's `verdict` onto an affirmative
2157/// Layer 0 probe success, and a timeout is neither.
2158///
2159/// # This function is 999.74's real defect site (D-15, ROADMAP criterion 4)
2160///
2161/// Until 34-01 the graft read Layer 1's `verdict` and nothing else. A marker of
2162/// `{"status":"failed","verdict":"pass"}` therefore produced `(Success,
2163/// Some(Pass), Some(0))`: an agent's self-reported FAILURE laundered into an
2164/// affirmative pair, which `outcome_policy::decide_action` advances and
2165/// `classify_validate_outcome` reads as `Passed` — Ship, in `Mode::Auto`, on a
2166/// run whose agent said it had failed. The status was never inspected, so
2167/// nothing downstream could see the contradiction; by the time the classifier
2168/// ran, the status genuinely WAS `Success`.
2169///
2170/// The fix consults Layer 1's own `AgentStatus` before transplanting its
2171/// verdict, because **a verdict attached to a self-reported failure is not a
2172/// pass**. Only `AgentStatus::Success` from Layer 1 may contribute a verdict;
2173/// everything else leaves `verdict: None` and the stage classifies `Ambiguous`,
2174/// which gates.
2175///
2176/// The classifier fix (plan 34-03, ROADMAP criterion 3) does **not** close this
2177/// and never could: gating `classify_validate_outcome`'s `Passed` arm on the
2178/// derived status passes cleanly here, because the derived status is `Success`.
2179/// Criterion 3 and criterion 4 are separate deliverables. Regression-pinned by
2180/// `layer0_verdict_graft_declines_when_layer1_status_is_not_success`, with
2181/// `layer0_verdict_graft_still_transplants_a_passing_layer1_verdict` as its
2182/// mandatory opposite-result control.
2183///
2184/// `evaluate_layer1` is called on `project_root`, NOT on the execution root,
2185/// and that asymmetry is deliberate rather than an oversight: Layer 1 reads the
2186/// stdout capture under `.devflow/`, which lives in the project root, while
2187/// Layer 0 above DISCOVERS declarations in `.planning/phases/` (project root)
2188/// and RUNS probes in the worktree. Plan 34-04 moves Layer 0's *discovery* to
2189/// the execution root; this call stays on `project_root` and is still correct
2190/// afterwards. Recorded here so a later reader does not "fix" the asymmetry.
2191fn reconcile_layer0_verdict(
2192 project_root: &Path,
2193 state: &State,
2194 result: AgentResult,
2195) -> AgentResult {
2196 if state.stage != Stage::Validate
2197 || result.status != AgentStatus::Success
2198 || result.decided_by_layer != Some(0)
2199 {
2200 return result;
2201 }
2202 let verdict = evaluate_layer1(project_root, state.phase)
2203 .filter(|layer1| layer1.status == AgentStatus::Success)
2204 .and_then(|layer1| layer1.verdict);
2205 AgentResult { verdict, ..result }
2206}
2207
2208/// Refuse to let a stream-derived `Success` outrank a contradicting exit code
2209/// (constraint 9's residual, T-31-15, 31-04).
2210///
2211/// # Why this cannot be a parser assertion
2212///
2213/// Constraint 9's items 1 and 2 — a torn line at or after the last surviving
2214/// top-level `result`, and provenance on verdict selection — were closed at the
2215/// root by the `a557805` refactor that made lossiness and capture kind
2216/// first-class ([`ParsedCapture`], [`classify`]). What survives is precisely
2217/// the case no parser can detect: **a capture cut at an exact line boundary is
2218/// byte-identical to a healthy shorter run.** There is nothing in the bytes to
2219/// assert on. The writer that died between flushing turn N and turn N+1 also
2220/// died non-zero, so the exit code is the only remaining signal — and it lives
2221/// one layer up, in the wiring, which is where this defence had to go.
2222///
2223/// # Why the fix is narrow rather than a cascade reordering
2224///
2225/// [`evaluate_agent_result_inner`] consults Layer 2 only when Layer 1 abstains,
2226/// which is why a Layer 1 `Success` wins over a contradicting exit code today.
2227/// That ordering is correct in the ordinary case: Layer 1 is authoritative
2228/// precisely so it does not need Layer 2's slower `git rev-list` fallback.
2229/// Making Layer 2 run first would trade a rare wrong answer for a slow one on
2230/// every stage. So this arbitrates one verdict rather than reordering anything.
2231///
2232/// # Scope
2233///
2234/// Fires ONLY on `AgentStatus::Success`. `RateLimited`, `IdleTimeout`,
2235/// `ResourceKilled`, `AgentUnavailable`, `Failed` and `Unknown` all return
2236/// untouched, each with a named test. Two of those exclusions are load-bearing
2237/// rather than tidy: a `RateLimited` downgraded to `Failed` would route the run
2238/// to a human gate instead of the auto-resume cron it needs, and an
2239/// `IdleTimeout` downgraded to `Failed` would erase the distinction plan 31-02
2240/// exists to create — 999.64 reborn inside its own fix.
2241///
2242/// 31-02 audit convention (non-exhaustive equality site): the `!= Success`
2243/// guard below is correct as-is for every current and future variant. Anything
2244/// that is not an affirmative claimed success has nothing to arbitrate, so
2245/// passing it through unchanged is the right default for a variant added later.
2246///
2247/// # `verdict: None` is load-bearing — do not carry it over for symmetry
2248///
2249/// `classify_validate_outcome` (`devflow-cli/src/pipeline_outcomes.rs`) matches
2250/// `(_, Some(Verdict::Pass)) => ValidateOutcome::Passed` FIRST, with `_`
2251/// discarding the status entirely. A downgraded result has no verdict to offer
2252/// and must not invent one. [`idle_timeout_result`] dodges the same trap the
2253/// same way, and says so. That instruction is unchanged and still binding.
2254///
2255/// **Correction (34-01, D-15).** An earlier version of this note went further
2256/// and claimed a kept `verdict: Pass` on a `status: Failed` "would still
2257/// classify Validate as **Passed**", making this function a no-op at Validate.
2258/// That overstated the reachability. `outcome_policy::decide_action` intercepts
2259/// every non-`Success` status and routes it to a gate BEFORE
2260/// `classify_validate_outcome` is ever reached, so THIS path is protected and
2261/// this function is not a no-op. The `verdict: None` above is defence in depth,
2262/// which is why it stays.
2263///
2264/// The route into the inversion that IS reachable is
2265/// [`reconcile_layer0_verdict`]'s graft — it produced `status: Success` with a
2266/// self-reported failure's verdict attached, so `decide_action` had nothing to
2267/// intercept. See that function's own doc comment for the full record. It is
2268/// closed in plan 34-01; the classifier's own structural fix (gating the
2269/// `Passed` arm on the derived status) lands in plan 34-03.
2270///
2271/// **999.74 / DEN-95** is therefore being CLOSED in Phase 34 rather than
2272/// deliberately deferred. The caution that motivated the earlier deferral still
2273/// applies to the classifier half and is discharged there, not here: changing
2274/// that match arm re-routes `Failed`, `Unknown` and `ResourceKilled`, so 34-03
2275/// audits all of them explicitly.
2276///
2277/// # Exit-code fidelity
2278///
2279/// 137 → `ResourceKilled` and 127 → `AgentUnavailable` are preserved rather
2280/// than collapsed into `Failed`, mirroring [`evaluate_layer2`] exactly:
2281/// `outcome_policy::decide_action` routes those two to `GateInfra` rather than
2282/// `GateReview`, and the same exit code must not reach two different operator
2283/// gates depending on whether a stale Layer 1 success happened to be present.
2284///
2285/// Note the `ResourceKilled` arm is currently **unreachable via the
2286/// `MonitorLaunch::PipeOwning` path**: `run_pipe_owning_monitor` records
2287/// `status.code().unwrap_or(-1)`, so a SIGKILLed child writes `-1`, not `137`.
2288/// Recorded rather than silently relabelling a real OOM as `Failed` — the arm
2289/// is still reachable from the `Legacy` arm's `sh` monitor, whose `$?` does
2290/// carry `128 + signal`.
2291///
2292/// Unreadable or unparseable exit-file content is tolerated exactly as
2293/// [`evaluate_layer2`] tolerates it — a missing file returns the result
2294/// unchanged (an absent file is not evidence of failure), and garbage parses to
2295/// `-1`. Neither is invented behaviour; both match the sibling reader.
2296fn reconcile_stream_success_against_exit_code(
2297 project_root: &Path,
2298 phase: u32,
2299 result: AgentResult,
2300) -> AgentResult {
2301 if result.status != AgentStatus::Success {
2302 return result;
2303 }
2304
2305 let Ok(raw) = std::fs::read_to_string(exit_code_path(project_root, phase)) else {
2306 return result;
2307 };
2308 let exit_code: i32 = raw.trim().parse().unwrap_or(-1);
2309 if exit_code == 0 {
2310 return result;
2311 }
2312
2313 let (status, lead) = if exit_code == 137 {
2314 (
2315 AgentStatus::ResourceKilled,
2316 format!(
2317 "the agent's output stream reported SUCCESS but the process was killed \
2318 (exit code {exit_code}, likely OOM)"
2319 ),
2320 )
2321 } else if exit_code == 127 {
2322 (
2323 AgentStatus::AgentUnavailable,
2324 format!(
2325 "the agent's output stream reported SUCCESS but the agent command was \
2326 unavailable (exit code {exit_code}, command not found)"
2327 ),
2328 )
2329 } else {
2330 (
2331 AgentStatus::Failed,
2332 format!(
2333 "the agent's output stream reported SUCCESS but the agent exited with \
2334 code {exit_code}"
2335 ),
2336 )
2337 };
2338
2339 AgentResult {
2340 status,
2341 exit_code: Some(exit_code),
2342 reason: Some(format!(
2343 "{lead}. A capture cut at an exact line boundary is byte-identical to a healthy \
2344 shorter run, so no parser assertion can tell the two apart — the exit code is the \
2345 only remaining signal, and it contradicts the claim. Review the phase branch before \
2346 deciding what to keep; nothing was rolled back."
2347 )),
2348 verdict: None,
2349 ..result
2350 }
2351}
2352
2353/// Full four-layer evaluation: returns the best available AgentResult.
2354pub fn evaluate_agent_result(
2355 project_root: &Path,
2356 state: &State,
2357 git_flow: &GitFlowConfig,
2358) -> Result<AgentResult, ResultError> {
2359 let approval = crate::verify::external_verification_approval();
2360 evaluate_agent_result_inner(project_root, state, git_flow, approval.as_deref())
2361}
2362
2363fn evaluate_agent_result_inner(
2364 project_root: &Path,
2365 state: &State,
2366 git_flow: &GitFlowConfig,
2367 approved_commands: Option<&[String]>,
2368) -> Result<AgentResult, ResultError> {
2369 // Layer 0: operator-authored external post-condition (authoritative failure)
2370 if let Some(result) = evaluate_layer0(project_root, state, approved_commands) {
2371 return Ok(reconcile_layer0_verdict(project_root, state, result));
2372 }
2373
2374 // Layer 1: DEVFLOW_RESULT marker (authoritative)
2375 //
2376 // Authoritative, but not unconditionally: a CLAIMED success is arbitrated
2377 // against the recorded exit code before it is returned (31-04, T-31-15).
2378 // The cascade below is deliberately NOT reordered — see
2379 // `reconcile_stream_success_against_exit_code` for why Layer 2 running
2380 // first would be the wrong trade.
2381 if let Some(result) = evaluate_layer1(project_root, state.phase) {
2382 return Ok(reconcile_stream_success_against_exit_code(
2383 project_root,
2384 state.phase,
2385 result,
2386 ));
2387 }
2388
2389 // Layer 2: Exit code + commit gate
2390 if let Some(result) = evaluate_layer2(project_root, state.phase, git_flow, state.stage)? {
2391 return Ok(result);
2392 }
2393
2394 // Layer 3: Process existence + commits
2395 evaluate_layer3(project_root, state.phase, git_flow)
2396}
2397
2398/// Path to the .devflow directory for a project root.
2399fn devflow_dir(project_root: &Path) -> PathBuf {
2400 project_root.join(".devflow")
2401}
2402
2403/// Path to the stdout file for a given phase.
2404pub fn stdout_path(project_root: &Path, phase: u32) -> PathBuf {
2405 devflow_dir(project_root).join(format!("phase-{:02}-stdout", phase))
2406}
2407
2408/// Path where the agent's stderr is captured for a given phase.
2409/// Lives alongside `stdout_path` under `.devflow/`.
2410pub fn stderr_path(project_root: &Path, phase: u32) -> PathBuf {
2411 devflow_dir(project_root).join(format!("phase-{phase:02}-stderr.log"))
2412}
2413
2414/// Path to the exit code file for a given phase.
2415pub fn exit_code_path(project_root: &Path, phase: u32) -> PathBuf {
2416 devflow_dir(project_root).join(format!("phase-{:02}-exit", phase))
2417}
2418
2419/// Path to the file where the monitor records the launched agent's PID.
2420pub fn agent_pid_path(project_root: &Path, phase: u32) -> PathBuf {
2421 devflow_dir(project_root).join(format!("phase-{:02}-agent-pid", phase))
2422}
2423
2424/// Path to the file holding the stage prompt handed to the pipe-owning
2425/// monitor (Phase 31).
2426///
2427/// The prompt travels `spawn_monitor` → detached monitor process as a FILE,
2428/// never as argv: DevFlow stage prompts are large and argv has a hard length
2429/// ceiling, so a prompt passed positionally would fail on exactly the
2430/// context-heavy stages that matter most.
2431pub fn prompt_path(project_root: &Path, phase: u32) -> PathBuf {
2432 devflow_dir(project_root).join(format!("phase-{:02}-prompt", phase))
2433}
2434
2435/// Path to the pipe-owning monitor's own log for a phase (Phase 31).
2436///
2437/// The monitor is a detached process whose stdio is not the operator's
2438/// terminal — anything it prints to its own stdout goes nowhere. Every "log
2439/// loudly" obligation in this phase (the D-04 idle-timeout clamp, the D-11
2440/// opt-out notice) writes here instead, so a loud message is actually
2441/// readable after the fact.
2442pub fn monitor_log_path(project_root: &Path, phase: u32) -> PathBuf {
2443 devflow_dir(project_root).join(format!("phase-{:02}-monitor.log", phase))
2444}
2445
2446/// Path to the pipe-owning monitor's idle-timeout verdict for a phase
2447/// (D-05/D-06, 31-02).
2448///
2449/// A SIDE CHANNEL, deliberately separate from the stdout capture: the capture
2450/// is the agent's own narration, and a verdict appended to it is shadowed by
2451/// any earlier genuine `result` event the stream already contained. See
2452/// [`parse_idle_timeout_side_channel`] — that separation is a correctness
2453/// requirement (T-31-06), not a filing convention.
2454///
2455/// Holds a JSON [`IdleTimeoutRecord`]. Written and fsynced by the monitor
2456/// BEFORE the child is signalled, so nothing can race the verdict.
2457pub fn idle_timeout_path(project_root: &Path, phase: u32) -> PathBuf {
2458 devflow_dir(project_root).join(format!("phase-{:02}-idle-timeout", phase))
2459}
2460
2461/// Path to the archived-capture-history directory for a phase (16b).
2462///
2463/// `.devflow/history/phase-NN/` holds retained per-stage capture generations
2464/// so a false-positive self-report can be diagnosed after the fact. Exposed
2465/// as a constructor (rather than inlined at each call site) so downstream
2466/// tooling (16h in 16-07's correlation, 16i in 16-05's enumeration) always
2467/// derives the path from here instead of hardcoding it.
2468pub fn history_dir(project_root: &Path, phase: u32) -> PathBuf {
2469 devflow_dir(project_root)
2470 .join("history")
2471 .join(format!("phase-{:02}", phase))
2472}
2473
2474/// Monotonically increasing tie-breaker appended to the nanosecond timestamp
2475/// used to stamp archived generations, so two archives issued within the
2476/// same nanosecond (possible in a tight test loop) never collide.
2477static ARCHIVE_SEQ: std::sync::atomic::AtomicU64 = std::sync::atomic::AtomicU64::new(0);
2478
2479/// A stamp unique within this process, used to name an archived generation.
2480/// The outgoing stage's name is not available at the `archive_phase_files`
2481/// call site (see `launch_stage` in main.rs), so a monotonic timestamp is
2482/// used instead — sufficient to order and identify generations.
2483fn archive_stamp() -> String {
2484 let nanos = std::time::SystemTime::now()
2485 .duration_since(std::time::UNIX_EPOCH)
2486 .map(|d| d.as_nanos())
2487 .unwrap_or(0);
2488 let seq = ARCHIVE_SEQ.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
2489 format!("{nanos}-{seq}")
2490}
2491
2492/// Archive the prior stage's stdout/exit captures into bounded per-phase
2493/// history instead of wiping them outright, so a false-positive self-report
2494/// can be diagnosed after the fact (16b). Replaces the old
2495/// `cleanup_phase_files`, which deleted these files unconditionally.
2496///
2497/// At most `retain` capture generations are kept per phase; older ones are
2498/// pruned (see [`prune_history`]). The agent-pid file is still removed
2499/// outright — it is process bookkeeping, not diagnostic output. When there
2500/// is nothing to archive (first launch), this is a no-op success.
2501pub fn archive_phase_files(
2502 project_root: &Path,
2503 evidence_root: &Path,
2504 phase: u32,
2505 retain: usize,
2506) -> Result<Option<String>, std::io::Error> {
2507 archive_phase_files_with_stamp(project_root, evidence_root, phase, retain, &archive_stamp())
2508}
2509
2510fn archive_phase_files_with_stamp(
2511 project_root: &Path,
2512 evidence_root: &Path,
2513 phase: u32,
2514 retain: usize,
2515 stamp: &str,
2516) -> Result<Option<String>, std::io::Error> {
2517 let _ = std::fs::remove_file(agent_pid_path(project_root, phase));
2518
2519 let stdout_src = stdout_path(project_root, phase);
2520 let exit_src = exit_code_path(project_root, phase);
2521 let stdout_exists = stdout_src.exists();
2522 let exit_exists = exit_src.exists();
2523 if !stdout_exists && !exit_exists {
2524 return Ok(None); // Nothing to archive — first launch.
2525 }
2526
2527 let history_dir = history_dir(project_root, phase);
2528 crate::workflow::ensure_devflow_dir(&history_dir)?;
2529
2530 let staging_dir = history_dir.join(format!(".pending-{stamp}"));
2531 std::fs::create_dir(&staging_dir)?;
2532 let stdout_stage = staging_dir.join("stdout");
2533 let exit_stage = staging_dir.join("exit");
2534 let review_stage = staging_dir.join("REVIEW.md");
2535 let stdout_dest = history_dir.join(format!("{stamp}-stdout"));
2536 let exit_dest = history_dir.join(format!("{stamp}-exit"));
2537 let review_dest = history_dir.join(format!("{stamp}-REVIEW.md"));
2538 let review_src = phase_review_path(evidence_root, phase);
2539
2540 let mut stdout_staged = false;
2541 let mut exit_staged = false;
2542 let mut stdout_published = false;
2543 let mut exit_published = false;
2544 let mut review_published = false;
2545
2546 let archive_result = (|| -> Result<(), std::io::Error> {
2547 if stdout_exists {
2548 std::fs::rename(&stdout_src, &stdout_stage)?;
2549 stdout_staged = true;
2550 }
2551 if exit_exists {
2552 std::fs::rename(&exit_src, &exit_stage)?;
2553 exit_staged = true;
2554 }
2555 if let Some(review) = &review_src {
2556 std::fs::copy(review, &review_stage)?;
2557 }
2558
2559 if stdout_exists {
2560 std::fs::rename(&stdout_stage, &stdout_dest)?;
2561 stdout_staged = false;
2562 stdout_published = true;
2563 }
2564 if exit_exists {
2565 std::fs::rename(&exit_stage, &exit_dest)?;
2566 exit_staged = false;
2567 exit_published = true;
2568 }
2569 if review_src.is_some() {
2570 std::fs::rename(&review_stage, &review_dest)?;
2571 review_published = true;
2572 }
2573 Ok(())
2574 })();
2575
2576 if let Err(error) = archive_result {
2577 let mut rollback_error = None;
2578 let mut restore = |from: &Path, to: &Path| {
2579 if let Err(error) = std::fs::rename(from, to)
2580 && rollback_error.is_none()
2581 {
2582 rollback_error = Some(error);
2583 }
2584 };
2585 if stdout_published {
2586 restore(&stdout_dest, &stdout_src);
2587 } else if stdout_staged {
2588 restore(&stdout_stage, &stdout_src);
2589 }
2590 if exit_published {
2591 restore(&exit_dest, &exit_src);
2592 } else if exit_staged {
2593 restore(&exit_stage, &exit_src);
2594 }
2595 if review_published {
2596 let _ = std::fs::remove_file(&review_dest);
2597 }
2598 let _ = std::fs::remove_dir_all(&staging_dir);
2599
2600 if let Some(rollback_error) = rollback_error {
2601 return Err(std::io::Error::new(
2602 error.kind(),
2603 format!("{error}; archive rollback failed: {rollback_error}"),
2604 ));
2605 }
2606 return Err(error);
2607 }
2608
2609 let _ = std::fs::remove_dir(&staging_dir);
2610
2611 prune_history(&history_dir, retain);
2612 Ok(Some(stamp.to_string()))
2613}
2614
2615fn phase_review_path(evidence_root: &Path, phase: u32) -> Option<PathBuf> {
2616 let phases = std::fs::read_dir(evidence_root.join(".planning/phases")).ok()?;
2617 let prefix = format!("{phase:02}-");
2618 for entry in phases.flatten() {
2619 if entry
2620 .file_name()
2621 .to_str()
2622 .is_some_and(|name| name.starts_with(&prefix))
2623 {
2624 let review = entry.path().join(format!("{phase:02}-REVIEW.md"));
2625 if review.exists() {
2626 return Some(review);
2627 }
2628 }
2629 }
2630 None
2631}
2632
2633/// Whether `/gsd-verify-work` has produced a `{phase:02}-VERIFICATION.md`
2634/// artifact for `phase` yet.
2635///
2636/// Per D-01 (33-CONTEXT.md), this is the sole mid-arc-vs-genuine-gaps signal
2637/// a Validate→Code loop-back consults: a phase with no verification artifact
2638/// is still mid-arc (its remaining plans have not been judged at all), so a
2639/// loop-back must re-run the phase in full rather than dispatch `--gaps-only`,
2640/// which matches zero plans and gates unresolvably. Mirrors
2641/// [`phase_review_path`]'s directory-prefix-scan idiom exactly, but returns a
2642/// `bool` — no caller needs the artifact's path, only whether it exists. A
2643/// missing `.planning/phases` directory returns `false` rather than panicking.
2644///
2645/// `evidence_root` is the root the Validate agent actually wrote to — the
2646/// phase's worktree when `state.worktree_path` is set, else the project root.
2647/// `.planning/` is tracked, so in worktree mode the artifact lands on
2648/// `feature/phase-N` and is invisible from the main checkout for the phase's
2649/// entire in-flight duration. Passing the project root in worktree mode is
2650/// exactly the defect this parameter name exists to prevent (33-CONTEXT.md
2651/// CR-01); it is NOT interchangeable with the root used for git reads such as
2652/// [`phase_commit_count`], whose refs and object database are shared across
2653/// worktrees and which therefore correctly takes the project root.
2654pub fn phase_verification_exists(evidence_root: &Path, phase: u32) -> bool {
2655 let Ok(phases) = std::fs::read_dir(evidence_root.join(".planning/phases")) else {
2656 return false;
2657 };
2658 let prefix = format!("{phase:02}-");
2659 for entry in phases.flatten() {
2660 if entry
2661 .file_name()
2662 .to_str()
2663 .is_some_and(|name| name.starts_with(&prefix))
2664 {
2665 let verification = entry.path().join(format!("{phase:02}-VERIFICATION.md"));
2666 if verification.exists() {
2667 return true;
2668 }
2669 }
2670 }
2671 false
2672}
2673
2674/// Keep only the newest `retain` capture generations under `history_dir`,
2675/// deleting older ones. Generations are grouped by their stamp (the shared
2676/// prefix of a `{stamp}-stdout`/`{stamp}-exit` pair, split off the trailing
2677/// `-stdout`/`-exit` suffix via `rsplit_once`) and ordered lexicographically,
2678/// which matches numeric/chronological order for the fixed-width nanosecond
2679/// stamps `archive_stamp` produces. Ordering parses both numeric components;
2680/// the process-local sequence is intentionally not fixed-width.
2681fn prune_history(history_dir: &Path, retain: usize) {
2682 let Ok(entries) = std::fs::read_dir(history_dir) else {
2683 return;
2684 };
2685
2686 let mut stamps: Vec<String> = entries
2687 .flatten()
2688 .filter_map(|entry| {
2689 let name = entry.file_name().to_str()?.to_string();
2690 name.rsplit_once('-')
2691 .map(|(stamp, _suffix)| stamp.to_string())
2692 })
2693 .collect();
2694 stamps.sort_by_key(|stamp| {
2695 let mut parts = stamp.split('-');
2696 let nanos = parts
2697 .next()
2698 .and_then(|part| part.parse::<u128>().ok())
2699 .unwrap_or(0);
2700 let sequence = parts
2701 .next()
2702 .and_then(|part| part.parse::<u64>().ok())
2703 .unwrap_or(0);
2704 (nanos, sequence)
2705 });
2706 stamps.dedup();
2707
2708 if stamps.len() <= retain {
2709 return;
2710 }
2711
2712 let to_remove = stamps.len() - retain;
2713 for stamp in &stamps[..to_remove] {
2714 let _ = std::fs::remove_file(history_dir.join(format!("{stamp}-stdout")));
2715 let _ = std::fs::remove_file(history_dir.join(format!("{stamp}-exit")));
2716 let _ = std::fs::remove_file(history_dir.join(format!("{stamp}-REVIEW.md")));
2717 }
2718}
2719
2720#[cfg(test)]
2721mod tests {
2722 use super::*;
2723 use crate::config::GitFlowConfig;
2724 use crate::mode::Mode;
2725 use crate::stage::Stage;
2726 use crate::state::{AgentKind, State};
2727
2728 fn state_in(root: &Path, phase: u32) -> State {
2729 let mut state = State::new(phase, AgentKind::Claude, Mode::Auto, root.to_path_buf());
2730 state.stage = Stage::Code;
2731 state
2732 }
2733
2734 fn git(root: &Path, args: &[&str]) {
2735 let output = crate::test_support::git_command(root)
2736 .args(args)
2737 .output()
2738 .unwrap();
2739 assert!(
2740 output.status.success(),
2741 "git {:?} failed\nstdout: {}\nstderr: {}",
2742 args,
2743 String::from_utf8_lossy(&output.stdout),
2744 String::from_utf8_lossy(&output.stderr)
2745 );
2746 }
2747
2748 fn init_repo_with_feature_commit(root: &Path, phase: u32) {
2749 git(root, &["init"]);
2750 git(root, &["config", "user.email", "devflow@example.com"]);
2751 git(root, &["config", "user.name", "DevFlow Tests"]);
2752 git(root, &["config", "commit.gpgsign", "false"]);
2753 git(root, &["config", "tag.gpgsign", "false"]);
2754 git(root, &["config", "core.hooksPath", "/dev/null"]);
2755 git(root, &["checkout", "-b", "develop"]);
2756 std::fs::write(root.join("README.md"), "base\n").unwrap();
2757 git(root, &["add", "README.md"]);
2758 git(root, &["commit", "-m", "base"]);
2759
2760 let branch = format!("feature/phase-{phase:02}");
2761 git(root, &["checkout", "-b", &branch]);
2762 std::fs::write(root.join("phase.txt"), "feature work\n").unwrap();
2763 git(root, &["add", "phase.txt"]);
2764 git(root, &["commit", "-m", "feature work"]);
2765 }
2766
2767 /// Like `init_repo_with_feature_commit`, but the feature branch sits at
2768 /// develop's tip with **no** extra commit (0 commits ahead).
2769 fn init_repo_with_feature_no_commit(root: &Path, phase: u32) {
2770 git(root, &["init"]);
2771 git(root, &["config", "user.email", "devflow@example.com"]);
2772 git(root, &["config", "user.name", "DevFlow Tests"]);
2773 git(root, &["config", "commit.gpgsign", "false"]);
2774 git(root, &["config", "tag.gpgsign", "false"]);
2775 git(root, &["config", "core.hooksPath", "/dev/null"]);
2776 git(root, &["checkout", "-b", "develop"]);
2777 std::fs::write(root.join("README.md"), "base\n").unwrap();
2778 git(root, &["add", "README.md"]);
2779 git(root, &["commit", "-m", "base"]);
2780
2781 let branch = format!("feature/phase-{phase:02}");
2782 git(root, &["checkout", "-b", &branch]);
2783 }
2784
2785 #[test]
2786 fn parse_success_marker() {
2787 let stdout = "some output\nDEVFLOW_RESULT: {\"status\":\"success\"}\n";
2788 let result = parse_devflow_result(stdout).unwrap();
2789 assert_eq!(result.status, AgentStatus::Success);
2790 }
2791
2792 #[test]
2793 fn parse_failed_marker_with_reason() {
2794 let stdout =
2795 "work done\nDEVFLOW_RESULT: {\"status\":\"failed\",\"reason\":\"clippy errors\"}\n";
2796 let result = parse_devflow_result(stdout).unwrap();
2797 assert_eq!(result.status, AgentStatus::Failed);
2798 assert_eq!(result.reason.unwrap(), "clippy errors");
2799 }
2800
2801 #[test]
2802 fn parse_missing_marker_returns_none() {
2803 let stdout = "just some output\nno marker here\n";
2804 assert!(parse_devflow_result(stdout).is_none());
2805 }
2806
2807 #[test]
2808 fn parse_malformed_json_returns_none() {
2809 let stdout = "DEVFLOW_RESULT: {not valid json}\n";
2810 assert!(parse_devflow_result(stdout).is_none());
2811 }
2812
2813 #[test]
2814 fn parse_lowercase_marker() {
2815 let stdout = "devflow_result: {\"status\":\"success\"}\n";
2816 let result = parse_devflow_result(stdout).unwrap();
2817 assert_eq!(result.status, AgentStatus::Success);
2818 }
2819
2820 #[test]
2821 fn parse_marker_without_space_after_colon() {
2822 let stdout = "DEVFLOW_RESULT:{\"status\":\"success\"}\n";
2823 let result = parse_devflow_result(stdout).unwrap();
2824 assert_eq!(result.status, AgentStatus::Success);
2825 }
2826
2827 #[test]
2828 fn parse_lowercase_no_space_marker() {
2829 // Lowercase prefix AND no space after the colon — the combination that
2830 // the Phase 6 review flagged as uncovered.
2831 let stdout = "devflow_result:{\"status\":\"success\"}\n";
2832 let result = parse_devflow_result(stdout).unwrap();
2833 assert_eq!(result.status, AgentStatus::Success);
2834 }
2835
2836 #[test]
2837 fn parse_finds_last_marker_in_tail() {
2838 // Multiple markers — should find the last one.
2839 let stdout = "DEVFLOW_RESULT: {\"status\":\"failed\"}\nsome more output\nDEVFLOW_RESULT: {\"status\":\"success\"}\n";
2840 let result = parse_devflow_result(stdout).unwrap();
2841 assert_eq!(result.status, AgentStatus::Success);
2842 }
2843
2844 #[test]
2845 fn parse_marker_lines_returns_last_marker_in_long_output() {
2846 let stdout = format!(
2847 "{}\nDEVFLOW_RESULT: {{\"status\":\"failed\"}}\n{}\n\
2848 DEVFLOW_RESULT: {{\"status\":\"success\"}}\n",
2849 "prefix".repeat(900),
2850 "tail output".repeat(100)
2851 );
2852
2853 let result = parse_marker_lines(&stdout).unwrap();
2854
2855 assert_eq!(result.status, AgentStatus::Success);
2856 }
2857
2858 #[test]
2859 fn parse_marker_only_in_last_4000_chars() {
2860 // Marker beyond 4000 chars from end should not be found.
2861 let prefix = "a".repeat(5000);
2862 let stdout = format!("DEVFLOW_RESULT: {{\"status\":\"success\"}}\n{prefix}");
2863 assert!(parse_devflow_result(&stdout).is_none());
2864 }
2865
2866 #[test]
2867 fn parse_marker_with_commits_and_summary() {
2868 let stdout = r#"DEVFLOW_RESULT: {"status":"success","commits":3,"summary":"added tests"}"#;
2869 let result = parse_devflow_result(stdout).unwrap();
2870 assert_eq!(result.status, AgentStatus::Success);
2871 assert_eq!(result.commits, Some(3));
2872 assert_eq!(result.summary.unwrap(), "added tests");
2873 }
2874
2875 #[test]
2876 fn parse_marker_inside_json_result_envelope() {
2877 // Claude --output-format json wraps the final text in a `result` field
2878 // with embedded newlines escaped.
2879 let stdout = r#"{"type":"result","subtype":"success","result":"All done.\nDEVFLOW_RESULT: {\"status\": \"success\", \"commits\": 2}","session_id":"abc"}"#;
2880 let result = parse_devflow_result(stdout).unwrap();
2881 assert_eq!(result.status, AgentStatus::Success);
2882 assert_eq!(result.commits, Some(2));
2883 }
2884
2885 #[test]
2886 fn parse_failed_marker_inside_json_envelope() {
2887 let stdout = r#"{"result":"work\nDEVFLOW_RESULT: {\"status\": \"failed\", \"reason\": \"tests failed\"}"}"#;
2888 let result = parse_devflow_result(stdout).unwrap();
2889 assert_eq!(result.status, AgentStatus::Failed);
2890 assert_eq!(result.reason.unwrap(), "tests failed");
2891 }
2892
2893 #[test]
2894 fn parse_json_envelope_without_marker_returns_none() {
2895 let stdout = r#"{"result":"did some work but forgot the marker","session_id":"x"}"#;
2896 assert!(parse_devflow_result(stdout).is_none());
2897 }
2898
2899 #[test]
2900 fn detect_claude_json_rate_limit_by_subtype() {
2901 let stdout = r#"{"type":"result","subtype":"error_rate_limit","retry_after":"2026-06-18T15:45:30Z","result":"rate limited"}"#;
2902 assert_eq!(
2903 detect_rate_limit(stdout).as_deref(),
2904 Some("2026-06-18T15:45:30Z")
2905 );
2906 }
2907
2908 #[test]
2909 fn detect_claude_json_rate_limit_by_429() {
2910 let stdout = r#"{"type":"result","api_error_status":429,"error":{"message":"Too many requests. Try later."}}"#;
2911 assert_eq!(
2912 detect_rate_limit(stdout).as_deref(),
2913 Some("Too many requests. Try later.")
2914 );
2915 }
2916
2917 #[test]
2918 fn detect_codex_try_again_rate_limit() {
2919 let stdout = "Usage limit reached. Try again at 3:45 PM.\n";
2920 assert_eq!(detect_rate_limit(stdout).as_deref(), Some("3:45 PM"));
2921 }
2922
2923 /// WR-12 (13-REVIEW.md), revised: `json_has_str`/`json_has_i64`/
2924 /// `json_find_key` run on the coding agent's raw stdout via
2925 /// `detect_claude_rate_limit`, which every `devflow advance` invocation
2926 /// goes through. Deeply nested JSON — accidental or adversarial — must
2927 /// not stack-overflow the process, and a real marker at any depth
2928 /// serde_json will parse (its default recursion limit is exactly 128)
2929 /// must still be FOUND — the first WR-12 fix capped traversal at 64 and
2930 /// silently misclassified rate-limit markers at depths 64–128.
2931 #[test]
2932 fn detect_rate_limit_finds_marker_in_deeply_nested_json_without_overflow() {
2933 // 100 levels: parseable by serde_json (limit 128), deeper than the
2934 // removed 64-level traversal cap that used to hide the marker.
2935 const DEPTH: usize = 100;
2936 let mut stdout = String::new();
2937 for _ in 0..DEPTH {
2938 stdout.push_str(r#"{"nested":"#);
2939 }
2940 stdout.push_str(r#"{"type":"result","subtype":"error_rate_limit","retry_after":"deep"}"#);
2941 for _ in 0..DEPTH {
2942 stdout.push('}');
2943 }
2944
2945 // Must return promptly without crashing AND find the buried marker —
2946 // the iterative worklist traversal has no silent-miss window.
2947 assert_eq!(detect_rate_limit(&stdout).as_deref(), Some("deep"));
2948 }
2949
2950 #[test]
2951 fn detect_rate_limit_ignores_normal_stdout() {
2952 let stdout = "implemented feature\nDEVFLOW_RESULT: {\"status\":\"success\"}\n";
2953 assert!(detect_rate_limit(stdout).is_none());
2954 }
2955
2956 #[test]
2957 fn claude_envelope_is_error_detected() {
2958 let stdout = r#"{"type":"result","subtype":"error","is_error":true,"num_turns":2,"result":"tool call failed","session_id":"abc"}"#;
2959 let result = detect_claude_envelope_failure(stdout).unwrap();
2960 assert_eq!(result.status, AgentStatus::Failed);
2961 }
2962
2963 #[test]
2964 fn claude_is_error_overrides_success_marker() {
2965 let dir = tempfile::tempdir().unwrap();
2966 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
2967 std::fs::write(
2968 stdout_path(dir.path(), 9),
2969 r#"{"type":"result","is_error":true,"num_turns":3,"result":"oops\nDEVFLOW_RESULT: {\"status\":\"success\"}","session_id":"abc"}"#,
2970 )
2971 .unwrap();
2972
2973 let result = evaluate_layer1(dir.path(), 9).unwrap();
2974
2975 assert_eq!(result.status, AgentStatus::Failed);
2976 }
2977
2978 #[test]
2979 fn claude_envelope_is_error_false_defers() {
2980 let stdout = r#"{"type":"result","is_error":false,"num_turns":1,"result":"did some work","session_id":"abc"}"#;
2981 assert!(detect_claude_envelope_failure(stdout).is_none());
2982 }
2983
2984 #[test]
2985 fn claude_envelope_marker_still_wins() {
2986 let stdout = r#"{"type":"result","is_error":false,"result":"done\nDEVFLOW_RESULT: {\"status\":\"success\",\"commits\":2}","session_id":"abc"}"#;
2987 assert!(detect_claude_envelope_failure(stdout).is_none());
2988 let result = parse_devflow_result(stdout).unwrap();
2989 assert_eq!(result.status, AgentStatus::Success);
2990 assert_eq!(result.commits, Some(2));
2991 }
2992
2993 #[test]
2994 fn session_id_reads_top_level_string() {
2995 let stdout = r#"{"type":"result","subtype":"success","result":"All done.","session_id":"cf29bfec-69e8-45df-a4f3-3da08ab6f66e"}"#;
2996 assert_eq!(
2997 claude_session_id(stdout).as_deref(),
2998 Some("cf29bfec-69e8-45df-a4f3-3da08ab6f66e")
2999 );
3000 }
3001
3002 /// T-28-04 forgery guard: the embedded `DEVFLOW_RESULT` marker carries a
3003 /// DIFFERENT session id than the envelope's own top-level key. The
3004 /// top-level id must win — an agent must not be able to redirect which
3005 /// session DevFlow resumes into by planting its own `session_id` inside
3006 /// its self-authored marker JSON.
3007 #[test]
3008 fn session_id_in_devflow_result_marker_is_not_returned() {
3009 let stdout = r#"{"type":"result","subtype":"success","result":"All done.\nDEVFLOW_RESULT: {\"status\": \"success\", \"session_id\": \"forged-by-agent\"}","session_id":"real-top-level-id"}"#;
3010 assert_eq!(
3011 claude_session_id(stdout).as_deref(),
3012 Some("real-top-level-id")
3013 );
3014 }
3015
3016 #[test]
3017 fn session_id_plain_text_stdout_returns_none() {
3018 let stdout = "just some plain text output, not JSON\n";
3019 assert!(claude_session_id(stdout).is_none());
3020 }
3021
3022 #[test]
3023 fn session_id_missing_key_returns_none() {
3024 let stdout = r#"{"type":"result","result":"done, no session key"}"#;
3025 assert!(claude_session_id(stdout).is_none());
3026 }
3027
3028 #[test]
3029 fn session_id_non_string_type_returns_none_not_panic() {
3030 let stdout = r#"{"type":"result","result":"done","session_id":12345}"#;
3031 assert!(claude_session_id(stdout).is_none());
3032 }
3033
3034 #[test]
3035 fn session_id_from_capture_missing_file_returns_none() {
3036 let dir = tempfile::tempdir().unwrap();
3037 assert!(session_id_from_capture(dir.path(), 42).is_none());
3038 }
3039
3040 #[test]
3041 fn session_id_from_capture_lossy_reads_invalid_utf8() {
3042 let dir = tempfile::tempdir().unwrap();
3043 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
3044 let mut bytes = br#"{"type":"result","result":"done "#.to_vec();
3045 bytes.push(0xFF); // invalid UTF-8 byte
3046 bytes.extend_from_slice(br#"","session_id":"lossy-ok"}"#);
3047 std::fs::write(stdout_path(dir.path(), 5), bytes).unwrap();
3048
3049 assert_eq!(
3050 session_id_from_capture(dir.path(), 5).as_deref(),
3051 Some("lossy-ok")
3052 );
3053 }
3054
3055 /// Positive fixture built from RESEARCH's *predicted* `**Gate:**`
3056 /// rendering (a bare, un-spanned value). Kept as a tolerated shape, but
3057 /// note this is NOT what a real run emits — see
3058 /// `blocking_human_checkpoint_reported_matches_live_observed_rendering`
3059 /// for the rendering actually captured on 2026-07-31, which this
3060 /// prediction missed.
3061 #[test]
3062 fn blocking_human_checkpoint_reported_detects_human_gate_line() {
3063 let stdout = format!(
3064 "## CHECKPOINT REACHED\n\n**Type:** human-verify\n**Gate:** {HUMAN_GATE_VALUE} — copy the task's `gate` attribute verbatim so the orchestrator's carve-out sees it\n"
3065 );
3066 assert!(blocking_human_checkpoint_reported(&stdout));
3067 }
3068
3069 /// The Phase 26 near-miss distinction: a plain `blocking` gate must NOT
3070 /// be classified as a human-blocking checkpoint. `PLAIN_GATE_VALUE` is
3071 /// local to this test (not a module-level const) — it has no production
3072 /// use, only this negative fixture's.
3073 #[test]
3074 fn blocking_human_checkpoint_reported_false_for_plain_blocking() {
3075 const PLAIN_GATE_VALUE: &str = "blocking";
3076 let stdout = format!(
3077 "## CHECKPOINT REACHED\n\n**Type:** human-verify\n**Gate:** {PLAIN_GATE_VALUE} — copy the task's `gate` attribute verbatim so the orchestrator's carve-out sees it\n"
3078 );
3079 assert!(!blocking_human_checkpoint_reported(&stdout));
3080 }
3081
3082 #[test]
3083 fn blocking_human_checkpoint_reported_false_when_no_gate_field() {
3084 let stdout = "some ordinary agent failure output, no checkpoint at all\n";
3085 assert!(!blocking_human_checkpoint_reported(stdout));
3086 }
3087
3088 /// The `Gate:` line arrives inside an escaped Claude JSON result
3089 /// envelope's `result` field — must be found via the unescaped inner
3090 /// text, not the raw (escaped) JSON string.
3091 #[test]
3092 fn blocking_human_checkpoint_reported_true_inside_escaped_envelope() {
3093 let inner = format!(
3094 "## CHECKPOINT REACHED\\n\\n**Gate:** {HUMAN_GATE_VALUE} — copy the task's `gate` attribute verbatim so the orchestrator's carve-out sees it\\n"
3095 );
3096 let stdout = format!(
3097 r#"{{"type":"result","subtype":"success","result":"{inner}","session_id":"abc"}}"#
3098 );
3099 assert!(blocking_human_checkpoint_reported(&stdout));
3100 }
3101
3102 #[test]
3103 fn blocking_human_checkpoint_reported_tolerates_whitespace_and_emphasis() {
3104 let stdout = format!(" **Gate:** {HUMAN_GATE_VALUE} \n");
3105 assert!(blocking_human_checkpoint_reported(&stdout));
3106 }
3107
3108 /// REGRESSION — the rendering a real headless run actually produces.
3109 ///
3110 /// Transcribed verbatim from `.devflow/phase-91-stdout` of the live A1
3111 /// run on 2026-07-31 (a genuine `gate="blocking-human"` task driven
3112 /// through DevFlow's own monitor). The value arrives as a markdown CODE
3113 /// SPAN, not the bare token RESEARCH.md predicted.
3114 ///
3115 /// Before the backtick was added to `text_reports_human_gate`'s trim set
3116 /// this returned `false`: the leading backtick survived the trim, so the
3117 /// value `take_while` terminated at once and yielded an empty token. A
3118 /// real checkpoint was therefore never recognized, and the run fell
3119 /// through to the generic gate. If this test ever goes red, DevFlow has
3120 /// stopped recognizing real checkpoints — do not "fix" it by relaxing
3121 /// the assertion.
3122 #[test]
3123 fn blocking_human_checkpoint_reported_matches_live_observed_rendering() {
3124 let stdout = format!(
3125 "---\n\n## Checkpoint: Decision\n\n**Plan:** 91-01 Emit the checkpoint\n**Gate:** `{HUMAN_GATE_VALUE}`\n**Progress:** 0/1 tasks complete\n**Task:** Task 1 — Ask the operator to authorize writing the marker file\n"
3126 );
3127 assert!(
3128 blocking_human_checkpoint_reported(&stdout),
3129 "the live-observed code-span rendering must be recognized; \
3130 a false negative here means real checkpoints fall through to \
3131 the generic gate (the 2026-07-31 A1 defect)"
3132 );
3133 }
3134
3135 /// The same live rendering as it actually crosses into DevFlow's capture:
3136 /// escaped inside the Claude JSON result envelope. This is the exact
3137 /// path `checkpoint_reported_in_capture` reads in production.
3138 #[test]
3139 fn blocking_human_checkpoint_reported_matches_live_rendering_in_envelope() {
3140 let inner = format!(
3141 "## Checkpoint: Decision\\n\\n**Gate:** `{HUMAN_GATE_VALUE}`\\n**Progress:** 0/1 tasks complete\\n"
3142 );
3143 let stdout = format!(
3144 r#"{{"type":"result","subtype":"success","result":"{inner}","session_id":"live-a1"}}"#
3145 );
3146 assert!(
3147 blocking_human_checkpoint_reported(&stdout),
3148 "the code-span rendering must also be found inside the escaped envelope"
3149 );
3150 }
3151
3152 /// The backtick tolerance must not erode the Phase 26 near-miss
3153 /// distinction: a code-spanned PLAIN `blocking` gate is still not a
3154 /// human-blocking checkpoint.
3155 #[test]
3156 fn blocking_human_checkpoint_reported_false_for_code_spanned_plain_blocking() {
3157 let stdout = "## Checkpoint: Decision\n\n**Gate:** `blocking`\n";
3158 assert!(!blocking_human_checkpoint_reported(stdout));
3159 }
3160
3161 #[test]
3162 fn checkpoint_reported_in_capture_missing_file_returns_false() {
3163 let dir = tempfile::tempdir().unwrap();
3164 assert!(!checkpoint_reported_in_capture(dir.path(), 42));
3165 }
3166
3167 #[test]
3168 fn checkpoint_reported_in_capture_reads_true_from_file() {
3169 let dir = tempfile::tempdir().unwrap();
3170 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
3171 std::fs::write(
3172 stdout_path(dir.path(), 11),
3173 format!("**Gate:** {HUMAN_GATE_VALUE}\n"),
3174 )
3175 .unwrap();
3176 assert!(checkpoint_reported_in_capture(dir.path(), 11));
3177 }
3178
3179 // ---- stream-capture gate scoping (plan 30-05) --------------------------
3180 //
3181 // Fixtures for this cluster live with the other v3 envelopes further down:
3182 // `V3_USER_EVENT`, `V3_ASSISTANT_TOP_LEVEL_EVENT`,
3183 // `V3_ASSISTANT_SUBAGENT_EVENT`, `gate_declaration_text` and
3184 // `gate_documenting_text`. Read their doc comments before adding a case —
3185 // they record which capture line each envelope came from and that every
3186 // gate payload is synthetic.
3187 //
3188 // Each negative asserts a NEGATIVE CONTROL first: `text_reports_human_gate`
3189 // must still match the raw capture. Without it a negative would also pass
3190 // against a fixture that simply contains no gate text, and would keep
3191 // passing if someone deleted the gate line from the fixture.
3192
3193 /// **REGRESSION — review constraint 3, the prompt-echo false positive.**
3194 ///
3195 /// Under a single-document envelope the only place gate text can appear is
3196 /// the one `result` field the agent authored, so scanning raw stdout is
3197 /// safe. A stream capture breaks that invariant: text DevFlow never
3198 /// authored is echoed back into the same stdout, and a substring scan
3199 /// cannot tell which event it is inside.
3200 ///
3201 /// A failure here means a checkpoint auto-decide can fire, or the resume
3202 /// ceiling be consumed, on a stage whose prompt merely DISCUSSED
3203 /// checkpoints — and DevFlow's own planning documents are exactly that kind
3204 /// of prompt content.
3205 #[test]
3206 fn blocking_human_checkpoint_reported_false_for_gate_text_in_user_event() {
3207 let capture = stream_capture_of(&[
3208 &v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3209 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3210 ]);
3211 assert!(
3212 text_reports_human_gate(&capture),
3213 "negative control: the raw capture must still contain matchable \
3214 gate text, or this test asserts nothing"
3215 );
3216 assert!(
3217 !blocking_human_checkpoint_reported(&capture),
3218 "gate text inside a `user` event is echoed input, not an agent \
3219 declaration (review constraint 3)"
3220 );
3221 }
3222
3223 /// **REGRESSION — T-30-22.** Gate text forwarded from a subagent is not the
3224 /// orchestrator declaring a gate. Mistaking subagent-forwarded content for
3225 /// orchestrator output is the error that invalidated the v1 experiment
3226 /// outright and got its whole capture discarded.
3227 ///
3228 /// Two independent guards reject this event — the type filter (it is an
3229 /// `assistant` event) and the provenance filter (its `parent_tool_use_id`
3230 /// is non-null). The case is kept even though either alone suffices: they
3231 /// are meant to fail independently, so a future widening of the type filter
3232 /// cannot silently inherit subagent content.
3233 #[test]
3234 fn blocking_human_checkpoint_reported_false_for_subagent_forwarded_gate_text() {
3235 let capture = stream_capture_of(&[
3236 &v3_message_event(V3_ASSISTANT_SUBAGENT_EVENT, &gate_documenting_text()),
3237 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3238 ]);
3239 assert!(
3240 text_reports_human_gate(&capture),
3241 "negative control: the raw capture must still contain matchable \
3242 gate text, or this test asserts nothing"
3243 );
3244 assert!(
3245 !blocking_human_checkpoint_reported(&capture),
3246 "a subagent's forwarded text cannot present as the orchestrator's \
3247 own gate declaration"
3248 );
3249 }
3250
3251 /// **REGRESSION — T-30-26, the surface cross-AI review found.**
3252 ///
3253 /// Narration is not declaration. The envelope is v3 line 6, whose captured
3254 /// payload is mid-turn narration appearing in NO `result` event of the
3255 /// capture — so admitting top-level `assistant` events would add a
3256 /// genuinely new trusted surface, not merely duplicate the result text.
3257 /// An agent narrating the checkpoint task it is about to work on would then
3258 /// trip the gate: the prompt-echo false positive, recreated one layer
3259 /// inward.
3260 ///
3261 /// Nothing observed is lost by excluding the class: turn-FINAL assistant
3262 /// text is duplicated verbatim into the `result` event that follows it
3263 /// (v3 lines 17→19, 36→37, 53→54).
3264 #[test]
3265 fn blocking_human_checkpoint_reported_false_for_top_level_assistant_narration() {
3266 let capture = stream_capture_of(&[
3267 &v3_message_event(V3_ASSISTANT_TOP_LEVEL_EVENT, &gate_documenting_text()),
3268 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3269 ]);
3270 assert!(
3271 text_reports_human_gate(&capture),
3272 "negative control: the raw capture must still contain matchable \
3273 gate text, or this test asserts nothing"
3274 );
3275 assert!(
3276 !blocking_human_checkpoint_reported(&capture),
3277 "intermediate assistant narration discussing a gate is not a live \
3278 gate declaration"
3279 );
3280 }
3281
3282 /// The positive that stops the scoping from degenerating into always-false
3283 /// — which would pass every negative above while silently dropping every
3284 /// real human authorization request (T-30-24).
3285 #[test]
3286 fn blocking_human_checkpoint_reported_true_for_top_level_result_declaration() {
3287 let capture = stream_capture_of(&[
3288 &v3_message_event(V3_USER_EVENT, "Execute the plan."),
3289 &v3_result_event(V3_RESULT_TURN1, &gate_declaration_text()),
3290 ]);
3291 assert!(
3292 blocking_human_checkpoint_reported(&capture),
3293 "a gate declared in a top-level `result` event's own result text \
3294 must still be detected under a stream capture"
3295 );
3296 }
3297
3298 /// **T-30-27.** Detection asks whether a gate fired ANYWHERE in the stage,
3299 /// so it deliberately does NOT inherit plan 30-01's last-result-wins
3300 /// verdict semantics. A gate declared in turn 1 followed by
3301 /// task-notification wake-up turns — the exact turn shape the v3 capture
3302 /// archives — must not be dropped in favour of the later, silent results.
3303 ///
3304 /// Losing a checkpoint report is the opposite-direction harm from the false
3305 /// positive this plan closes, and the worse of the two: it silently drops a
3306 /// request for human authorization to the generic gate.
3307 #[test]
3308 fn blocking_human_checkpoint_reported_true_when_only_first_result_declares_gate() {
3309 let capture = v3_stream_capture(&gate_declaration_text(), NO_MARKER, NO_MARKER);
3310 assert!(
3311 blocking_human_checkpoint_reported(&capture),
3312 "detection must scan every top-level `result` event, not only the \
3313 last one"
3314 );
3315 }
3316
3317 /// The overcorrection guard: an echo and a genuine declaration can coexist
3318 /// in one capture, and the scoping must resolve per event rather than
3319 /// suppressing any capture that contains an echo.
3320 #[test]
3321 fn blocking_human_checkpoint_reported_true_when_echo_co_occurs_with_declaration() {
3322 let capture = stream_capture_of(&[
3323 &v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3324 &v3_result_event(V3_RESULT_TURN1, &gate_declaration_text()),
3325 ]);
3326 assert!(
3327 blocking_human_checkpoint_reported(&capture),
3328 "an echoed prompt in the same capture must not suppress a genuine \
3329 declaration"
3330 );
3331 }
3332
3333 /// The same scoping, proven on the path production actually consumes —
3334 /// `checkpoint_reported_in_capture` reading `.devflow/phase-NN-stdout` from
3335 /// disk. Both directions are asserted in one test on purpose: the negative
3336 /// alone cannot distinguish correct scoping from a wrapper that stopped
3337 /// reading the file at all.
3338 #[test]
3339 fn checkpoint_reported_in_capture_scopes_stream_gate_text_to_result_events() {
3340 let dir = tempfile::tempdir().unwrap();
3341 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
3342
3343 let echo_only = stream_capture_of(&[
3344 &v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3345 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3346 ]);
3347 std::fs::write(stdout_path(dir.path(), 30), &echo_only).unwrap();
3348 assert!(
3349 !checkpoint_reported_in_capture(dir.path(), 30),
3350 "an echoed gate mention read from the capture file must not report \
3351 a checkpoint"
3352 );
3353
3354 let declared = stream_capture_of(&[
3355 &v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3356 &v3_result_event(V3_RESULT_TURN1, &gate_declaration_text()),
3357 ]);
3358 std::fs::write(stdout_path(dir.path(), 31), &declared).unwrap();
3359 assert!(
3360 checkpoint_reported_in_capture(dir.path(), 31),
3361 "a genuine declaration read from the capture file must still \
3362 report a checkpoint"
3363 );
3364 }
3365
3366 /// **The fail-open regression.** A torn `system`/`init` line must not send
3367 /// gate scanning back to raw stdout.
3368 ///
3369 /// `claude_stream_events` silently drops any line that fails to parse, and
3370 /// recognition used to require a successfully parsed `init`. So one
3371 /// truncated first line — a partial write, or a read of a capture still
3372 /// being appended to — made the whole capture unrecognised, and
3373 /// `blocking_human_checkpoint_reported` fell back to scanning raw stdout,
3374 /// which under a stream capture contains the echoed prompt. The constraint-3
3375 /// scoping failed OPEN, into the exact false positive it exists to close.
3376 /// Found by cross-AI code review (gpt-5.6-sol, 2026-08-02, High finding 2).
3377 ///
3378 /// Envelopes are real (v3 `user` + `result`); the `init` line is a real one
3379 /// truncated mid-token, and the gate text payload is synthetic — no archived
3380 /// capture contains gate text or a prompt echo.
3381 #[test]
3382 fn blocking_human_checkpoint_reported_false_when_init_is_torn() {
3383 let torn_init = &V3_INIT_EVENT[..40];
3384 assert!(
3385 serde_json::from_str::<serde_json::Value>(torn_init).is_err(),
3386 "fixture precondition: the truncated init must actually fail to parse"
3387 );
3388
3389 let capture = format!(
3390 "{}\n{}\n{}\n",
3391 torn_init,
3392 v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3393 v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3394 );
3395 assert!(
3396 !blocking_human_checkpoint_reported(&capture),
3397 "a torn init must not re-enable the raw-stdout scan and let the \
3398 echoed prompt read as a gate declaration"
3399 );
3400
3401 // Same capture, init intact — proves the negative above is the torn-init
3402 // path being handled, not the fixture simply lacking gate text.
3403 let intact = stream_capture_of(&[
3404 &v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3405 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3406 ]);
3407 assert!(
3408 !blocking_human_checkpoint_reported(&intact),
3409 "control: the same capture with a valid init is also false"
3410 );
3411
3412 // And a real declaration is still detected with the init torn, so the
3413 // fix did not degenerate into always-false (T-30-24).
3414 let declared = format!(
3415 "{}\n{}\n{}\n",
3416 torn_init,
3417 v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3418 v3_result_event(V3_RESULT_TURN1, &gate_declaration_text()),
3419 );
3420 assert!(
3421 blocking_human_checkpoint_reported(&declared),
3422 "a genuine declaration must still be detected when init is torn"
3423 );
3424 }
3425
3426 /// A stream with NO `init` at all is likewise scoped rather than raw-scanned.
3427 /// Same fail-open class as the torn-init case; reported by the same review.
3428 #[test]
3429 fn blocking_human_checkpoint_reported_false_when_init_is_absent() {
3430 let capture = format!(
3431 "{}\n{}\n",
3432 v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3433 v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3434 );
3435 assert!(
3436 !blocking_human_checkpoint_reported(&capture),
3437 "an init-less stream must still scope the gate scan to result events"
3438 );
3439 }
3440
3441 /// **The mandatory over-correction controls.** Widening stream recognition
3442 /// must not divert the three non-stream inputs off the raw-scan path they
3443 /// have always used (T-30-25). Each carries genuine gate text and must
3444 /// still report `true`; if any flips to `false`, the widening has started
3445 /// suppressing real gates.
3446 #[test]
3447 fn non_stream_captures_still_use_the_raw_scan_after_widening() {
3448 let plain = format!("Some narration.\n{}\n", gate_declaration_text());
3449 assert!(
3450 blocking_human_checkpoint_reported(&plain),
3451 "plain text must still be raw-scanned"
3452 );
3453
3454 let single_doc = v3_result_event(V3_RESULT_TURN1, &gate_declaration_text());
3455 assert!(
3456 blocking_human_checkpoint_reported(&single_doc),
3457 "a single-document envelope must still be raw-scanned — it is \
3458 `{{\"type\":\"result\"}}`, which claude_stream_gate_shape excludes"
3459 );
3460
3461 let codex = format!(
3462 "{{\"type\":\"thread.started\",\"thread_id\":\"t1\"}}\n\
3463 {{\"type\":\"item.completed\",\"item\":{{\"type\":\"agent_message\",\
3464 \"text\":\"{}\"}}}}\n",
3465 gate_declaration_text().replace('"', "\\\"")
3466 );
3467 assert!(
3468 blocking_human_checkpoint_reported(&codex),
3469 "a Codex stream must still be raw-scanned — its top-level types are \
3470 dotted, so claude_stream_gate_shape excludes it"
3471 );
3472 }
3473
3474 /// **Fourth-pass High.** Decoding must never JOIN tokens across corrupt
3475 /// bytes. The third pass's remediation dropped invalid bytes, and
3476 /// `DEVFLOW_RESULT: {"status":"suc<FF>cess"}` with exit 1 decoded to a
3477 /// fabricated, VALID success marker — Layer 1 then short-circuited the
3478 /// nonzero exit. Replacement (U+FFFD) keeps the corruption visible: the
3479 /// status reads `suc\u{FFFD}cess`, no parser trusts it, and the exit code
3480 /// decides. Edge corruption stays covered by [`strip_corruption_padding`]
3481 /// — see the sibling third-pass test, which must pass alongside this one.
3482 #[test]
3483 fn corrupt_byte_inside_a_marker_is_never_repaired_into_success() {
3484 let dir = tempfile::tempdir().unwrap();
3485 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
3486
3487 let mut poisoned = b"DEVFLOW_RESULT: {\"status\":\"suc".to_vec();
3488 poisoned.push(0xff);
3489 poisoned.extend_from_slice(b"cess\"}");
3490 std::fs::write(stdout_path(dir.path(), 30), &poisoned).unwrap();
3491 assert_ne!(
3492 evaluate_layer1(dir.path(), 30).map(|r| r.status),
3493 Some(AgentStatus::Success),
3494 "a corrupt capture with no valid success marker must not be \
3495 repaired into an authoritative one"
3496 );
3497
3498 // Control: the same marker with the byte absent IS a real success.
3499 std::fs::write(
3500 stdout_path(dir.path(), 31),
3501 br#"DEVFLOW_RESULT: {"status":"success"}"#,
3502 )
3503 .unwrap();
3504 assert_eq!(
3505 evaluate_layer1(dir.path(), 31).map(|r| r.status),
3506 Some(AgentStatus::Success),
3507 "control: the intact marker must still parse as success"
3508 );
3509 }
3510
3511 /// **Third-pass High.** A stray invalid byte outside the JSON envelope must
3512 /// not convert an authoritative failure into a Layer-2 success.
3513 ///
3514 /// `from_utf8_lossy` substitutes U+FFFD, which survives `trim()`, so
3515 /// `detect_claude_envelope_failure`'s `starts_with('{')` guard went false and
3516 /// Layer 1 abstained on `is_error: true`. The cascade then fell through to
3517 /// the exit-code check — Ship proceeding on a reported failure. Reachable on
3518 /// the shipped `--output-format json` envelope; nothing to do with
3519 /// stream-json.
3520 #[test]
3521 fn stray_invalid_byte_does_not_hide_an_envelope_failure() {
3522 let envelope = br#"{"type":"result","subtype":"error","is_error":true,"result":"boom","session_id":"s"}"#;
3523 let dir = tempfile::tempdir().unwrap();
3524 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
3525
3526 std::fs::write(stdout_path(dir.path(), 30), envelope).unwrap();
3527 assert_eq!(
3528 evaluate_layer1(dir.path(), 30).map(|r| r.status),
3529 Some(AgentStatus::Failed),
3530 "control: the intact envelope is an authoritative Layer-1 failure"
3531 );
3532
3533 let mut poisoned = vec![0xffu8];
3534 poisoned.extend_from_slice(envelope);
3535 std::fs::write(stdout_path(dir.path(), 31), &poisoned).unwrap();
3536 assert_eq!(
3537 evaluate_layer1(dir.path(), 31).map(|r| r.status),
3538 Some(AgentStatus::Failed),
3539 "one invalid byte before the envelope must not make Layer 1 abstain \
3540 and hand a FAILURE to the exit-code fallback"
3541 );
3542 }
3543
3544 /// **Third-pass Medium.** A torn gate-bearing `user` event must not reopen
3545 /// raw-stdout scanning.
3546 ///
3547 /// `claude_stream_gate_shape` keyed stream recognition on system/user/
3548 /// assistant events. If the echoed `user` event tore *after* carrying the
3549 /// full gate text and only a later `result` parsed, none of those types
3550 /// survived, the capture stopped looking like a stream, and the raw scan
3551 /// read the echoed prompt as a declaration. Every line is still `{`-shaped,
3552 /// so this is neither the torn-`init` case nor V-01.
3553 #[test]
3554 fn torn_gate_bearing_user_event_does_not_reopen_raw_scanning() {
3555 let echo = v3_message_event(V3_USER_EVENT, &gate_documenting_text());
3556 let quiet_result = v3_result_event(V3_RESULT_TURN1, NO_MARKER);
3557
3558 let closed = format!("{}\n{}\n{}\n", V3_INIT_EVENT, echo, quiet_result);
3559 assert!(
3560 !blocking_human_checkpoint_reported(&closed),
3561 "control: with the echo intact the gate mention is correctly scoped out"
3562 );
3563
3564 let torn = format!("{}\n{}\n", &echo[..echo.len() - 12], quiet_result);
3565 assert!(
3566 !blocking_human_checkpoint_reported(&torn),
3567 "a torn echo leaving only a result must stay scoped, not fall back to \
3568 the raw scan that reads the echoed prompt as a declaration"
3569 );
3570
3571 // The shipped single-document envelope is ONE result line and must keep
3572 // taking the raw path (T-30-25).
3573 let single_doc = v3_result_event(V3_RESULT_TURN1, &gate_declaration_text());
3574 assert!(
3575 blocking_human_checkpoint_reported(&single_doc),
3576 "control: the single-document envelope still uses the raw scan"
3577 );
3578 }
3579
3580 /// **Fourth-pass Medium 3.** Benign prose noise must not block session
3581 /// recovery — only a torn JSON line can conceal a newer `init`.
3582 ///
3583 /// The first fail-closed guard rejected the capture when ANY non-empty line
3584 /// failed to parse, so one interleaved progress line disabled checkpoint
3585 /// auto-resume while the verdict parser accepted the same capture. An
3586 /// `init` is a JSON line; a non-`{` line can never be a torn one.
3587 #[test]
3588 fn prose_noise_does_not_block_session_recovery() {
3589 let stream = format!(
3590 "{}\nprogress: still working…\n{}\n",
3591 V3_INIT_EVENT,
3592 v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3593 );
3594 assert!(
3595 claude_stream_session_id(&stream).is_some(),
3596 "a prose progress line must not fail session recovery closed"
3597 );
3598
3599 // Control: the same capture with the noise line made JSON-shaped-but-torn
3600 // MUST fail closed — that shape could be a torn newer init.
3601 let torn = format!(
3602 "{}\n{{\"type\":\"system\",\"subty\n{}\n",
3603 V3_INIT_EVENT,
3604 v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3605 );
3606 assert!(
3607 claude_stream_session_id(&torn).is_none(),
3608 "a torn JSON line could be a newer init and must fail closed"
3609 );
3610 }
3611
3612 /// **Third-pass High.** A torn *later* `init` must not resurrect an earlier
3613 /// session's id.
3614 ///
3615 /// Each turn opens its own `init`; the last carries the id a resume must
3616 /// target. Dropped lines are invisible, so the scan returned the last
3617 /// PARSEABLE init — a stale token that looks entirely valid. Fails closed
3618 /// now: `None` costs a resume, the wrong id corrupts one.
3619 #[test]
3620 fn torn_later_init_does_not_resurrect_a_stale_session_id() {
3621 let init =
3622 |id: &str| format!(r#"{{"type":"system","subtype":"init","session_id":"{id}"}}"#);
3623
3624 let rotated = format!("{}\n{}\n", init("session-a"), init("session-b"));
3625 assert_eq!(
3626 claude_stream_session_id(&rotated).as_deref(),
3627 Some("session-b"),
3628 "control: with both init events intact the LAST id wins"
3629 );
3630
3631 let init_c = init("session-c");
3632 let torn = format!(
3633 "{}\n{}\n{}\n",
3634 init("session-a"),
3635 init("session-b"),
3636 &init_c[..init_c.len() - 10],
3637 );
3638 assert_ne!(
3639 claude_stream_session_id(&torn).as_deref(),
3640 Some("session-b"),
3641 "a torn newer init must not hand back the previous session's id"
3642 );
3643 }
3644
3645 /// **V-01 regression.** One stray JSONL-shaped line must not divert a
3646 /// plain-text capture onto the stream branch and suppress a real gate.
3647 ///
3648 /// The first `claude_stream_gate_shape` asked only whether ANY event carried
3649 /// a stream type. Since the stream branch never consults raw stdout, a single
3650 /// `{"type":"assistant",…}` line was enough to hide a genuine declaration
3651 /// sitting in the surrounding plain text — turning the fail-OPEN this
3652 /// predicate was written to close into a fail-CLOSED that drops a human
3653 /// authorization request. Found by phase-30 verification after the fix
3654 /// shipped in `06675da`.
3655 #[test]
3656 fn one_stray_json_line_does_not_suppress_a_plain_text_gate() {
3657 let gate = gate_declaration_text();
3658
3659 assert!(
3660 blocking_human_checkpoint_reported(&gate),
3661 "positive control: the gate text alone must be detected"
3662 );
3663
3664 let poisoned =
3665 format!("{gate}\n{{\"type\":\"assistant\",\"message\":{{\"content\":[]}}}}\n");
3666 assert!(
3667 blocking_human_checkpoint_reported(&poisoned),
3668 "one stray JSONL line must not suppress a real plain-text gate (V-01)"
3669 );
3670
3671 // The torn-init capture is still recognised as a stream — the majority
3672 // rule must not undo the fail-open fix it was added to preserve.
3673 let torn_init = &V3_INIT_EVENT[..40];
3674 let torn = format!(
3675 "{}\n{}\n{}\n",
3676 torn_init,
3677 v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3678 v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3679 );
3680 assert!(
3681 !blocking_human_checkpoint_reported(&torn),
3682 "control: a torn-init stream must still be scoped, not raw-scanned"
3683 );
3684 }
3685
3686 /// Every byte-prefix of a capture, fed to the gate scanner.
3687 ///
3688 /// **Why a sweep and not more hand-written cases.** Phase 30 shipped 116
3689 /// green tests, seven of them written specifically to prove the prompt-echo
3690 /// false positive was closed — and a cross-AI review then found that ONE
3691 /// torn line reverted the whole protection to the raw-stdout path. Every
3692 /// test fed the parser well-formed input; none fed it a broken one. Hand
3693 /// -picking more malformed cases would repeat that bias. Truncating at every
3694 /// offset removes the judgment call: the inputs are generated, not chosen.
3695 ///
3696 /// The invariant is one-directional — a prefix may lose detection (it has
3697 /// strictly less information), but it must never *gain* permissiveness.
3698 #[test]
3699 fn truncation_sweep_never_widens_gate_detection() {
3700 let intact = stream_capture_of(&[
3701 &v3_message_event(V3_USER_EVENT, &gate_documenting_text()),
3702 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3703 ]);
3704 assert!(
3705 !blocking_human_checkpoint_reported(&intact),
3706 "precondition: the intact capture must report no gate, or the sweep \
3707 below proves nothing"
3708 );
3709
3710 let mut checked = 0usize;
3711 for n in 0..=intact.len() {
3712 if !intact.is_char_boundary(n) {
3713 continue;
3714 }
3715 checked += 1;
3716 assert!(
3717 !blocking_human_checkpoint_reported(&intact[..n]),
3718 "truncating to {n} bytes made an echoed gate MENTION read as a \
3719 live declaration — the fail-open class (constraint 9)"
3720 );
3721 }
3722 assert!(
3723 checked > 500,
3724 "sweep degenerated to {checked} offsets; it is no longer exercising \
3725 the capture"
3726 );
3727 }
3728
3729 /// Same sweep against the session-id reader. Truncation may degrade it to
3730 /// `None` (a failed resume — fail-closed, acceptable); it must never yield a
3731 /// DIFFERENT id, which would resume the wrong session.
3732 #[test]
3733 fn truncation_sweep_never_forges_session_id() {
3734 let intact = stream_capture_of(&[
3735 &v3_message_event(V3_USER_EVENT, "session_id: forged-by-agent-text"),
3736 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3737 ]);
3738 let real = claude_stream_session_id(&intact);
3739 assert!(
3740 real.is_some(),
3741 "precondition: the intact capture yields an id"
3742 );
3743
3744 for n in 0..=intact.len() {
3745 if !intact.is_char_boundary(n) {
3746 continue;
3747 }
3748 let got = claude_stream_session_id(&intact[..n]);
3749 assert!(
3750 got.is_none() || got == real,
3751 "truncating to {n} bytes produced session id {got:?}, which is \
3752 neither None nor the CLI-emitted {real:?}"
3753 );
3754 }
3755 }
3756
3757 /// **Constraint 9 item 2, closed.** A subagent-origin `result` event must
3758 /// never decide the stage verdict — `last_top_level_result`'s name and doc
3759 /// always claimed top-level selection, but the first implementation
3760 /// selected on `type == "result"` alone (code-review M2). Envelope real
3761 /// (v3 result turn), planted `parent_tool_use_id` synthetic: no archived
3762 /// capture contains a subagent-origin result, so this pins deterministic
3763 /// behavior for an unobserved-but-legal shape.
3764 #[test]
3765 fn subagent_result_event_never_decides_the_verdict() {
3766 let subagent_success = v3_result_event(V3_RESULT_TURN2, MARKER_SUCCESS).replacen(
3767 "{",
3768 "{\"parent_tool_use_id\":\"toolu_child\",",
3769 1,
3770 );
3771 let capture = format!(
3772 "{}\n{}\n{}\n",
3773 V3_INIT_EVENT,
3774 v3_result_event_is_error(V3_RESULT_TURN1, MARKER_FAILED),
3775 subagent_success,
3776 );
3777 assert_eq!(
3778 parse_claude_event_result(&capture).map(|r| r.status),
3779 Some(AgentStatus::Failed),
3780 "a subagent-origin success result must not override the last \
3781 top-level failure"
3782 );
3783
3784 // Control: the same final event WITHOUT the planted parent id is
3785 // top-level and legitimately wins.
3786 let top_level = format!(
3787 "{}\n{}\n{}\n",
3788 V3_INIT_EVENT,
3789 v3_result_event_is_error(V3_RESULT_TURN1, MARKER_FAILED),
3790 v3_result_event(V3_RESULT_TURN2, MARKER_SUCCESS),
3791 );
3792 assert_eq!(
3793 parse_claude_event_result(&top_level).map(|r| r.status),
3794 Some(AgentStatus::Success),
3795 "control: the same event without a parent id is the final verdict"
3796 );
3797 }
3798
3799 /// D-13 trap 1, pinned: the delivery canary's declared token appears in the
3800 /// stream as a PROMPT ECHO before it can ever appear as an answer, so a
3801 /// naive text scan reports delivery on every run — including runs where the
3802 /// notification path is dead. That echo is what produced the checkpoint
3803 /// false positive 30-05 fixed.
3804 ///
3805 /// Three cases, and the first two are the negative controls that give the
3806 /// third its meaning: the same token, in the same capture shape, must read
3807 /// `false` from an echo and from a subagent-origin result, and `true` only
3808 /// from a top-level `result`.
3809 #[test]
3810 fn token_matches_only_inside_top_level_result() {
3811 const TOKEN: &str = "DEVFLOW-CANARY-7f3a";
3812
3813 // 1. Echo only: the token is in the operator's own turn, forwarded back
3814 // into stdout, and in no result at all.
3815 let echoed = format!(
3816 "{}\n{}\n{}\n",
3817 V3_INIT_EVENT,
3818 V3_USER_EVENT.replace("__MARKER__", &format!("please return {TOKEN} when done")),
3819 v3_result_event(V3_RESULT_TURN1, NO_MARKER),
3820 );
3821 assert!(
3822 !token_reported_in_capture(&echoed, TOKEN),
3823 "a token echoed back in the prompt is not delivery evidence — \
3824 the CLI forwards the operator's own turn into the same stdout"
3825 );
3826
3827 // 2. Subagent-origin result: right event type, wrong provenance.
3828 let subagent = format!(
3829 "{}\n{}\n",
3830 V3_INIT_EVENT,
3831 v3_result_event(V3_RESULT_TURN2, TOKEN).replacen(
3832 "{",
3833 "{\"parent_tool_use_id\":\"toolu_child\",",
3834 1,
3835 ),
3836 );
3837 assert!(
3838 !token_reported_in_capture(&subagent, TOKEN),
3839 "a subagent-origin result must not satisfy the canary — it is the \
3840 same provenance hole constraint 9 item 2 closed for the verdict"
3841 );
3842
3843 // 3. Authoritative: a top-level `result` carrying the token.
3844 let authoritative = format!(
3845 "{}\n{}\n",
3846 V3_INIT_EVENT,
3847 v3_result_event(V3_RESULT_TURN1, TOKEN),
3848 );
3849 assert!(
3850 token_reported_in_capture(&authoritative, TOKEN),
3851 "a token inside a top-level result IS the canary's answer"
3852 );
3853 }
3854
3855 /// The Codex arm of the trailing-torn rule — same R1 root cause, and the
3856 /// Codex adapter is live in production.
3857 ///
3858 /// The resurrection shape here is a torn SUPERSEDING marker: codex verdict
3859 /// precedence is marker-over-`turn.failed` by design (13-06 dogfood
3860 /// finding), and last-marker-wins — so the tear that matters is one that
3861 /// conceals a LATER marker contradicting an earlier success.
3862 #[test]
3863 fn codex_torn_tail_does_not_resurrect_earlier_success_marker() {
3864 let intact = concat!(
3865 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
3866 "{\"type\":\"item.completed\",\"item\":{\"type\":\"agent_message\",",
3867 "\"text\":\"DEVFLOW_RESULT: {\\\"status\\\":\\\"success\\\"}\"}}\n",
3868 "{\"type\":\"item.completed\",\"item\":{\"type\":\"agent_message\",",
3869 "\"text\":\"DEVFLOW_RESULT: {\\\"status\\\":\\\"failed\\\"}\"}}\n",
3870 );
3871 assert_eq!(
3872 parse_codex_event_result(intact).map(|r| r.status),
3873 Some(AgentStatus::Failed),
3874 "control: intact capture — the LAST marker wins and it is a failure"
3875 );
3876
3877 let torn = &intact[..intact.len() - 20];
3878 assert_ne!(
3879 parse_codex_event_result(torn).map(|r| r.status),
3880 Some(AgentStatus::Success),
3881 "a torn superseding marker must not let the earlier success marker \
3882 decide the stage"
3883 );
3884 }
3885
3886 /// **Sixth-pass Highs 1–3.** The marker tail scanner — the reader that
3887 /// decides most production stages today — must survive edge corruption, a
3888 /// marker line longer than the tail budget, and mixed-case prefixes (its
3889 /// contract has always said case-insensitive).
3890 #[test]
3891 fn marker_tail_scan_survives_corruption_length_and_case() {
3892 let m = "DEVFLOW_RESULT: {\"status\":\"failed\"}";
3893 assert_eq!(
3894 parse_devflow_result(m).map(|r| r.status),
3895 Some(AgentStatus::Failed),
3896 "control: the plain marker parses"
3897 );
3898
3899 // High 1 — edge corruption on either side must not hide the marker.
3900 for poisoned in [format!("\u{FFFD}{m}"), format!("{m}\u{FFFD}")] {
3901 assert_eq!(
3902 parse_devflow_result(&poisoned).map(|r| r.status),
3903 Some(AgentStatus::Failed),
3904 "one stray byte at a line edge must not hide a failure marker"
3905 );
3906 }
3907 // …while interior corruption stays untrusted (fourth-pass hazard).
3908 assert!(
3909 parse_devflow_result("DEVFLOW_RESULT: {\"status\":\"fai\u{FFFD}led\"}").is_none(),
3910 "interior corruption must not parse as a valid status"
3911 );
3912
3913 // High 2 — a marker line longer than the tail budget is scanned whole.
3914 let long_reason = "x".repeat(5000);
3915 let long =
3916 format!("DEVFLOW_RESULT: {{\"status\":\"failed\",\"reason\":\"{long_reason}\"}}");
3917 assert_eq!(
3918 parse_devflow_result(&long).map(|r| r.status),
3919 Some(AgentStatus::Failed),
3920 "the tail budget must never bisect the final marker line"
3921 );
3922 // …and the budget still bounds the walk: a marker buried beyond the
3923 // budget with newer non-marker output after it stays out of reach.
3924 let buried = format!("{m}\n{}\n", "y\n".repeat(4100));
3925 assert!(
3926 parse_devflow_result(&buried).is_none(),
3927 "control: the budget still cuts off markers deep in old output"
3928 );
3929
3930 // High 3 — mixed case matches, per the documented contract.
3931 assert_eq!(
3932 parse_devflow_result("DevFlow_Result: {\"status\":\"failed\"}").map(|r| r.status),
3933 Some(AgentStatus::Failed),
3934 "mixed-case prefix must match — the contract says case-insensitive"
3935 );
3936 }
3937
3938 /// **Sixth-pass Mediums 4–5.** The codex plain-text rate-limit heuristic:
3939 /// an edge-corrupt JSON event line must stay excluded from prose scanning,
3940 /// and "429" only counts as a standalone token.
3941 #[test]
3942 fn codex_rate_limit_heuristic_excludes_recovered_json_and_embedded_429() {
3943 // M4 — a corrupt-prefixed event line is still a JSON line, not prose.
3944 let doc_line = concat!(
3945 "{\"type\":\"item.completed\",\"item\":{\"type\":\"agent_message\",",
3946 "\"text\":\"docs mention rate limiting policies\"}}",
3947 );
3948 let poisoned =
3949 format!("{{\"type\":\"thread.started\",\"thread_id\":\"t\"}}\n\u{FFFD}{doc_line}\n");
3950 assert!(
3951 detect_codex_rate_limit(&poisoned).is_none(),
3952 "an edge-corrupt event line must not be prose-scanned for \
3953 rate-limit vocabulary"
3954 );
3955 // Control: genuine plain-text rate-limit output is still detected.
3956 assert!(
3957 detect_codex_rate_limit("Rate limit exceeded. Try again at 17:00.").is_some(),
3958 "control: real plain-text rate-limit output must still be detected"
3959 );
3960
3961 // M5 — embedded digits are not rate-limit evidence…
3962 assert!(
3963 detect_codex_rate_limit("processed issue #429 successfully").is_none(),
3964 "'#429' is an issue number, not a rate limit"
3965 );
3966 assert!(
3967 detect_codex_rate_limit("transferred 14290 bytes").is_none(),
3968 "digits containing 429 are not a rate limit"
3969 );
3970 // …while a genuine standalone 429 still is.
3971 assert!(
3972 detect_codex_rate_limit("HTTP 429 Too Many Requests").is_some(),
3973 "control: a standalone 429 status is still detected"
3974 );
3975 }
3976
3977 /// **Fifth-pass High 1.** A replacement-character-prefixed event line must
3978 /// not classify as prose Noise and slip past the torn-tail guard.
3979 ///
3980 /// `read_capture` turns an invalid byte into U+FFFD; a line reading
3981 /// `\u{FFFD}{"type":…}` fails to parse and does not start with `{`, so it
3982 /// became Noise — invisible to `torn_json_after_last_matching`. A corrupt
3983 /// byte in front of a superseding failed marker let the earlier success
3984 /// marker decide the stage, with the contradicting exit code never
3985 /// consulted. Live today on the Codex `--json` adapter. The fix recovers
3986 /// an edge-corrupt-but-intact event by re-parsing the stripped line, so
3987 /// the TRUE verdict decides — better than merely failing indeterminate.
3988 #[test]
3989 fn corruption_prefixed_event_line_is_not_prose_noise() {
3990 let good = concat!(
3991 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
3992 "{\"type\":\"item.completed\",\"item\":{\"type\":\"agent_message\",",
3993 "\"text\":\"DEVFLOW_RESULT: {\\\"status\\\":\\\"success\\\"}\"}}\n",
3994 );
3995 let failed_line = concat!(
3996 "{\"type\":\"item.completed\",\"item\":{\"type\":\"agent_message\",",
3997 "\"text\":\"DEVFLOW_RESULT: {\\\"status\\\":\\\"failed\\\"}\"}}\n",
3998 );
3999
4000 let intact = format!("{good}{failed_line}");
4001 assert_eq!(
4002 parse_codex_event_result(&intact).map(|r| r.status),
4003 Some(AgentStatus::Failed),
4004 "control: intact capture — the last (failed) marker decides"
4005 );
4006
4007 let poisoned = format!("{good}\u{FFFD}{failed_line}");
4008 assert_eq!(
4009 parse_codex_event_result(&poisoned).map(|r| r.status),
4010 Some(AgentStatus::Failed),
4011 "an edge-corrupt superseding marker must be recovered (or at worst \
4012 fail indeterminate) — never let the earlier success decide"
4013 );
4014
4015 // Interior corruption stays visible and untrusted: a FFFD INSIDE the
4016 // marker's status string must not parse as a valid status (the
4017 // fourth-pass fabrication hazard, still guarded).
4018 let interior = concat!(
4019 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
4020 "{\"type\":\"item.completed\",\"item\":{\"type\":\"agent_message\",",
4021 "\"text\":\"DEVFLOW_RESULT: {\\\"status\\\":\\\"suc\u{FFFD}cess\\\"}\"}}\n",
4022 );
4023 assert_ne!(
4024 parse_codex_event_result(interior).map(|r| r.status),
4025 Some(AgentStatus::Success),
4026 "interior corruption must never be repaired into a success"
4027 );
4028 }
4029
4030 /// **Fifth-pass Medium 1.** An edge-corrupt rate-limit envelope must stay
4031 /// `RateLimited`, not decay into a generic `Failed`.
4032 ///
4033 /// The rate-limit detector outranks the generic envelope-failure detector
4034 /// precisely because rate-limit envelopes carry `is_error: true`. It was
4035 /// the one single-document reader without `strip_corruption_padding`, so a
4036 /// stray byte inverted the precedence — auto-resume became review/gating.
4037 #[test]
4038 fn edge_corrupt_rate_limit_envelope_stays_rate_limited() {
4039 let envelope = r#"{"type":"result","subtype":"error_rate_limit","is_error":true,"result":"rate limited","retry_after":"17:00"}"#;
4040 assert!(
4041 detect_claude_rate_limit(envelope).is_some(),
4042 "control: the intact envelope is detected as a rate limit"
4043 );
4044 assert!(
4045 detect_claude_rate_limit(&format!("\u{FFFD}{envelope}")).is_some(),
4046 "one stray byte must not demote RateLimited to generic Failed"
4047 );
4048 }
4049
4050 /// **Fourth-pass Medium 1.** The generic marker path — the one production
4051 /// hits today — must overwrite a planted `decided_by_layer`, exactly as the
4052 /// stream path has since 30-01. `Some(0)` is Layer-0 external-verification
4053 /// provenance, which `classify_validate_outcome` trusts when classifying a
4054 /// Validate stage: an agent could force the ambiguous-validation gate with
4055 /// a verdict it never earned (999.67's class, live instance).
4056 #[test]
4057 fn generic_marker_cannot_forge_layer0_provenance() {
4058 let stdout = r#"DEVFLOW_RESULT: {"status":"success","decided_by_layer":0}"#;
4059 let result = parse_devflow_result(stdout).unwrap();
4060 assert_eq!(
4061 result.decided_by_layer,
4062 Some(1),
4063 "a planted decided_by_layer:0 must be overwritten to Layer 1"
4064 );
4065
4066 // Control: an honest marker without the field also normalises to
4067 // Some(1) — provenance is DERIVED here, never deserialized.
4068 let honest = r#"DEVFLOW_RESULT: {"status":"success"}"#;
4069 assert_eq!(
4070 parse_devflow_result(honest).unwrap().decided_by_layer,
4071 Some(1)
4072 );
4073 }
4074
4075 /// Codex arm of the T-30-26 provenance overwrite (fourth-pass Medium 1's
4076 /// class): a `decided_by_layer` planted in the codex marker JSON must be
4077 /// overwritten, exactly as on the generic and Claude-stream marker paths.
4078 #[test]
4079 fn codex_marker_cannot_forge_layer0_provenance() {
4080 let capture = concat!(
4081 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
4082 "{\"type\":\"item.completed\",\"item\":{\"type\":\"agent_message\",",
4083 "\"text\":\"DEVFLOW_RESULT: {\\\"status\\\":\\\"success\\\",\\\"decided_by_layer\\\":0}\"}}\n",
4084 );
4085 let result = parse_codex_event_result(capture).unwrap();
4086 assert_eq!(
4087 result.decided_by_layer,
4088 Some(1),
4089 "a planted decided_by_layer:0 must be overwritten to Layer 1"
4090 );
4091 }
4092
4093 /// **Constraint 9 item 1, closed for every DETECTABLE truncation**
4094 /// (originally committed `#[ignore]`d as a known-red deferral to Phase 31;
4095 /// the operator's "fix root causes before proceeding" decision pulled it
4096 /// back into phase 30).
4097 ///
4098 /// A truncated terminal `result` used to vanish from the parsed events, so
4099 /// `last_top_level_result` returned an EARLIER turn's result — a stale
4100 /// SUCCESS advancing a stage whose real terminal turn failed. Now every
4101 /// prefix with a torn trailing line yields an indeterminate FAILURE.
4102 ///
4103 /// **The named residual — line-boundary truncation is UNDETECTABLE from
4104 /// content.** A prefix cut exactly at the newline after the success turn is
4105 /// a well-formed capture: two parsed events, no torn line, byte-identical
4106 /// to a healthy one-turn-success capture plus nothing. The evidence of loss
4107 /// is in the bytes that never arrived, so no parser assertion can exist for
4108 /// it. The remaining defense belongs to the layer that HAS the missing
4109 /// information: Phase 31's wiring must not let a stream-derived Success
4110 /// short-circuit a contradicting exit code (a writer that died between
4111 /// flushing turn N and turn N+1 also died with a non-zero exit). Recorded
4112 /// in ROADMAP constraint 9.
4113 #[test]
4114 fn truncation_sweep_never_upgrades_verdict_to_success() {
4115 let intact = format!(
4116 "{}\n{}\n{}\n",
4117 V3_INIT_EVENT,
4118 v3_result_event(V3_RESULT_TURN1, MARKER_SUCCESS),
4119 v3_result_event_is_error(V3_RESULT_TURN2, MARKER_FAILED),
4120 );
4121 assert_eq!(
4122 parse_claude_event_result(&intact).map(|r| r.status),
4123 Some(AgentStatus::Failed),
4124 "precondition: intact capture ends in a failure verdict"
4125 );
4126
4127 let mut torn_prefixes = 0usize;
4128 let mut clean_prefixes = 0usize;
4129 for n in 0..=intact.len() {
4130 if !intact.is_char_boundary(n) {
4131 continue;
4132 }
4133 let prefix = &intact[..n];
4134 let got = parse_claude_event_result(prefix).map(|r| r.status);
4135 if ParsedCapture::parse(prefix).torn_json_line_present() {
4136 torn_prefixes += 1;
4137 assert_ne!(
4138 got,
4139 Some(AgentStatus::Success),
4140 "truncating to {n} bytes left a torn tail yet resurrected \
4141 an earlier turn's SUCCESS over a failed terminal turn"
4142 );
4143 } else {
4144 clean_prefixes += 1;
4145 }
4146 }
4147 // Negative controls on the sweep itself: both branches must have been
4148 // exercised, or the loop is asserting over nothing.
4149 assert!(
4150 torn_prefixes > 500,
4151 "sweep degenerated: only {torn_prefixes} torn prefixes"
4152 );
4153 assert!(
4154 clean_prefixes > 2,
4155 "sweep never produced a well-formed prefix; the residual case \
4156 documented above is not being exercised"
4157 );
4158 }
4159
4160 #[test]
4161 fn codex_event_stream_parses_turn_failed() {
4162 let stdout = concat!(
4163 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
4164 "{\"type\":\"turn.started\"}\n",
4165 "{\"type\":\"item.started\",\"item\":{}}\n",
4166 "{\"type\":\"turn.failed\",\"error\":{\"message\":\"sandbox denied write\"}}\n",
4167 );
4168 let result = parse_codex_event_result(stdout).unwrap();
4169 assert_eq!(result.status, AgentStatus::Failed);
4170 assert_eq!(result.reason.as_deref(), Some("sandbox denied write"));
4171 }
4172
4173 #[test]
4174 fn codex_turn_completed_no_marker_defers() {
4175 let stdout = concat!(
4176 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
4177 "{\"type\":\"turn.started\"}\n",
4178 "{\"type\":\"turn.completed\",\"usage\":{\"input_tokens\":10,\"output_tokens\":5}}\n",
4179 );
4180 assert!(parse_codex_event_result(stdout).is_none());
4181 }
4182
4183 /// 13-06 dogfood regression: Codex delivers the DEVFLOW_RESULT marker
4184 /// inside an `agent_message` item's text, never as a raw stdout line. A
4185 /// self-reported failure followed by a bare `turn.completed` must parse
4186 /// as Failed with the agent's reason — not defer to Layer 2 (which would
4187 /// see exit 0 and call it a success).
4188 #[test]
4189 fn codex_agent_message_marker_failed_wins_over_bare_turn_completed() {
4190 let stdout = concat!(
4191 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
4192 "{\"type\":\"item.completed\",\"item\":{\"id\":\"item_7\",\"type\":\"agent_message\",\"text\":\"DEVFLOW_RESULT: {\\\"status\\\": \\\"failed\\\", \\\"reason\\\": \\\"interactive input unavailable\\\"}\"}}\n",
4193 "{\"type\":\"turn.completed\",\"usage\":{\"input_tokens\":10,\"output_tokens\":5}}\n",
4194 );
4195 let result = parse_codex_event_result(stdout).unwrap();
4196 assert_eq!(result.status, AgentStatus::Failed);
4197 assert_eq!(
4198 result.reason.as_deref(),
4199 Some("interactive input unavailable")
4200 );
4201 }
4202
4203 #[test]
4204 fn codex_agent_message_marker_success_short_circuits() {
4205 let stdout = concat!(
4206 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
4207 "{\"type\":\"item.completed\",\"item\":{\"id\":\"item_2\",\"type\":\"agent_message\",\"text\":\"DEVFLOW_RESULT: {\\\"status\\\": \\\"success\\\"}\"}}\n",
4208 "{\"type\":\"turn.completed\",\"usage\":{\"input_tokens\":10,\"output_tokens\":5}}\n",
4209 );
4210 let result = parse_codex_event_result(stdout).unwrap();
4211 assert_eq!(result.status, AgentStatus::Success);
4212 }
4213
4214 /// 13-06 dogfood regression: document content echoed into a JSONL event
4215 /// (GSD reference tables mentioning "rate limiting") must not trip the
4216 /// plain-text rate-limit heuristic — it returned the entire multi-KB
4217 /// event line as the "retry time" and that reached the desktop
4218 /// notification verbatim.
4219 #[test]
4220 fn detect_rate_limit_ignores_json_event_lines() {
4221 let stdout = concat!(
4222 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
4223 "{\"type\":\"item.completed\",\"item\":{\"id\":\"item_4\",\"type\":\"command_execution\",\"aggregated_output\":\"| API keys | Rate limiting per key? |\"}}\n",
4224 "{\"type\":\"turn.completed\",\"usage\":{\"input_tokens\":10,\"output_tokens\":5}}\n",
4225 );
4226 assert_eq!(detect_rate_limit(stdout), None);
4227 }
4228
4229 #[test]
4230 fn detect_rate_limit_still_reads_codex_plain_text() {
4231 let stdout = "Rate limit reached.\nTry again at 3:45 PM.\n";
4232 assert_eq!(detect_rate_limit(stdout).as_deref(), Some("3:45 PM"));
4233 }
4234
4235 #[test]
4236 fn codex_event_stream_ignores_progress_and_unparseable_lines() {
4237 let stdout = concat!(
4238 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
4239 "not json at all\n",
4240 "{\"type\":\"item.started\",\"item\":{}}\n",
4241 "{\"type\":\"item.updated\",\"item\":{}}\n",
4242 "{\"type\":\"turn.failed\",\"error\":{\"message\":\"boom\"}}\n",
4243 );
4244 let result = parse_codex_event_result(stdout).unwrap();
4245 assert_eq!(result.status, AgentStatus::Failed);
4246 assert_eq!(result.reason.as_deref(), Some("boom"));
4247 }
4248
4249 #[test]
4250 fn claude_envelope_not_consumed_by_codex_parser() {
4251 let stdout = r#"{"type":"result","subtype":"success","is_error":false,"num_turns":4,"result":"All done.","session_id":"abc"}"#;
4252 assert!(parse_codex_event_result(stdout).is_none());
4253 }
4254
4255 /// The highest-value isolation test in plan 30-01 (T-30-02).
4256 ///
4257 /// The single-document `--output-format json` envelope that ships TODAY
4258 /// carries `type: "result"` AND a `session_id` — precisely the gate shape
4259 /// 30-RESEARCH.md offered as an alternative to `system`/`init`. If anyone
4260 /// widens [`is_claude_event_stream`] to accept it, the stream parser starts
4261 /// consuming every production capture in use and silently displaces
4262 /// `parse_devflow_result` in the Layer-1 cascade. This test fails first.
4263 ///
4264 /// The first literal is reused verbatim from
4265 /// `claude_envelope_not_consumed_by_codex_parser` above so the two read as
4266 /// a matched pair.
4267 #[test]
4268 fn single_doc_envelope_not_consumed_by_claude_stream_parser() {
4269 let stdout = r#"{"type":"result","subtype":"success","is_error":false,"num_turns":4,"result":"All done.","session_id":"abc"}"#;
4270 assert!(parse_claude_event_result(stdout).is_none());
4271
4272 // Non-vacuity: the literal above carries no marker, so it would return
4273 // None even from a WRONGLY-widened gate — on its own it proves little.
4274 // This envelope does carry one, so it can only return None because the
4275 // gate declined the document, not because the marker scan came up dry.
4276 let with_marker = r#"{"type":"result","subtype":"success","is_error":false,"num_turns":4,"result":"Done.\nDEVFLOW_RESULT: {\"status\":\"success\"}","session_id":"abc"}"#;
4277 assert!(parse_claude_event_result(with_marker).is_none());
4278
4279 // ...and the shipped path still owns it, so declining costs no verdict.
4280 assert_eq!(
4281 parse_devflow_result(with_marker).unwrap().status,
4282 AgentStatus::Success
4283 );
4284 }
4285
4286 /// Cross-adapter isolation: a Codex `--json` event stream is not consumed
4287 /// by the Claude stream parser. The two gates are mutually exclusive by
4288 /// construction — Codex keys on `thread.started`/`turn.*`, Claude on
4289 /// `system`/`init` — and this pins that.
4290 #[test]
4291 fn codex_stream_not_consumed_by_claude_stream_parser() {
4292 let stdout = concat!(
4293 "{\"type\":\"thread.started\",\"thread_id\":\"t1\"}\n",
4294 "{\"type\":\"item.completed\",\"item\":{\"id\":\"item_2\",\"type\":\"agent_message\",\"text\":\"DEVFLOW_RESULT: {\\\"status\\\": \\\"success\\\"}\"}}\n",
4295 "{\"type\":\"turn.completed\",\"usage\":{\"input_tokens\":10,\"output_tokens\":5}}\n",
4296 );
4297 assert!(parse_claude_event_result(stdout).is_none());
4298
4299 // The Codex parser still decides it — isolation costs no verdict.
4300 assert_eq!(
4301 parse_codex_event_result(stdout).unwrap().status,
4302 AgentStatus::Success
4303 );
4304 }
4305
4306 /// The same isolation claim in the other direction: a Claude stream capture
4307 /// is not consumed by the Codex parser, so the two never collide.
4308 #[test]
4309 fn claude_stream_not_consumed_by_codex_parser() {
4310 let capture = v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS);
4311 assert!(parse_codex_event_result(&capture).is_none());
4312 }
4313
4314 /// Plain-text stdout is not consumed by the Claude stream parser.
4315 ///
4316 /// Non-vacuous by construction: the text carries a real marker, so a gate
4317 /// that wrongly fired on non-JSON input would change the verdict rather
4318 /// than merely returning None. The second assertion pins that the marker
4319 /// path still decides it — the cascade must lose nothing.
4320 #[test]
4321 fn plain_text_not_consumed_by_claude_stream_parser() {
4322 let stdout = "Running the plan...\nDEVFLOW_RESULT: {\"status\":\"success\"}\n";
4323 assert!(parse_claude_event_result(stdout).is_none());
4324 assert_eq!(
4325 parse_devflow_result(stdout).unwrap().status,
4326 AgentStatus::Success
4327 );
4328 }
4329
4330 // ---- Claude `--output-format stream-json` fixtures (plan 30-01) --------
4331 //
4332 // Sourced from the archived capture
4333 // `.planning/phases/30-keep-the-session-alive-past-turn-end/30a-evidence/raw_output_v3.jsonl`,
4334 // a real 54-line stream from a session that survived three orchestrator
4335 // turns via task-notification wake-ups. The `init` event is line 5; the
4336 // three `result` events are lines 19, 37 and 54.
4337 //
4338 // TWO documented modifications, both labelled where they occur:
4339 // 1. Each envelope's `result` string value is replaced with the sentinel
4340 // `__MARKER__`, which each test fills in. NO archived capture contains
4341 // a real `DEVFLOW_RESULT` marker — the v3 harness produced
4342 // acknowledgment prose, not GSD stage output — so every marker payload
4343 // below is SYNTHETIC. Envelope shape is real; marker text is not.
4344 // 2. The `init` event's three inert array payloads are truncated and its
4345 // `cwd` is redacted (see `V3_INIT_EVENT`).
4346 // Everything else is byte-for-byte as captured, including field ORDER —
4347 // note that `"type":"result"` appears near the END of each result line,
4348 // long after `result` itself, which is exactly why the parser must key on
4349 // the parsed object rather than on textual position.
4350
4351 /// v3 line 5 — the `system`/`init` event that opens the stream and is the
4352 /// ONLY thing `is_claude_event_stream` gates on.
4353 ///
4354 /// Modification 2: verbatim except that `tools`, `mcp_servers` and
4355 /// `slash_commands` are truncated to a real prefix (verbatim they run to
4356 /// 5,523 characters of tool and slash-command names that no code path here
4357 /// reads) and `cwd` is redacted to a neutral path — the captured value
4358 /// embeds a developer's home directory, and `devflow-core` is published to
4359 /// crates.io. Both fields are inert for every function under test.
4360 const V3_INIT_EVENT: &str = r#"{"type":"system","subtype":"init","cwd":"/tmp/scratchpad/999.64-experiment","session_id":"559fef4d-2053-459e-b7a7-f3200c3b3790","tools":["Task","Bash","Read","Write"],"mcp_servers":[{"name":"github","status":"pending"}],"model":"claude-opus-5[1m]","permissionMode":"bypassPermissions","slash_commands":["gsd-execute-phase"],"capabilities":["interrupt_receipt_v1","interrupt_cancel_queued_v1","msg_lifecycle_v1"],"uuid":"597e1613-77cb-4cdd-a716-2aa75dc58c0b"}"#;
4361
4362 /// v3 line 19 — the FIRST turn's terminal `result` event.
4363 const V3_RESULT_TURN1: &str = r#"{"is_error":false,"duration_api_ms":8087,"num_turns":3,"stop_reason":"end_turn","session_id":"559fef4d-2053-459e-b7a7-f3200c3b3790","total_cost_usd":0.2401795,"usage":{"input_tokens":4,"cache_creation_input_tokens":20120,"cache_read_input_tokens":49219,"output_tokens":574,"service_tier":"standard","inference_geo":"not_available","speed":"standard"},"permission_denials":[],"terminal_reason":"completed","fast_mode_state":"off","subtype":"success","api_error_status":null,"result":"__MARKER__","ttft_ms":1381,"time_to_request_ms":91,"type":"result","duration_ms":8315,"uuid":"3dce3044-2d33-4c4d-bfcb-80e1756a5522"}"#;
4364
4365 /// v3 line 37 — the SECOND turn's terminal `result` event, produced after a
4366 /// task-notification wake-up. Carries the `origin` key the later turns have
4367 /// and the first does not.
4368 const V3_RESULT_TURN2: &str = r#"{"is_error":false,"duration_api_ms":27809,"num_turns":1,"stop_reason":"end_turn","session_id":"559fef4d-2053-459e-b7a7-f3200c3b3790","total_cost_usd":0.53654625,"usage":{"input_tokens":2,"cache_creation_input_tokens":3147,"cache_read_input_tokens":35393,"output_tokens":124,"service_tier":"standard","inference_geo":"not_available","speed":"standard"},"permission_denials":[],"terminal_reason":"completed","fast_mode_state":"off","origin":{"kind":"task-notification"},"subtype":"success","api_error_status":null,"result":"__MARKER__","ttft_ms":5476,"time_to_request_ms":18,"type":"result","duration_ms":6195,"uuid":"ca58693c-2599-4eb6-955b-e9d1e7444255"}"#;
4369
4370 /// v3 line 54 — the THIRD and LAST turn's terminal `result` event. This is
4371 /// the one whose marker must decide the stage.
4372 const V3_RESULT_TURN3: &str = r#"{"is_error":false,"duration_api_ms":39273,"num_turns":2,"stop_reason":"end_turn","session_id":"559fef4d-2053-459e-b7a7-f3200c3b3790","total_cost_usd":0.6599295,"usage":{"input_tokens":4,"cache_creation_input_tokens":999,"cache_read_input_tokens":77871,"output_tokens":302,"service_tier":"standard","inference_geo":"not_available","speed":"standard"},"permission_denials":[],"terminal_reason":"completed","fast_mode_state":"off","origin":{"kind":"task-notification"},"subtype":"success","api_error_status":null,"result":"__MARKER__","ttft_ms":2099,"time_to_request_ms":14,"type":"result","duration_ms":5276,"uuid":"dc76186e-3e9a-4d52-9152-27aa5012bc41"}"#;
4373
4374 // ---- prompt-echo regression fixtures (plan 30-05) ----------------------
4375 //
4376 // Message-event envelopes from the same archived capture. Same sentinel
4377 // discipline as the `result` envelopes above — the innermost text payload
4378 // is replaced with `__MARKER__` and each test fills it — plus a third
4379 // documented modification noted per constant where inert bulk is dropped.
4380 // The ENVELOPE is real: every `type`, `parent_tool_use_id`, `session_id`
4381 // and `uuid` value, and the nesting shape the extraction path walks, is
4382 // exactly as captured.
4383 //
4384 // NO archived capture contains checkpoint gate text at all — the 30a
4385 // harness prompt was about background tasks and never mentioned gates. So
4386 // every gate payload below is SYNTHETIC and must not be described as an
4387 // observed rendering. What IS observed is the gate VALUE's markdown
4388 // code-span rendering, transcribed from the live 2026-07-31 A1 run (see
4389 // `HUMAN_GATE_VALUE`), which every fixture here reproduces.
4390
4391 /// v3 line 10 — a TOP-LEVEL `user` event (`parent_tool_use_id` null).
4392 ///
4393 /// Modification 3: the trailing `tool_use_result` object is dropped. It is
4394 /// inert for every function under test and embeds both a developer home
4395 /// directory and the child agent's full prompt; `devflow-core` is published
4396 /// to crates.io.
4397 ///
4398 /// **The archived capture contains no echoed prompt.** Every `user` event
4399 /// in it is a `tool_result` relay, because the 30a harness ran a single
4400 /// prompt with no re-injection. This fixture's payload therefore STANDS IN
4401 /// for an echoed prompt rather than reproducing one. The substitution is
4402 /// sound for what is under test: the scan's first filter keys on the
4403 /// event's `type`, which is `user` in both cases, and
4404 /// `claude_stream_reports_human_gate` excludes that whole class — an echoed
4405 /// prompt and a re-injected notification summary are the two members of it.
4406 const V3_USER_EVENT: &str = r#"{"type":"user","message":{"role":"user","content":[{"tool_use_id":"toolu_01FVk15W8zxiazXutJYn8rsv","type":"tool_result","content":[{"type":"text","text":"__MARKER__"}]}]},"parent_tool_use_id":null,"session_id":"559fef4d-2053-459e-b7a7-f3200c3b3790","uuid":"60c5839e-40b3-492a-83e7-00882189f1d3","timestamp":"2026-08-02T00:22:22.603Z"}"#;
4407
4408 /// v3 line 6 — a TOP-LEVEL `assistant` event (`parent_tool_use_id` null).
4409 ///
4410 /// Its captured payload is `I'll spawn both subagents in the background
4411 /// now.` — mid-turn narration that appears in NO `result` event of the
4412 /// capture, re-confirmed by re-parsing all 54 lines at execution time. That
4413 /// property is the entire reason this envelope was chosen: it proves
4414 /// top-level assistant text is not merely a preview of the result text, so
4415 /// admitting the class would add a genuinely new trusted surface.
4416 ///
4417 /// Modification 3: the `usage.cache_creation` sub-object is dropped (inert).
4418 const V3_ASSISTANT_TOP_LEVEL_EVENT: &str = r#"{"type":"assistant","message":{"model":"claude-opus-5","id":"msg_011Cdcy3oC1a4rcmbp3avDYX","type":"message","role":"assistant","content":[{"type":"text","text":"__MARKER__"}],"stop_reason":null,"stop_sequence":null,"stop_details":null,"usage":{"input_tokens":2,"cache_creation_input_tokens":18673,"cache_read_input_tokens":15273,"output_tokens":1,"service_tier":"standard","inference_geo":"not_available"},"diagnostics":null,"context_management":null},"parent_tool_use_id":null,"session_id":"559fef4d-2053-459e-b7a7-f3200c3b3790","uuid":"85e8747b-e551-47b7-af38-fcd3bb1e06f8","timestamp":"2026-08-02T00:22:18.742Z","request_id":"req_011Cdcy3ngzpMCk3bijt1nkE"}"#;
4419
4420 /// v3 line 11 — a SUBAGENT-forwarded `assistant` event. Its captured
4421 /// `parent_tool_use_id` (`toolu_01FVk15W8zxiazXutJYn8rsv`, the Task call
4422 /// that spawned child A) is preserved verbatim: it is the whole point of
4423 /// the fixture, and the discrimination whose absence invalidated the v1
4424 /// experiment outright.
4425 ///
4426 /// Modification 3: the `usage.cache_creation` sub-object is dropped (inert).
4427 const V3_ASSISTANT_SUBAGENT_EVENT: &str = r#"{"type":"assistant","message":{"model":"claude-opus-5","id":"msg_011Cdcy4BNkfziogNMFM8V7K","type":"message","role":"assistant","content":[{"type":"text","text":"__MARKER__"}],"stop_reason":null,"stop_sequence":null,"stop_details":null,"usage":{"input_tokens":2,"cache_creation_input_tokens":17705,"cache_read_input_tokens":0,"output_tokens":1,"service_tier":"standard","inference_geo":"not_available"},"diagnostics":null,"context_management":null},"parent_tool_use_id":"toolu_01FVk15W8zxiazXutJYn8rsv","session_id":"559fef4d-2053-459e-b7a7-f3200c3b3790","uuid":"3fb37d43-86af-48b1-ace4-55147ed47b15","timestamp":"2026-08-02T00:22:23.850Z","request_id":"req_011Cdcy4ASSvG8gf8fRwiWZW","subagent_type":"general-purpose","task_description":"Signal A after 10s"}"#;
4428
4429 /// Fill a message envelope's innermost text payload. Mirrors
4430 /// [`v3_result_event`] and is kept separate from it so the assertion names
4431 /// the right fixture family when a sentinel is lost.
4432 fn v3_message_event(envelope: &str, text: &str) -> String {
4433 assert!(
4434 envelope.contains("__MARKER__"),
4435 "fixture envelope lost its message-text sentinel"
4436 );
4437 envelope.replace("__MARKER__", text)
4438 }
4439
4440 /// A checkpoint DECLARATION, as an agent's final message would render it,
4441 /// escaped for a JSON string field (literal `\n`, the way `claude` emits
4442 /// an agent's result text).
4443 ///
4444 /// The gate value carries the markdown CODE SPAN the live 2026-07-31 run
4445 /// captured — see [`HUMAN_GATE_VALUE`]. A bare unquoted value would test a
4446 /// rendering that has never been observed in production.
4447 fn gate_declaration_text() -> String {
4448 format!(
4449 "## CHECKPOINT REACHED\\n\\n**Type:** decision\\n**Gate:** `{HUMAN_GATE_VALUE}`\\n**Plan:** 30-05\\n"
4450 )
4451 }
4452
4453 /// Text that merely DOCUMENTS a gate rendering — the shape a plan file, a
4454 /// GSD reference document, or an agent narrating its next task carries.
4455 /// Same code-span rendering as a real declaration, which is precisely why a
4456 /// substring scan cannot tell the two apart and the EVENT must decide.
4457 ///
4458 /// Single line, no double quotes, so it drops into a JSON string field
4459 /// without further escaping.
4460 fn gate_documenting_text() -> String {
4461 format!(
4462 "The next task is declared **Gate:** `{HUMAN_GATE_VALUE}` in the plan, so the executor must stop rather than auto-select."
4463 )
4464 }
4465
4466 // Synthetic `result`-text payloads (modification 1). Written exactly as
4467 // they appear INSIDE the envelope's `result` JSON string — escaped quotes
4468 // and an escaped newline — because that is how `claude` emits an agent's
4469 // final message. Once serde decodes the field the `\n` becomes a real
4470 // newline and `parse_marker_lines`' line scan works on it unmodified.
4471 const MARKER_SUCCESS: &str = r#"Plan complete.\nDEVFLOW_RESULT: {\"status\":\"success\"}"#;
4472 const MARKER_FAILED: &str =
4473 r#"Blocked.\nDEVFLOW_RESULT: {\"status\":\"failed\",\"reason\":\"earlier turn aborted\"}"#;
4474 const MARKER_PLANTED_LAYER: &str =
4475 r#"Done.\nDEVFLOW_RESULT: {\"status\":\"success\",\"decided_by_layer\":0}"#;
4476 const NO_MARKER: &str = r#"Acknowledged; nothing to report."#;
4477
4478 /// Fill one real envelope's `result` field with a synthetic payload.
4479 fn v3_result_event(envelope: &str, escaped_result_text: &str) -> String {
4480 assert!(
4481 envelope.contains("__MARKER__"),
4482 "fixture envelope lost its result-text sentinel"
4483 );
4484 envelope.replace("__MARKER__", escaped_result_text)
4485 }
4486
4487 /// Assemble a three-turn Claude stream capture: the real `init` event
4488 /// followed by all three real `result` envelopes, each carrying the given
4489 /// payload. Three result events (not two) is load-bearing — a two-event
4490 /// fixture cannot tell "last wins" apart from "highest index of two".
4491 fn v3_stream_capture(turn1: &str, turn2: &str, turn3: &str) -> String {
4492 format!(
4493 "{}\n{}\n{}\n{}\n",
4494 V3_INIT_EVENT,
4495 v3_result_event(V3_RESULT_TURN1, turn1),
4496 v3_result_event(V3_RESULT_TURN2, turn2),
4497 v3_result_event(V3_RESULT_TURN3, turn3),
4498 )
4499 }
4500
4501 // ---- rate-limit / envelope-failure fixtures (plan 30-03) --------------
4502
4503 /// v3 line 15, **VERBATIM** — the only `rate_limit_event` in any archived
4504 /// capture, and the reason this plan exists in its current form.
4505 ///
4506 /// Read it before touching [`detect_claude_stream_rate_limit`]: its
4507 /// `rate_limit_info.status` is **`allowed`**. The CLI emits these events as
4508 /// routine quota telemetry on healthy streams — this one sits at line 15 of
4509 /// a capture that then completed three turns successfully (results at 19,
4510 /// 37 and 54). Presence of the event type carries NO information about
4511 /// whether the run was blocked.
4512 ///
4513 /// Note the second trap one level down: `overageStatus` is `rejected`. Any
4514 /// nested search for the token `rejected` (e.g. via [`json_find_key`]) also
4515 /// misclassifies this healthy event, which is why the classifier reads
4516 /// `rate_limit_info.status` and nothing else, by direct `.get()`.
4517 const V3_RATE_LIMIT_EVENT_ALLOWED: &str = r#"{"type":"rate_limit_event","rate_limit_info":{"status":"allowed","resetsAt":1785645600,"rateLimitType":"five_hour","overageStatus":"rejected","overageDisabledReason":"out_of_credits","isUsingOverage":false},"uuid":"e73e6774-a79d-4cdf-90bd-53a695f44f5a","session_id":"559fef4d-2053-459e-b7a7-f3200c3b3790"}"#;
4518
4519 /// A `rate_limit_event` with the given `rate_limit_info.status`, built by
4520 /// substituting one field of the real archived event above.
4521 ///
4522 /// **SYNTHETIC for every status except `allowed`.** No archived capture
4523 /// contains a blocked stream — the denial fixtures below are constructed,
4524 /// not observed, and are labelled as such at each use. Every other field
4525 /// (including `resetsAt`, which supplies the retry hint) is exactly as
4526 /// captured.
4527 fn v3_rate_limit_event(status: &str) -> String {
4528 assert!(
4529 V3_RATE_LIMIT_EVENT_ALLOWED.contains(r#""status":"allowed""#),
4530 "fixture lost its status field"
4531 );
4532 V3_RATE_LIMIT_EVENT_ALLOWED
4533 .replace(r#""status":"allowed""#, &format!(r#""status":"{status}""#))
4534 }
4535
4536 /// One real `result` envelope with its captured `is_error":false` flipped
4537 /// to `true`, every other field untouched. The assertion makes the
4538 /// substitution non-silent: if the fixture text ever changes, the test
4539 /// fails loudly rather than quietly testing an `is_error: false` envelope.
4540 fn v3_result_event_is_error(envelope: &str, escaped_result_text: &str) -> String {
4541 let filled = v3_result_event(envelope, escaped_result_text);
4542 assert!(
4543 filled.contains(r#""is_error":false"#),
4544 "fixture envelope lost its is_error field"
4545 );
4546 filled.replace(r#""is_error":false"#, r#""is_error":true"#)
4547 }
4548
4549 /// Assemble a capture from the real `init` event followed by the given
4550 /// lines in order. Unlike [`v3_stream_capture`] this lets a test position a
4551 /// `rate_limit_event` at an arbitrary index, which is the whole point of
4552 /// the final-turn scoping assertions.
4553 fn stream_capture_of(lines: &[&str]) -> String {
4554 let mut out = String::from(V3_INIT_EVENT);
4555 for line in lines {
4556 out.push('\n');
4557 out.push_str(line);
4558 }
4559 out.push('\n');
4560 out
4561 }
4562
4563 /// **The mandatory negative regression.** The real archived stream — whose
4564 /// `rate_limit_event` says `status: "allowed"` and which then completed
4565 /// three turns — must NOT classify as `RateLimited`.
4566 ///
4567 /// This event is routine quota telemetry, not a block. Classifying its mere
4568 /// presence as a rate limit would route EVERY healthy Claude stream stage
4569 /// into `Action::AutoResume` against a fabricated retry time, instead of
4570 /// advancing the pipeline. That mapping is
4571 /// `crates/devflow-core/src/outcome_policy.rs:41` — `AgentStatus::RateLimited
4572 /// => Action::AutoResume`, re-read in this crate at execution time; 30-03's
4573 /// plan and threat register cite it as `outcome_policy.rs:41` without a
4574 /// crate, and it is NOT in `devflow-cli`. This is a denial of service on
4575 /// the whole product, produced by a one-line "detect the event type"
4576 /// shortcut.
4577 ///
4578 /// Two independent guards must both hold here, and the second assertion
4579 /// pins the one the positioning guard alone would hide: the event is placed
4580 /// at its real position (before the first `result`, mirroring line 15 vs
4581 /// 19), AND its status is not a denial. `detect_claude_stream_rate_limit`
4582 /// is asserted directly on a final-turn placement of the same real event so
4583 /// the status guard cannot be dropped without this test failing.
4584 #[test]
4585 fn claude_stream_real_allowed_rate_limit_event_is_not_rate_limited() {
4586 let capture = stream_capture_of(&[
4587 V3_RATE_LIMIT_EVENT_ALLOWED,
4588 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4589 &v3_result_event(V3_RESULT_TURN2, NO_MARKER),
4590 &v3_result_event(V3_RESULT_TURN3, MARKER_SUCCESS),
4591 ]);
4592
4593 let result = parse_claude_event_result(&capture)
4594 .expect("the final turn's success marker still decides this stream");
4595 assert_eq!(result.status, AgentStatus::Success);
4596 assert_ne!(result.status, AgentStatus::RateLimited);
4597
4598 // The status guard on its own: the SAME real event moved into the final
4599 // turn (after the second-to-last `result`) is still not a rate limit.
4600 // Without this, deleting the status check would leave the test green.
4601 let final_turn = stream_capture_of(&[
4602 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4603 &v3_result_event(V3_RESULT_TURN2, NO_MARKER),
4604 V3_RATE_LIMIT_EVENT_ALLOWED,
4605 &v3_result_event(V3_RESULT_TURN3, MARKER_SUCCESS),
4606 ]);
4607 assert!(
4608 detect_claude_stream_rate_limit(&ParsedCapture::parse(&final_turn).events).is_none()
4609 );
4610 }
4611
4612 /// The positive: an explicit quota DENIAL inside the final turn classifies
4613 /// as `RateLimited`, so the rate-limit resume path stays reachable under
4614 /// `stream-json`.
4615 ///
4616 /// **The denial fixture is SYNTHETIC.** No archived capture contains a
4617 /// blocked stream, so the `rejected` status is constructed from the
4618 /// observed vocabulary of this schema rather than observed in the wild —
4619 /// the same honest-fixture rule this phase applies to marker payloads. The
4620 /// retry hint comes from the real `resetsAt` value.
4621 #[test]
4622 fn claude_stream_final_turn_denial_rate_limit_event_is_rate_limited() {
4623 let denial = v3_rate_limit_event("rejected");
4624 let capture = stream_capture_of(&[
4625 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4626 &v3_result_event(V3_RESULT_TURN2, NO_MARKER),
4627 &denial,
4628 &v3_result_event(V3_RESULT_TURN3, NO_MARKER),
4629 ]);
4630
4631 let result = parse_claude_event_result(&capture)
4632 .expect("a final-turn quota denial must produce a Layer-1 verdict");
4633 assert_eq!(result.status, AgentStatus::RateLimited);
4634 assert_eq!(
4635 result.reason.as_deref(),
4636 Some("rate limited until 1785645600")
4637 );
4638 assert_eq!(result.decided_by_layer, Some(1));
4639
4640 // Fewer than two `result` events means the whole stream IS the final
4641 // turn — a run blocked before it ever completed a turn must still
4642 // classify, or the boundary logic silently swallows the common case.
4643 let single_turn =
4644 stream_capture_of(&[&denial, &v3_result_event(V3_RESULT_TURN1, NO_MARKER)]);
4645 assert_eq!(
4646 parse_claude_event_result(&single_turn).map(|r| r.status),
4647 Some(AgentStatus::RateLimited)
4648 );
4649 }
4650
4651 /// Scoping: a denial that predates the final turn cannot outrank the final
4652 /// turn's own outcome. Rate-limit chatter from an earlier turn must not
4653 /// decide a stream that later completed — in the real capture the rate
4654 /// event (line 15) precedes all three results, so an unscoped detector
4655 /// would let a first-turn event decide a stream that finished forty seconds
4656 /// later.
4657 ///
4658 /// The denial status here is the SAME one the positive test proves does
4659 /// classify, so this test can only pass because of the POSITION guard.
4660 #[test]
4661 fn claude_stream_denial_before_final_turn_does_not_outrank_final_result() {
4662 let capture = stream_capture_of(&[
4663 &v3_rate_limit_event("rejected"),
4664 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4665 &v3_result_event(V3_RESULT_TURN2, NO_MARKER),
4666 &v3_result_event(V3_RESULT_TURN3, MARKER_SUCCESS),
4667 ]);
4668
4669 let result = parse_claude_event_result(&capture)
4670 .expect("the final turn's success marker decides this stream");
4671 assert_eq!(result.status, AgentStatus::Success);
4672 }
4673
4674 /// An unrecognised `rate_limit_info.status` DEFERS rather than classifying.
4675 ///
4676 /// Deferring is the deliberately safe direction: an unknown denial status
4677 /// falls through to the envelope/marker paths and is reported `Failed` — a
4678 /// real degradation (the operator loses automatic resume) but a never-silent
4679 /// one that still gates. The opposite error auto-resumes a healthy stream
4680 /// against a retry time the parser invented.
4681 ///
4682 /// Positioned in the FINAL turn, so only the status check can decline it.
4683 #[test]
4684 fn claude_stream_unrecognised_rate_limit_status_defers() {
4685 let capture = stream_capture_of(&[
4686 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4687 &v3_result_event(V3_RESULT_TURN2, NO_MARKER),
4688 &v3_rate_limit_event("some_future_status"),
4689 &v3_result_event(V3_RESULT_TURN3, MARKER_SUCCESS),
4690 ]);
4691
4692 assert!(detect_claude_stream_rate_limit(&ParsedCapture::parse(&capture).events).is_none());
4693 let result = parse_claude_event_result(&capture)
4694 .expect("the parser must fall through to the marker path");
4695 assert_eq!(result.status, AgentStatus::Success);
4696 }
4697
4698 /// Precedence (T-30-13): when the detector fires, rate limit outranks the
4699 /// marker path. A rate-limited run classified as generic `Failed` kills the
4700 /// primary rate-limit resume cron — the one path that exists to recover
4701 /// from it — which is exactly why `evaluate_layer1` already orders
4702 /// `detect_claude_rate_limit` ahead of `detect_claude_envelope_failure` for
4703 /// the single-document path.
4704 ///
4705 /// Non-vacuous: the same capture WITHOUT the rate event yields `Failed`, so
4706 /// this test fails the moment the ordering is reshuffled.
4707 #[test]
4708 fn claude_stream_final_turn_denial_outranks_failed_marker() {
4709 let with_denial = stream_capture_of(&[
4710 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4711 &v3_result_event(V3_RESULT_TURN2, NO_MARKER),
4712 &v3_rate_limit_event("rejected"),
4713 &v3_result_event(V3_RESULT_TURN3, MARKER_FAILED),
4714 ]);
4715 assert_eq!(
4716 parse_claude_event_result(&with_denial).map(|r| r.status),
4717 Some(AgentStatus::RateLimited)
4718 );
4719
4720 let without_denial = v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_FAILED);
4721 assert_eq!(
4722 parse_claude_event_result(&without_denial).map(|r| r.status),
4723 Some(AgentStatus::Failed)
4724 );
4725 }
4726
4727 /// A last `result` event with `is_error: true` and NO marker is an
4728 /// authoritative Layer-1 failure, not a deferral to Layer 2's coarse
4729 /// exit-code heuristic — matching `detect_claude_envelope_failure` for the
4730 /// single-document envelope. The reason is drawn from the event's own
4731 /// `result` text with the `num_turns` suffix, the same shape that function
4732 /// produces.
4733 #[test]
4734 fn claude_stream_last_result_is_error_without_marker_is_failed() {
4735 let capture = stream_capture_of(&[
4736 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4737 &v3_result_event(V3_RESULT_TURN2, NO_MARKER),
4738 &v3_result_event_is_error(V3_RESULT_TURN3, r#"Execution error: context exhausted"#),
4739 ]);
4740
4741 let result = parse_claude_event_result(&capture)
4742 .expect("is_error on the last result must not defer to Layer 2");
4743 assert_eq!(result.status, AgentStatus::Failed);
4744 assert_eq!(
4745 result.reason.as_deref(),
4746 Some("Execution error: context exhausted (num_turns: 2)")
4747 );
4748 assert_eq!(result.decided_by_layer, Some(1));
4749 }
4750
4751 /// Envelope-over-marker (T-30-15): `is_error: true` overrides a SUCCESS
4752 /// marker in the same event, matching `detect_claude_envelope_failure`'s
4753 /// documented precedence over a stale or echoed success marker.
4754 ///
4755 /// Non-vacuous: the identical capture with `is_error: false` yields
4756 /// `Success`, so the assertion below can only pass because the envelope
4757 /// check overrode the marker.
4758 #[test]
4759 fn claude_stream_is_error_overrides_success_marker() {
4760 let capture = stream_capture_of(&[
4761 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4762 &v3_result_event(V3_RESULT_TURN2, NO_MARKER),
4763 &v3_result_event_is_error(V3_RESULT_TURN3, MARKER_SUCCESS),
4764 ]);
4765 let result = parse_claude_event_result(&capture)
4766 .expect("is_error must produce a verdict even with a success marker");
4767 assert_eq!(result.status, AgentStatus::Failed);
4768
4769 let healthy = v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS);
4770 assert_eq!(
4771 parse_claude_event_result(&healthy).map(|r| r.status),
4772 Some(AgentStatus::Success)
4773 );
4774 }
4775
4776 // ---- session id from a stream capture (plan 30-03 Task 2) -------------
4777
4778 /// The single `session_id` every event in the archived v3 capture carries —
4779 /// all three `init` events (lines 5, 32 and 47) and all three `result`
4780 /// events agree on it, confirmed by reading the capture.
4781 const V3_SESSION_ID: &str = "559fef4d-2053-459e-b7a7-f3200c3b3790";
4782
4783 /// The real `init` event with its `session_id` substituted. Used only to
4784 /// build a SYNTHETIC mid-stream rotation — no archived capture rotates.
4785 fn v3_init_event_with_session(session_id: &str) -> String {
4786 assert!(
4787 V3_INIT_EVENT.contains(V3_SESSION_ID),
4788 "fixture lost its session_id"
4789 );
4790 V3_INIT_EVENT.replace(V3_SESSION_ID, session_id)
4791 }
4792
4793 /// `claude_stream_session_id` reads the CLI-emitted id out of a JSONL
4794 /// capture built from the archived `init` events (v3 lines 5, 32 and 47 —
4795 /// all three carry this same value).
4796 ///
4797 /// The second half pins LAST-init-wins with a synthetic rotation: the real
4798 /// capture's three `init` events are identical, so first-wins and last-wins
4799 /// agree on today's evidence and a fixture built only from it cannot tell
4800 /// the two apart. Three `init` events do NOT mean three sessions.
4801 #[test]
4802 fn claude_stream_session_id_reads_cli_emitted_init_value() {
4803 let capture = stream_capture_of(&[
4804 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4805 V3_INIT_EVENT,
4806 &v3_result_event(V3_RESULT_TURN2, NO_MARKER),
4807 V3_INIT_EVENT,
4808 &v3_result_event(V3_RESULT_TURN3, MARKER_SUCCESS),
4809 ]);
4810 assert_eq!(
4811 claude_stream_session_id(&capture).as_deref(),
4812 Some(V3_SESSION_ID)
4813 );
4814
4815 let rotated = stream_capture_of(&[
4816 &v3_result_event(V3_RESULT_TURN1, NO_MARKER),
4817 &v3_init_event_with_session("second-session-id"),
4818 &v3_result_event(V3_RESULT_TURN2, MARKER_SUCCESS),
4819 ]);
4820 assert_eq!(
4821 claude_stream_session_id(&rotated).as_deref(),
4822 Some("second-session-id")
4823 );
4824 }
4825
4826 /// D-04 / T-28-04 forgery guard for the stream path — the analog of
4827 /// `session_id_in_devflow_result_marker_is_not_returned`, which pins the
4828 /// same contract for the single-document envelope.
4829 ///
4830 /// The fixture defeats BOTH plausible wrong implementations at once: a
4831 /// nested traversal (`json_find_key`/`json_scan`) would reach the
4832 /// `session_id` the agent planted inside its own `DEVFLOW_RESULT` marker
4833 /// text, and a "last event carrying a `session_id`" scan would return the
4834 /// final `result` event's own key. Both are wrong; only the `init` event's
4835 /// top-level value is CLI-emitted. The divergence between the `result`
4836 /// event's id and the `init` event's is synthetic — no archived capture
4837 /// diverges — and exists purely so those two implementations cannot pass.
4838 #[test]
4839 fn claude_stream_session_id_ignores_agent_planted_value() {
4840 const PLANTED_MARKER: &str =
4841 r#"Done.\nDEVFLOW_RESULT: {\"status\":\"success\",\"session_id\":\"forged-by-agent\"}"#;
4842
4843 let last_result = v3_result_event(V3_RESULT_TURN3, PLANTED_MARKER)
4844 .replace(V3_SESSION_ID, "result-event-session-id");
4845 let capture =
4846 stream_capture_of(&[&v3_result_event(V3_RESULT_TURN1, NO_MARKER), &last_result]);
4847
4848 // Non-vacuity: both decoys really are present in the capture text, so a
4849 // wrong implementation has something wrong to find.
4850 assert!(capture.contains("forged-by-agent"));
4851 assert!(capture.contains("result-event-session-id"));
4852
4853 assert_eq!(
4854 claude_stream_session_id(&capture).as_deref(),
4855 Some(V3_SESSION_ID)
4856 );
4857 }
4858
4859 /// The stream reader does not shadow or duplicate `claude_session_id`: it
4860 /// declines the single-document envelope (the exact literal
4861 /// `session_id_reads_top_level_string` asserts on) and plain text, so the
4862 /// wrapper's stream-first ordering cannot change today's behavior.
4863 #[test]
4864 fn claude_stream_session_id_declines_non_stream_shapes() {
4865 let envelope = r#"{"type":"result","subtype":"success","result":"All done.","session_id":"cf29bfec-69e8-45df-a4f3-3da08ab6f66e"}"#;
4866 assert!(claude_stream_session_id(envelope).is_none());
4867 // ...and the shipped reader still owns it, so declining costs nothing.
4868 assert_eq!(
4869 claude_session_id(envelope).as_deref(),
4870 Some("cf29bfec-69e8-45df-a4f3-3da08ab6f66e")
4871 );
4872
4873 assert!(claude_stream_session_id("just some plain text output\n").is_none());
4874 }
4875
4876 /// The wiring that matters: `session_id_from_capture` — the Phase 28
4877 /// checkpoint-resume reader (`claude --resume` needs an id DevFlow can
4878 /// read) — returns an id for a JSONL capture, where before this plan it
4879 /// returned `None` for every stream capture.
4880 #[test]
4881 fn claude_stream_session_id_from_capture_reads_jsonl() {
4882 let dir = tempfile::tempdir().unwrap();
4883 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
4884 std::fs::write(
4885 stdout_path(dir.path(), 30),
4886 v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS),
4887 )
4888 .unwrap();
4889
4890 assert_eq!(
4891 session_id_from_capture(dir.path(), 30).as_deref(),
4892 Some(V3_SESSION_ID)
4893 );
4894 }
4895
4896 /// The other half of the wiring claim: a single-document envelope capture
4897 /// still yields exactly what it did before the stream reader was inserted
4898 /// ahead of `claude_session_id` in the fallback chain.
4899 #[test]
4900 fn claude_stream_wiring_leaves_single_document_capture_unchanged() {
4901 let dir = tempfile::tempdir().unwrap();
4902 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
4903 let envelope = r#"{"type":"result","subtype":"success","result":"All done.","session_id":"cf29bfec-69e8-45df-a4f3-3da08ab6f66e"}"#;
4904 std::fs::write(stdout_path(dir.path(), 8), envelope).unwrap();
4905
4906 assert_eq!(
4907 session_id_from_capture(dir.path(), 8).as_deref(),
4908 claude_session_id(envelope).as_deref()
4909 );
4910 assert_eq!(
4911 session_id_from_capture(dir.path(), 8).as_deref(),
4912 Some("cf29bfec-69e8-45df-a4f3-3da08ab6f66e")
4913 );
4914 }
4915
4916 /// The tracer: a real archived `stream-json` capture written to
4917 /// `.devflow/phase-NN-stdout` produces a Layer-1 verdict out of
4918 /// `evaluate_layer1`. Before plan 30-01 this returned `None` for every
4919 /// JSONL capture — `serde_json::from_str` on the whole multi-line document
4920 /// is a hard "trailing characters" error, so all four single-document
4921 /// parsers declined it and the stage fell through to Layer 2's coarse
4922 /// exit-code+commit heuristic.
4923 ///
4924 /// Fixture provenance and its two modifications are documented on
4925 /// `V3_INIT_EVENT` / `V3_RESULT_TURN1..3` above.
4926 #[test]
4927 fn evaluate_layer1_parses_claude_stream_capture() {
4928 let dir = tempfile::tempdir().unwrap();
4929 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
4930 std::fs::write(
4931 stdout_path(dir.path(), 30),
4932 v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS),
4933 )
4934 .unwrap();
4935
4936 let result = evaluate_layer1(dir.path(), 30).unwrap();
4937
4938 assert_eq!(result.status, AgentStatus::Success);
4939 assert_eq!(result.decided_by_layer, Some(1));
4940
4941 // Non-vacuity guard for the assertion above: this marker omits
4942 // `decided_by_layer`, and the field is `#[serde(default)]`, so
4943 // `parse_marker_lines` alone yields `None`. `Some(1)` can therefore
4944 // only have come from the parser's explicit overwrite.
4945 assert_eq!(
4946 parse_marker_lines(r#"DEVFLOW_RESULT: {"status":"success"}"#)
4947 .unwrap()
4948 .decided_by_layer,
4949 None
4950 );
4951 }
4952
4953 // ---- idle-timeout side channel (31-02, D-05/D-06/D-07) ---------------
4954
4955 /// Write a monitor-shaped idle-timeout record. Field names and types match
4956 /// `IdleTimeoutRecord` exactly; the monitor writes it via serde, so a drift
4957 /// between the two shows up as a failing deserialize here.
4958 fn write_idle_timeout_record(root: &Path, phase: u32, commits: &[(&str, &str)]) {
4959 let record = IdleTimeoutRecord {
4960 status: AgentStatus::IdleTimeout.as_wire_str().to_string(),
4961 idle_secs: 30,
4962 agent_pid: 4242,
4963 written_at: 1_700_000_000,
4964 commits: commits
4965 .iter()
4966 .map(|(sha, subject)| IdleTimeoutCommit {
4967 sha: (*sha).to_string(),
4968 subject: (*subject).to_string(),
4969 })
4970 .collect(),
4971 };
4972 std::fs::write(
4973 idle_timeout_path(root, phase),
4974 serde_json::to_string(&record).unwrap(),
4975 )
4976 .unwrap();
4977 }
4978
4979 /// T-31-06, and the single most important test in plan 31-02.
4980 ///
4981 /// The fixture is a REAL archived three-turn capture in which every
4982 /// top-level `result` event carries a success marker — the normal shape of
4983 /// a run that got far enough to idle out. A fixture without a prior
4984 /// `result` event would pass vacuously while the same mechanism silently
4985 /// failed in production.
4986 ///
4987 /// The negative control is encoded INSIDE the test rather than described in
4988 /// prose: the same fixture is evaluated first WITHOUT the side channel and
4989 /// must return `Success`. If that ever stops holding, the `IdleTimeout`
4990 /// assertion below is proving a verdict nothing was competing with.
4991 #[test]
4992 fn idle_timeout_side_channel_wins_over_stale_stream_result() {
4993 let dir = tempfile::tempdir().unwrap();
4994 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
4995 std::fs::write(
4996 stdout_path(dir.path(), 40),
4997 v3_stream_capture(MARKER_SUCCESS, MARKER_SUCCESS, MARKER_SUCCESS),
4998 )
4999 .unwrap();
5000
5001 // NEGATIVE CONTROL — must produce the OPPOSITE result.
5002 assert_eq!(
5003 evaluate_layer1(dir.path(), 40).unwrap().status,
5004 AgentStatus::Success,
5005 "negative control: without the side channel this fixture must decide Success, \
5006 otherwise the assertion below is vacuous"
5007 );
5008
5009 write_idle_timeout_record(
5010 dir.path(),
5011 40,
5012 &[("aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", "feat: partial")],
5013 );
5014
5015 let result = evaluate_layer1(dir.path(), 40).unwrap();
5016 assert_eq!(
5017 result.status,
5018 AgentStatus::IdleTimeout,
5019 "a stale success already in the capture must not shadow the monitor's verdict"
5020 );
5021 assert_eq!(result.decided_by_layer, Some(1));
5022 }
5023
5024 /// The read must precede `read_capture`'s early `return None`, so a
5025 /// timeout that fired before the child emitted anything at all is still
5026 /// authoritative rather than discarded.
5027 #[test]
5028 fn idle_timeout_side_channel_is_read_even_when_the_capture_is_missing() {
5029 let dir = tempfile::tempdir().unwrap();
5030 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5031 assert!(
5032 !stdout_path(dir.path(), 41).exists(),
5033 "fixture precondition: there must be no capture at all"
5034 );
5035
5036 // NEGATIVE CONTROL: with neither file present Layer 1 abstains, so the
5037 // verdict below can only have come from the side channel.
5038 assert!(evaluate_layer1(dir.path(), 41).is_none());
5039
5040 write_idle_timeout_record(dir.path(), 41, &[]);
5041
5042 let result = evaluate_layer1(dir.path(), 41).unwrap();
5043 assert_eq!(result.status, AgentStatus::IdleTimeout);
5044 assert_eq!(result.commits, Some(0));
5045 }
5046
5047 /// D-07: the verdict names the commits, and says they were not rolled back.
5048 #[test]
5049 fn idle_timeout_result_carries_the_commits_it_enumerated() {
5050 let dir = tempfile::tempdir().unwrap();
5051 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5052 write_idle_timeout_record(
5053 dir.path(),
5054 42,
5055 &[
5056 ("1111111abcdef0000000000000000000000000000", "feat: first"),
5057 ("2222222abcdef0000000000000000000000000000", "fix: second"),
5058 ],
5059 );
5060
5061 let result = evaluate_layer1(dir.path(), 42).unwrap();
5062
5063 assert_eq!(result.commits, Some(2));
5064 let reason = result.reason.expect("an idle timeout must explain itself");
5065 for fragment in [
5066 "1111111", // short sha, first commit
5067 "feat: first", // its subject
5068 "2222222",
5069 "fix: second",
5070 "30s", // how long the stream was silent
5071 "NONE of them were rolled back", // D-07's non-destruction promise
5072 ] {
5073 assert!(
5074 reason.contains(fragment),
5075 "reason must name {fragment:?}; got: {reason}"
5076 );
5077 }
5078 // The full sha must not be what is printed — a 40-char sha in a gate
5079 // message is noise, and the short form is what an operator pastes.
5080 assert!(!reason.contains("1111111abcdef0000000000000000000000000000"));
5081 }
5082
5083 /// Nothing about the pre-existing cascade changes when no timeout fired.
5084 /// Three shapes, each asserted against the verdict it produced before this
5085 /// plan existed, with the side channel confirmed absent in every one.
5086 #[test]
5087 fn absent_side_channel_leaves_the_cascade_unchanged() {
5088 let dir = tempfile::tempdir().unwrap();
5089 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5090
5091 std::fs::write(
5092 stdout_path(dir.path(), 43),
5093 v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS),
5094 )
5095 .unwrap();
5096 std::fs::write(
5097 stdout_path(dir.path(), 44),
5098 v3_stream_capture(MARKER_SUCCESS, MARKER_SUCCESS, MARKER_FAILED),
5099 )
5100 .unwrap();
5101
5102 for (phase, expected) in [
5103 (43, Some(AgentStatus::Success)),
5104 (44, Some(AgentStatus::Failed)),
5105 (45, None), // no capture, no side channel
5106 ] {
5107 assert!(
5108 !idle_timeout_path(dir.path(), phase).exists(),
5109 "fixture precondition: phase {phase} must have no side channel"
5110 );
5111 assert_eq!(
5112 evaluate_layer1(dir.path(), phase).map(|r| r.status),
5113 expected,
5114 "the cascade changed for phase {phase} with no timeout on disk"
5115 );
5116 }
5117 }
5118
5119 /// The file's PRESENCE is the signal; its contents are enrichment.
5120 ///
5121 /// A corrupt record must NOT fall back into the cascade — that would let
5122 /// the stale success in the capture win, converting a damaged file into a
5123 /// silent wrong advance. This is the same fixture as
5124 /// `idle_timeout_side_channel_wins_over_stale_stream_result`, so the
5125 /// Success it would otherwise decide is real and not hypothetical.
5126 #[test]
5127 fn an_unreadable_idle_timeout_record_still_produces_the_verdict() {
5128 let dir = tempfile::tempdir().unwrap();
5129 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5130 std::fs::write(
5131 stdout_path(dir.path(), 46),
5132 v3_stream_capture(MARKER_SUCCESS, MARKER_SUCCESS, MARKER_SUCCESS),
5133 )
5134 .unwrap();
5135
5136 // NEGATIVE CONTROL: this capture decides Success on its own.
5137 assert_eq!(
5138 evaluate_layer1(dir.path(), 46).unwrap().status,
5139 AgentStatus::Success
5140 );
5141
5142 std::fs::write(idle_timeout_path(dir.path(), 46), "{ this is not json").unwrap();
5143
5144 let result = evaluate_layer1(dir.path(), 46).unwrap();
5145 assert_eq!(result.status, AgentStatus::IdleTimeout);
5146 assert_eq!(
5147 result.commits, None,
5148 "an unreadable record must not invent a commit count"
5149 );
5150 assert!(result.reason.unwrap().contains("unreadable"));
5151 }
5152
5153 /// Last-result-wins. A session kept alive across turns emits one `result`
5154 /// event per turn; only the final one is the session's verdict.
5155 ///
5156 /// Asserts BOTH directions so the test cannot pass by a parser that merely
5157 /// prefers `success`: failed-then-success yields Success, and
5158 /// success-then-failed yields Failed. The middle event carries the same
5159 /// payload as the first, so a parser that stopped at index 1 would also
5160 /// fail.
5161 #[test]
5162 fn claude_stream_last_result_event_wins_over_earlier_results() {
5163 let last_success = v3_stream_capture(MARKER_FAILED, MARKER_FAILED, MARKER_SUCCESS);
5164 let result = parse_claude_event_result(&last_success).unwrap();
5165 assert_eq!(result.status, AgentStatus::Success);
5166
5167 let last_failed = v3_stream_capture(MARKER_SUCCESS, MARKER_SUCCESS, MARKER_FAILED);
5168 let result = parse_claude_event_result(&last_failed).unwrap();
5169 assert_eq!(result.status, AgentStatus::Failed);
5170 assert_eq!(result.reason.as_deref(), Some("earlier turn aborted"));
5171 }
5172
5173 /// T-30-26: `decided_by_layer` is provenance, not decoration.
5174 /// `crates/devflow-cli/src/pipeline_outcomes.rs` (`classify_validate_outcome`)
5175 /// computes `external = decided_by_layer == Some(0) && status == Success`
5176 /// and uses it to tell an externally-probe-verified Validate stage apart
5177 /// from an ordinary one. An agent that writes `"decided_by_layer": 0` into
5178 /// its own marker is claiming a Layer-0 probe provenance it did not earn,
5179 /// so the stream parser overwrites the field unconditionally.
5180 ///
5181 /// This is a runtime assertion on the returned struct, not a source grep —
5182 /// it fails the moment the overwrite is dropped.
5183 #[test]
5184 fn claude_stream_overwrites_agent_planted_decided_by_layer() {
5185 // Non-vacuity guard: prove the planted value really would survive
5186 // deserialization, so the `Some(1)` below is the overwrite at work and
5187 // not an artifact of a marker that failed to parse.
5188 assert_eq!(
5189 parse_marker_lines(r#"DEVFLOW_RESULT: {"status":"success","decided_by_layer":0}"#)
5190 .unwrap()
5191 .decided_by_layer,
5192 Some(0)
5193 );
5194
5195 let capture = v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_PLANTED_LAYER);
5196 let result = parse_claude_event_result(&capture).unwrap();
5197
5198 assert_eq!(result.status, AgentStatus::Success);
5199 assert_eq!(result.decided_by_layer, Some(1));
5200 }
5201
5202 /// A marker-less final turn defers to Layer 2 rather than reporting an
5203 /// unconditional Success — the same convention `parse_codex_event_result`
5204 /// applies to a bare `turn.completed`. A marker-less turn must never
5205 /// silently advance a stage.
5206 ///
5207 /// The FIRST turn carries a success marker, so this also proves the parser
5208 /// does not fall back to an earlier turn's marker when the last one has
5209 /// none.
5210 ///
5211 /// Plan 30-03 addendum: the deferral must hold specifically for
5212 /// `is_error: false`, which is what the real captured envelope carries —
5213 /// asserted below so this reads as a deliberate is_error case rather than
5214 /// an incidental one. Only `is_error: true` may promote a marker-less turn
5215 /// to `Failed`.
5216 #[test]
5217 fn claude_stream_last_result_without_marker_defers() {
5218 let capture = v3_stream_capture(MARKER_SUCCESS, NO_MARKER, NO_MARKER);
5219 assert!(
5220 capture.contains(r#""is_error":false"#),
5221 "the archived envelopes carry is_error:false; this test is about that case"
5222 );
5223 assert!(parse_claude_event_result(&capture).is_none());
5224 }
5225
5226 #[test]
5227 fn evaluate_layer1_reports_rate_limited_without_marker() {
5228 let dir = tempfile::tempdir().unwrap();
5229 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5230 std::fs::write(
5231 stdout_path(dir.path(), 7),
5232 r#"{"type":"result","subtype":"error_rate_limit","retry_after":"2026-06-18T15:45:30Z"}"#,
5233 )
5234 .unwrap();
5235
5236 let result = evaluate_layer1(dir.path(), 7).unwrap();
5237
5238 assert_eq!(result.status, AgentStatus::RateLimited);
5239 assert_eq!(
5240 result.reason.as_deref(),
5241 Some("rate limited until 2026-06-18T15:45:30Z")
5242 );
5243 }
5244
5245 /// A real Claude rate-limit envelope carries `is_error: true` alongside
5246 /// `subtype: "error_rate_limit"`. The specific RateLimited classification
5247 /// must outrank the generic is_error → Failed path, or the primary
5248 /// rate-limit resume cron never triggers for the exact case it exists for.
5249 #[test]
5250 fn evaluate_layer1_rate_limit_envelope_with_is_error_is_rate_limited() {
5251 let dir = tempfile::tempdir().unwrap();
5252 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5253 std::fs::write(
5254 stdout_path(dir.path(), 7),
5255 r#"{"type":"result","subtype":"error_rate_limit","is_error":true,"retry_after":"2026-06-18T15:45:30Z"}"#,
5256 )
5257 .unwrap();
5258
5259 let result = evaluate_layer1(dir.path(), 7).unwrap();
5260
5261 assert_eq!(result.status, AgentStatus::RateLimited);
5262 assert_eq!(
5263 result.reason.as_deref(),
5264 Some("rate limited until 2026-06-18T15:45:30Z")
5265 );
5266 }
5267
5268 /// CR-01 (13-REVIEW.md) completion: the monitor path writes raw agent
5269 /// bytes to the stdout file via sh redirection, so evaluate_layer1 must
5270 /// tolerate invalid UTF-8 rather than silently disabling all Layer-1
5271 /// detection (the blocking-mode capture was fixed; the file read here is
5272 /// the other half of the same bug).
5273 #[test]
5274 fn evaluate_layer1_finds_marker_despite_invalid_utf8_bytes() {
5275 let dir = tempfile::tempdir().unwrap();
5276 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5277 let mut bytes = b"progress \xff\xfe garbage\n".to_vec();
5278 bytes.extend_from_slice(
5279 b"DEVFLOW_RESULT: {\"status\":\"failed\",\"reason\":\"review: bad\"}\n",
5280 );
5281 std::fs::write(stdout_path(dir.path(), 5), bytes).unwrap();
5282
5283 let result = evaluate_layer1(dir.path(), 5).unwrap();
5284
5285 assert_eq!(result.status, AgentStatus::Failed);
5286 assert_eq!(result.reason.as_deref(), Some("review: bad"));
5287 }
5288
5289 #[test]
5290 fn failing_external_probe_outranks_success_marker() {
5291 let dir = tempfile::tempdir().unwrap();
5292 let phase_dir = dir
5293 .path()
5294 .join(".planning/phases/16-pipeline-reliability-hardening");
5295 std::fs::create_dir_all(&phase_dir).unwrap();
5296 std::fs::write(
5297 phase_dir.join("16-03-PLAN.md"),
5298 "---\nphase: 16\nexternal_verify: \"test -f externally-shipped\"\n---\n",
5299 )
5300 .unwrap();
5301 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5302 std::fs::write(
5303 stdout_path(dir.path(), 16),
5304 "DEVFLOW_RESULT: {\"status\":\"success\"}\n",
5305 )
5306 .unwrap();
5307 let state = state_in(dir.path(), 16);
5308
5309 let approval = vec!["test -f externally-shipped".to_string()];
5310 let result = evaluate_agent_result_inner(
5311 dir.path(),
5312 &state,
5313 &GitFlowConfig::default(),
5314 Some(&approval),
5315 )
5316 .unwrap();
5317
5318 assert_eq!(result.status, AgentStatus::Failed);
5319 assert!(
5320 result
5321 .reason
5322 .as_deref()
5323 .is_some_and(|reason| reason.contains("external verification failed"))
5324 );
5325 }
5326
5327 /// D-05 gap 1 / D-06 (17-03): Layer 0 now evaluates on every stage, not
5328 /// only Code.
5329 ///
5330 /// This is the MAIN-CHECKOUT MIRROR of
5331 /// `external_probe_discovers_from_the_worktree_when_the_main_checkout_lacks_the_plan`,
5332 /// and the two must be read together: with no worktree set, discovery and
5333 /// probe execution resolve to the SAME root, so 999.76's relocation of
5334 /// discovery to `execution_root` provably leaves this path untouched.
5335 /// Without this mirror the worktree fixture alone could not distinguish
5336 /// "discovery reads the execution root" from "discovery reads any root
5337 /// that happens to hold the PLAN".
5338 ///
5339 /// It previously set `state.worktree_path` and asserted the opposite
5340 /// direction — that discovery must read `project_root` while probes run in
5341 /// the worktree (review Plan 03 MEDIUM, OpenCode). 999.76 overturned that
5342 /// premise (see [`evaluate_layer0`]'s doc comment), so the fixture was
5343 /// converted rather than deleted: every assertion below is the original
5344 /// one, including the `"external verification failed"` reason text and the
5345 /// final `Success` assertion. Only the two roots' coincidence changed.
5346 #[test]
5347 fn external_probe_discovers_from_project_root_across_every_stage_without_a_worktree() {
5348 let dir = tempfile::tempdir().unwrap();
5349 let phase_dir = dir.path().join(".planning/phases/16-reliability");
5350 std::fs::create_dir_all(&phase_dir).unwrap();
5351 std::fs::write(
5352 phase_dir.join("16-01-PLAN.md"),
5353 "---\nexternal_verify: \"test -f implemented\"\n---\n",
5354 )
5355 .unwrap();
5356 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5357 std::fs::write(
5358 stdout_path(dir.path(), 16),
5359 "DEVFLOW_RESULT: {\"status\":\"success\"}\n",
5360 )
5361 .unwrap();
5362 let mut state = state_in(dir.path(), 16);
5363 // No worktree: `execution_root` falls back to `project_root`, so
5364 // discovery and probe execution read the same directory.
5365 state.worktree_path = None;
5366 state.stage = Stage::Plan;
5367
5368 let approval = vec!["test -f implemented".to_string()];
5369
5370 // Layer 0 now fires on Plan too — the probe file does not yet exist,
5371 // so this must fail on the probe itself (NOT a false PLAN-removed
5372 // veto, which would mean discovery silently returned zero commands).
5373 let plan_result = evaluate_agent_result_inner(
5374 dir.path(),
5375 &state,
5376 &GitFlowConfig::default(),
5377 Some(&approval),
5378 )
5379 .unwrap();
5380 assert_eq!(plan_result.status, AgentStatus::Failed);
5381 assert!(
5382 plan_result
5383 .reason
5384 .as_deref()
5385 .is_some_and(|reason| reason.contains("external verification failed")),
5386 "expected a failing-probe reason, not a false PLAN-removed veto: {:?}",
5387 plan_result.reason
5388 );
5389
5390 state.stage = Stage::Code;
5391 let code_result = evaluate_agent_result_inner(
5392 dir.path(),
5393 &state,
5394 &GitFlowConfig::default(),
5395 Some(&approval),
5396 )
5397 .unwrap();
5398 assert_eq!(code_result.status, AgentStatus::Failed);
5399
5400 // The probe executes against execution_root, which without a worktree
5401 // IS project_root — the coincidence this mirror exists to pin.
5402 std::fs::write(dir.path().join("implemented"), "done").unwrap();
5403 let passing = evaluate_agent_result_inner(
5404 dir.path(),
5405 &state,
5406 &GitFlowConfig::default(),
5407 Some(&approval),
5408 )
5409 .unwrap();
5410 assert_eq!(passing.status, AgentStatus::Success);
5411 assert_eq!(passing.decided_by_layer, Some(0));
5412 }
5413
5414 /// 999.76 (ROADMAP criterion 6): the INVERSE of the fixture above. The
5415 /// PLAN lives only under the worktree and `project_root`'s own
5416 /// `.planning/phases/` is absent entirely — which is what an in-flight
5417 /// phase actually looks like. `.planning/` is tracked content, so a phase's
5418 /// `{N}-PLAN.md` sits on `feature/phase-{N}` INSIDE the worktree and is
5419 /// absent from the main checkout for the phase's whole duration.
5420 ///
5421 /// The live provenance measurement for that layout claim is **NC-7**,
5422 /// recorded in this phase's `34-04-SUMMARY.md`: `git ls-tree -r develop`
5423 /// vs `git ls-tree -r HEAD` over `.planning/phases`, reported with both
5424 /// refs' counts. NC-7 is evidence that the layout manufactured here is the
5425 /// real one — it says nothing about whether this code is correct. That
5426 /// claim is carried by this fixture and by its main-checkout mirror
5427 /// `external_probe_discovers_from_project_root_across_every_stage_without_a_worktree`,
5428 /// which must be read together with it.
5429 #[test]
5430 fn external_probe_discovers_from_the_worktree_when_the_main_checkout_lacks_the_plan() {
5431 let dir = tempfile::tempdir().unwrap();
5432 let worktree = dir.path().join("phase-worktree");
5433 // The PLAN exists ONLY under the worktree — `dir.path()`'s own
5434 // `.planning/phases/` is deliberately never created.
5435 let phase_dir = worktree.join(".planning/phases/16-reliability");
5436 std::fs::create_dir_all(&phase_dir).unwrap();
5437 std::fs::write(
5438 phase_dir.join("16-01-PLAN.md"),
5439 "---\nexternal_verify: \"test -f implemented\"\n---\n",
5440 )
5441 .unwrap();
5442 // Captures live in the project root, not the worktree.
5443 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5444 std::fs::write(
5445 stdout_path(dir.path(), 16),
5446 "DEVFLOW_RESULT: {\"status\":\"success\"}\n",
5447 )
5448 .unwrap();
5449 let mut state = state_in(dir.path(), 16);
5450 state.worktree_path = Some(worktree.clone());
5451
5452 let approval = vec!["test -f implemented".to_string()];
5453
5454 // The probe file does not exist yet, so this must fail ON THE PROBE.
5455 let failing = evaluate_agent_result_inner(
5456 dir.path(),
5457 &state,
5458 &GitFlowConfig::default(),
5459 Some(&approval),
5460 )
5461 .unwrap();
5462 assert_eq!(failing.status, AgentStatus::Failed);
5463 assert!(
5464 failing
5465 .reason
5466 .as_deref()
5467 .is_some_and(|reason| reason.contains("external verification failed")),
5468 "expected a failing-probe reason; a PLAN-removed reason means discovery \
5469 silently returned zero commands — i.e. discovery still reads project_root \
5470 and 999.76's fix did not land: {:?}",
5471 failing.reason
5472 );
5473
5474 std::fs::write(worktree.join("implemented"), "done").unwrap();
5475 let passing = evaluate_agent_result_inner(
5476 dir.path(),
5477 &state,
5478 &GitFlowConfig::default(),
5479 Some(&approval),
5480 )
5481 .unwrap();
5482 assert_eq!(passing.status, AgentStatus::Success);
5483 assert_eq!(passing.decided_by_layer, Some(0));
5484 }
5485
5486 #[test]
5487 fn changed_external_probe_never_inherits_prior_approval() {
5488 let dir = tempfile::tempdir().unwrap();
5489 let phase_dir = dir.path().join(".planning/phases/16-reliability");
5490 std::fs::create_dir_all(&phase_dir).unwrap();
5491 std::fs::write(
5492 phase_dir.join("16-01-PLAN.md"),
5493 "---\nexternal_verify: \"touch escaped\"\n---\n",
5494 )
5495 .unwrap();
5496 let state = state_in(dir.path(), 16);
5497 let approved = vec!["test -f reviewed-artifact".to_string()];
5498
5499 let result = evaluate_agent_result_inner(
5500 dir.path(),
5501 &state,
5502 &GitFlowConfig::default(),
5503 Some(&approved),
5504 )
5505 .unwrap();
5506
5507 assert_eq!(result.status, AgentStatus::Failed);
5508 assert!(result.reason.unwrap().contains("approval mismatch"));
5509 assert!(!dir.path().join("escaped").exists());
5510 }
5511
5512 #[test]
5513 fn removed_external_probe_fails_closed_against_prior_approval() {
5514 let dir = tempfile::tempdir().unwrap();
5515 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5516 std::fs::write(
5517 stdout_path(dir.path(), 16),
5518 "DEVFLOW_RESULT: {\"status\":\"success\"}\n",
5519 )
5520 .unwrap();
5521 let state = state_in(dir.path(), 16);
5522 let approved = vec!["test -f shipped".to_string()];
5523
5524 let result = evaluate_agent_result_inner(
5525 dir.path(),
5526 &state,
5527 &GitFlowConfig::default(),
5528 Some(&approved),
5529 )
5530 .unwrap();
5531
5532 assert_eq!(result.status, AgentStatus::Failed);
5533 assert!(result.reason.unwrap().contains("declaration was removed"));
5534 }
5535
5536 #[test]
5537 fn no_external_declaration_preserves_layer1_result() {
5538 let dir = tempfile::tempdir().unwrap();
5539 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5540 std::fs::write(
5541 stdout_path(dir.path(), 16),
5542 "DEVFLOW_RESULT: {\"status\":\"success\",\"commits\":2,\"summary\":\"done\"}\n",
5543 )
5544 .unwrap();
5545 let state = state_in(dir.path(), 16);
5546 let layer1 = evaluate_layer1(dir.path(), 16).unwrap();
5547
5548 let full = evaluate_agent_result(dir.path(), &state, &GitFlowConfig::default()).unwrap();
5549
5550 assert_eq!(
5551 serde_json::to_value(full).unwrap(),
5552 serde_json::to_value(layer1).unwrap()
5553 );
5554 }
5555
5556 /// D-05 gap 2 (17-03): a declared, operator-approved external
5557 /// post-condition whose probe passes is affirmative Success evidence on
5558 /// its own — even with zero commits and on a non-Code stage (Define
5559 /// here). No agent stdout is written at all, so if Layer 0 did not
5560 /// short-circuit, there would be nothing for Layer 1 to find and Layer 2
5561 /// would fall through for lack of an exit-code file.
5562 #[test]
5563 fn layer0_affirmative_success_on_non_code_stage_with_zero_commits() {
5564 let dir = tempfile::tempdir().unwrap();
5565 let phase_dir = dir.path().join(".planning/phases/16-reliability");
5566 std::fs::create_dir_all(&phase_dir).unwrap();
5567 std::fs::write(
5568 phase_dir.join("16-01-PLAN.md"),
5569 "---\nexternal_verify: \"test -f shipped\"\n---\n",
5570 )
5571 .unwrap();
5572 std::fs::write(dir.path().join("shipped"), "done").unwrap();
5573 let mut state = state_in(dir.path(), 16);
5574 state.stage = Stage::Define;
5575
5576 let approval = vec!["test -f shipped".to_string()];
5577 let result = evaluate_agent_result_inner(
5578 dir.path(),
5579 &state,
5580 &GitFlowConfig::default(),
5581 Some(&approval),
5582 )
5583 .unwrap();
5584
5585 assert_eq!(result.status, AgentStatus::Success);
5586 assert_eq!(result.decided_by_layer, Some(0));
5587 assert_eq!(result.commits, None);
5588 // Off-Validate stage: verdict reconciliation does not apply (18e).
5589 assert_eq!(result.verdict, None);
5590 }
5591
5592 /// Review Plan 03 LOW (Codex+OpenCode), 16a: an approved all-passing
5593 /// Layer 0 probe intentionally outranks a Layer 1 self-reported failure
5594 /// marker — proven here at the cascade level (`evaluate_agent_result_inner`),
5595 /// not merely in isolation on `evaluate_layer0`.
5596 #[test]
5597 fn layer0_affirmative_success_outranks_layer1_failure_marker() {
5598 let dir = tempfile::tempdir().unwrap();
5599 let phase_dir = dir
5600 .path()
5601 .join(".planning/phases/16-pipeline-reliability-hardening");
5602 std::fs::create_dir_all(&phase_dir).unwrap();
5603 std::fs::write(
5604 phase_dir.join("16-03-PLAN.md"),
5605 "---\nphase: 16\nexternal_verify: \"test -f externally-shipped\"\n---\n",
5606 )
5607 .unwrap();
5608 std::fs::write(dir.path().join("externally-shipped"), "done").unwrap();
5609 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5610 std::fs::write(
5611 stdout_path(dir.path(), 16),
5612 "DEVFLOW_RESULT: {\"status\":\"failed\",\"reason\":\"agent self-reported failure\"}\n",
5613 )
5614 .unwrap();
5615 let state = state_in(dir.path(), 16);
5616
5617 let approval = vec!["test -f externally-shipped".to_string()];
5618 let result = evaluate_agent_result_inner(
5619 dir.path(),
5620 &state,
5621 &GitFlowConfig::default(),
5622 Some(&approval),
5623 )
5624 .unwrap();
5625
5626 assert_eq!(result.status, AgentStatus::Success);
5627 assert_eq!(result.decided_by_layer, Some(0));
5628 // Off-Validate stage (Code): verdict reconciliation does not apply,
5629 // even though Layer 1's marker here reports a (failure) status (18e).
5630 assert_eq!(result.verdict, None);
5631 }
5632
5633 /// D-05/18e: Layer 0's affirmative-success arm at `Stage::Validate` must
5634 /// consult Layer 1's verdict rather than discard it — the two-signal
5635 /// reconciliation `reconcile_layer0_verdict` adds. Covers all three
5636 /// verdict states Layer 1 can produce: pass, gaps, and no marker at all.
5637 ///
5638 /// D-15 (34-01) adds a FOURTH case: the self-contradictory marker
5639 /// `{"status":"failed","verdict":"pass"}`. "Consult Layer 1's verdict" was
5640 /// implemented as "read Layer 1's verdict and nothing else", so an agent
5641 /// that reported its own failure while claiming a passing verdict had that
5642 /// verdict grafted onto Layer 0's `Success` — 999.74's real route. The
5643 /// fourth case pins `verdict: None` for it; before the fix it observed
5644 /// `Some(Pass)`.
5645 #[test]
5646 fn layer0_affirmative_success_consults_layer1_verdict_at_validate() {
5647 let dir = tempfile::tempdir().unwrap();
5648 let phase_dir = dir.path().join(".planning/phases/16-reliability");
5649 std::fs::create_dir_all(&phase_dir).unwrap();
5650 std::fs::write(
5651 phase_dir.join("16-01-PLAN.md"),
5652 "---\nexternal_verify: \"test -f shipped\"\n---\n",
5653 )
5654 .unwrap();
5655 std::fs::write(dir.path().join("shipped"), "done").unwrap();
5656 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5657 let mut state = state_in(dir.path(), 16);
5658 state.stage = Stage::Validate;
5659 let approval = vec!["test -f shipped".to_string()];
5660
5661 std::fs::write(
5662 stdout_path(dir.path(), 16),
5663 "DEVFLOW_RESULT: {\"status\":\"success\",\"verdict\":\"pass\"}\n",
5664 )
5665 .unwrap();
5666 let result = evaluate_agent_result_inner(
5667 dir.path(),
5668 &state,
5669 &GitFlowConfig::default(),
5670 Some(&approval),
5671 )
5672 .unwrap();
5673 assert_eq!(result.status, AgentStatus::Success);
5674 assert_eq!(result.decided_by_layer, Some(0));
5675 assert_eq!(result.verdict, Some(Verdict::Pass));
5676
5677 std::fs::write(
5678 stdout_path(dir.path(), 16),
5679 "DEVFLOW_RESULT: {\"status\":\"success\",\"verdict\":\"gaps\"}\n",
5680 )
5681 .unwrap();
5682 let result = evaluate_agent_result_inner(
5683 dir.path(),
5684 &state,
5685 &GitFlowConfig::default(),
5686 Some(&approval),
5687 )
5688 .unwrap();
5689 assert_eq!(result.verdict, Some(Verdict::Gaps));
5690
5691 std::fs::remove_file(stdout_path(dir.path(), 16)).unwrap();
5692 let result = evaluate_agent_result_inner(
5693 dir.path(),
5694 &state,
5695 &GitFlowConfig::default(),
5696 Some(&approval),
5697 )
5698 .unwrap();
5699 assert_eq!(result.verdict, None);
5700
5701 // D-15: the self-contradictory marker. Layer 1 reports its own run
5702 // FAILED and simultaneously claims a passing verdict. Pre-fix the graft
5703 // read only `.verdict` and produced `Some(Pass)`, i.e. an affirmative
5704 // pair `decide_action` advances and `classify_validate_outcome` reads
5705 // as Passed — Ship, unattended, on a run whose agent reported failure.
5706 std::fs::write(
5707 stdout_path(dir.path(), 16),
5708 "DEVFLOW_RESULT: {\"status\":\"failed\",\"verdict\":\"pass\"}\n",
5709 )
5710 .unwrap();
5711 let result = evaluate_agent_result_inner(
5712 dir.path(),
5713 &state,
5714 &GitFlowConfig::default(),
5715 Some(&approval),
5716 )
5717 .unwrap();
5718 assert_eq!(
5719 result.verdict, None,
5720 "a verdict attached to a self-reported failure must not be grafted (D-15)"
5721 );
5722 // The fix touches `.verdict` only — Layer 0 still decided the status.
5723 assert_eq!(result.status, AgentStatus::Success);
5724 assert_eq!(result.decided_by_layer, Some(0));
5725 }
5726
5727 /// D-15 / ROADMAP criterion 4: `reconcile_layer0_verdict` must consult
5728 /// Layer 1's own `AgentStatus` before transplanting its `verdict`.
5729 ///
5730 /// A regression here costs an unattended Ship on a run whose agent reported
5731 /// failure: the graft would rebuild `(Success, Some(Pass), Some(0))` from a
5732 /// self-contradictory marker, `decide_action` would advance it, and
5733 /// `classify_validate_outcome` would classify Validate as `Passed`.
5734 ///
5735 /// Also carries NC-5's two discrimination cases, which share this fixture.
5736 /// The exploit needs BOTH marker fields; removing either must not reach an
5737 /// affirmative pair. The mandatory opposite-result control lives in
5738 /// `layer0_verdict_graft_still_transplants_a_passing_layer1_verdict` — if
5739 /// that test also produced `None` the fix would be indiscriminate and this
5740 /// one would prove nothing.
5741 #[test]
5742 fn layer0_verdict_graft_declines_when_layer1_status_is_not_success() {
5743 let dir = tempfile::tempdir().unwrap();
5744 let phase_dir = dir.path().join(".planning/phases/16-reliability");
5745 std::fs::create_dir_all(&phase_dir).unwrap();
5746 std::fs::write(
5747 phase_dir.join("16-01-PLAN.md"),
5748 "---\nexternal_verify: \"test -f shipped\"\n---\n",
5749 )
5750 .unwrap();
5751 std::fs::write(dir.path().join("shipped"), "done").unwrap();
5752 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5753 let mut state = state_in(dir.path(), 16);
5754 state.stage = Stage::Validate;
5755 let approval = vec!["test -f shipped".to_string()];
5756
5757 // The exploit itself: both fields present and mutually contradictory.
5758 std::fs::write(
5759 stdout_path(dir.path(), 16),
5760 "DEVFLOW_RESULT: {\"status\":\"failed\",\"verdict\":\"pass\"}\n",
5761 )
5762 .unwrap();
5763 let result = evaluate_agent_result_inner(
5764 dir.path(),
5765 &state,
5766 &GitFlowConfig::default(),
5767 Some(&approval),
5768 )
5769 .unwrap();
5770 assert_eq!(
5771 result.verdict, None,
5772 "self-contradictory marker: the verdict must be declined (D-15)"
5773 );
5774 assert_eq!(result.status, AgentStatus::Success);
5775 assert_eq!(result.decided_by_layer, Some(0));
5776
5777 // NC-5a: removes the `verdict` FIELD, keeps the failed status. `None`
5778 // both pre- and post-fix, so this case cannot discriminate the fix —
5779 // that is the point. The failed status alone is not the exploit.
5780 std::fs::write(
5781 stdout_path(dir.path(), 16),
5782 "DEVFLOW_RESULT: {\"status\":\"failed\"}\n",
5783 )
5784 .unwrap();
5785 let result = evaluate_agent_result_inner(
5786 dir.path(),
5787 &state,
5788 &GitFlowConfig::default(),
5789 Some(&approval),
5790 )
5791 .unwrap();
5792 assert_eq!(
5793 result.verdict, None,
5794 "NC-5a removes the `verdict` field: there is no verdict to graft, \
5795 so the result must be None whether or not the fix is present"
5796 );
5797
5798 // NC-5b: removes `verdict: pass` SPECIFICALLY by downgrading it to
5799 // `gaps`, keeping both fields present. Pre-fix this grafted
5800 // `Some(Gaps)`; post-fix it declines like any other non-Success
5801 // Layer 1. Neither state is an affirmative pair — the exploit needs
5802 // `pass`, not merely any verdict.
5803 std::fs::write(
5804 stdout_path(dir.path(), 16),
5805 "DEVFLOW_RESULT: {\"status\":\"failed\",\"verdict\":\"gaps\"}\n",
5806 )
5807 .unwrap();
5808 let result = evaluate_agent_result_inner(
5809 dir.path(),
5810 &state,
5811 &GitFlowConfig::default(),
5812 Some(&approval),
5813 )
5814 .unwrap();
5815 assert_ne!(
5816 result.verdict,
5817 Some(Verdict::Pass),
5818 "NC-5b removes `verdict: pass` by downgrading it to `gaps`: this \
5819 case must never reach an affirmative pair"
5820 );
5821 assert_eq!(result.verdict, None);
5822 }
5823
5824 /// NC-5's positive half: the fix declines ONLY when Layer 1's own status is
5825 /// not `Success`, never indiscriminately.
5826 ///
5827 /// This is the case that must produce the OPPOSITE result from
5828 /// `layer0_verdict_graft_declines_when_layer1_status_is_not_success`. If
5829 /// both produced `None` the fix would have disabled 18e's legitimate
5830 /// reconciliation wholesale — re-introducing the 17-03 regression that
5831 /// `reconcile_layer0_verdict` exists to fix — and the pair would prove
5832 /// nothing about D-15, because a measurement whose two arms agree is
5833 /// broken rather than informative.
5834 #[test]
5835 fn layer0_verdict_graft_still_transplants_a_passing_layer1_verdict() {
5836 let dir = tempfile::tempdir().unwrap();
5837 let phase_dir = dir.path().join(".planning/phases/16-reliability");
5838 std::fs::create_dir_all(&phase_dir).unwrap();
5839 std::fs::write(
5840 phase_dir.join("16-01-PLAN.md"),
5841 "---\nexternal_verify: \"test -f shipped\"\n---\n",
5842 )
5843 .unwrap();
5844 std::fs::write(dir.path().join("shipped"), "done").unwrap();
5845 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5846 let mut state = state_in(dir.path(), 16);
5847 state.stage = Stage::Validate;
5848 let approval = vec!["test -f shipped".to_string()];
5849
5850 std::fs::write(
5851 stdout_path(dir.path(), 16),
5852 "DEVFLOW_RESULT: {\"status\":\"success\",\"verdict\":\"pass\"}\n",
5853 )
5854 .unwrap();
5855 let result = evaluate_agent_result_inner(
5856 dir.path(),
5857 &state,
5858 &GitFlowConfig::default(),
5859 Some(&approval),
5860 )
5861 .unwrap();
5862 assert_eq!(
5863 result.verdict,
5864 Some(Verdict::Pass),
5865 "a passing verdict from a Layer 1 that reported its OWN success \
5866 must still be transplanted (18e); a None here would mean the \
5867 D-15 fix is indiscriminate"
5868 );
5869 assert_eq!(result.status, AgentStatus::Success);
5870 assert_eq!(result.decided_by_layer, Some(0));
5871 }
5872
5873 /// NC-6: with Layer 0 disabled, the same self-contradictory marker never
5874 /// gets laundered at all — Layer 1 reports `Failed` verbatim and
5875 /// `decide_action` routes it to `GateReview`.
5876 ///
5877 /// What the control proves: the GRAFT is the mechanism, not the classifier
5878 /// and not `decide_action`. Removing Layer 0 removes the laundering
5879 /// entirely, so the exploit's precondition is an affirmative Layer-0 probe
5880 /// success — which is exactly why plan 34-04 (999.76), by making
5881 /// `decided_by_layer == Some(0)` common in worktree mode, must not land
5882 /// without the fix this test pins.
5883 ///
5884 /// The routing consequence is asserted here rather than assumed, so a
5885 /// future change to `decide_action`'s `Failed` arm breaks this test rather
5886 /// than silently invalidating the control.
5887 #[test]
5888 fn layer0_disabled_routes_a_self_reported_failure_to_gate_review() {
5889 let dir = tempfile::tempdir().unwrap();
5890 let phase_dir = dir.path().join(".planning/phases/16-reliability");
5891 std::fs::create_dir_all(&phase_dir).unwrap();
5892 // The difference from the fixtures above: Layer 0 is switched off, so
5893 // the cascade falls through to Layer 1 instead of short-circuiting on
5894 // an affirmative probe success.
5895 std::fs::write(
5896 dir.path().join("devflow.toml"),
5897 "external_verify_enabled = false\n",
5898 )
5899 .unwrap();
5900 // Belt AND braces, deliberately. `config::external_verify_enabled`
5901 // consults `DEVFLOW_EXTERNAL_VERIFY_ENABLED` BEFORE `devflow.toml`, and
5902 // `config::tests::env_overrides_file_external_verification` sets that
5903 // variable to "true" process-globally under a mutex private to its own
5904 // module — which cannot serialize against this one. A PLAN declaring
5905 // `external_verify` would therefore let a parallel run of that test
5906 // re-enable Layer 0 here and flake this control into a green.
5907 // Declaring no probe closes that window: with no declared commands and
5908 // no approval vector, `evaluate_layer0` abstains whatever the env says,
5909 // so this test is deterministic under every value of the variable.
5910 std::fs::write(phase_dir.join("16-01-PLAN.md"), "---\nplan: 01\n---\n").unwrap();
5911 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5912 let mut state = state_in(dir.path(), 16);
5913 state.stage = Stage::Validate;
5914
5915 std::fs::write(
5916 stdout_path(dir.path(), 16),
5917 "DEVFLOW_RESULT: {\"status\":\"failed\",\"verdict\":\"pass\"}\n",
5918 )
5919 .unwrap();
5920 // No approval vector — Layer 0 is disabled, so there is nothing to
5921 // approve, and supplying one would re-arm the very arm being removed.
5922 let result =
5923 evaluate_agent_result_inner(dir.path(), &state, &GitFlowConfig::default(), None)
5924 .unwrap();
5925
5926 assert_eq!(
5927 result.status,
5928 AgentStatus::Failed,
5929 "with Layer 0 disabled, Layer 1's self-reported failure stands \
5930 verbatim — there is no affirmative probe success to graft onto"
5931 );
5932 assert_eq!(result.decided_by_layer, Some(1));
5933 assert_eq!(
5934 crate::outcome_policy::decide_action(Stage::Validate, result.status),
5935 crate::outcome_policy::Action::GateReview,
5936 "a self-reported failure must gate for review, never advance"
5937 );
5938 }
5939
5940 /// 18e's reconciliation is scoped to `Stage::Validate` only (flagged
5941 /// assumption in 18-05-PLAN.md): at every other stage an affirmative
5942 /// Layer 0 success must keep `verdict: None`, even when Layer 1's marker
5943 /// carries an explicit verdict.
5944 #[test]
5945 fn layer0_affirmative_success_keeps_none_verdict_off_validate() {
5946 let dir = tempfile::tempdir().unwrap();
5947 let phase_dir = dir.path().join(".planning/phases/16-reliability");
5948 std::fs::create_dir_all(&phase_dir).unwrap();
5949 std::fs::write(
5950 phase_dir.join("16-01-PLAN.md"),
5951 "---\nexternal_verify: \"test -f shipped\"\n---\n",
5952 )
5953 .unwrap();
5954 std::fs::write(dir.path().join("shipped"), "done").unwrap();
5955 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
5956 std::fs::write(
5957 stdout_path(dir.path(), 16),
5958 "DEVFLOW_RESULT: {\"status\":\"success\",\"verdict\":\"pass\"}\n",
5959 )
5960 .unwrap();
5961 let state = state_in(dir.path(), 16); // Stage::Code by default
5962 let approval = vec!["test -f shipped".to_string()];
5963
5964 let result = evaluate_agent_result_inner(
5965 dir.path(),
5966 &state,
5967 &GitFlowConfig::default(),
5968 Some(&approval),
5969 )
5970 .unwrap();
5971
5972 assert_eq!(result.status, AgentStatus::Success);
5973 assert_eq!(result.decided_by_layer, Some(0));
5974 assert_eq!(result.verdict, None);
5975 }
5976
5977 /// Ordering edge (17a): with multiple declared probes, ALL must pass for
5978 /// affirmative Success — the first failing probe vetoes the outcome
5979 /// regardless of which position it occupies among the declarations.
5980 #[test]
5981 fn multiple_declared_probes_first_failure_vetoes_regardless_of_order() {
5982 let dir = tempfile::tempdir().unwrap();
5983 let phase_dir = dir.path().join(".planning/phases/16-reliability");
5984 std::fs::create_dir_all(&phase_dir).unwrap();
5985 // 16-01 comes first alphabetically and passes; 16-02 comes second and fails.
5986 std::fs::write(
5987 phase_dir.join("16-01-PLAN.md"),
5988 "---\nexternal_verify: \"test -f passing-artifact\"\n---\n",
5989 )
5990 .unwrap();
5991 std::fs::write(
5992 phase_dir.join("16-02-PLAN.md"),
5993 "---\nexternal_verify: \"test -f never-created\"\n---\n",
5994 )
5995 .unwrap();
5996 std::fs::write(dir.path().join("passing-artifact"), "done").unwrap();
5997 let mut state = state_in(dir.path(), 16);
5998 state.stage = Stage::Define;
5999
6000 let approval = vec![
6001 "test -f passing-artifact".to_string(),
6002 "test -f never-created".to_string(),
6003 ];
6004 let result_a = evaluate_agent_result_inner(
6005 dir.path(),
6006 &state,
6007 &GitFlowConfig::default(),
6008 Some(&approval),
6009 )
6010 .unwrap();
6011 assert_eq!(result_a.status, AgentStatus::Failed);
6012 assert!(
6013 result_a
6014 .reason
6015 .as_deref()
6016 .is_some_and(|reason| reason.contains("never-created")),
6017 "unexpected reason: {:?}",
6018 result_a.reason
6019 );
6020
6021 // Swap which position fails: 16-01 now fails, 16-02 passes. The
6022 // overall outcome must still veto — order of declaration must not
6023 // matter.
6024 std::fs::write(
6025 phase_dir.join("16-01-PLAN.md"),
6026 "---\nexternal_verify: \"test -f still-missing\"\n---\n",
6027 )
6028 .unwrap();
6029 std::fs::write(
6030 phase_dir.join("16-02-PLAN.md"),
6031 "---\nexternal_verify: \"test -f passing-artifact\"\n---\n",
6032 )
6033 .unwrap();
6034 let approval_swapped = vec![
6035 "test -f still-missing".to_string(),
6036 "test -f passing-artifact".to_string(),
6037 ];
6038 let result_b = evaluate_agent_result_inner(
6039 dir.path(),
6040 &state,
6041 &GitFlowConfig::default(),
6042 Some(&approval_swapped),
6043 )
6044 .unwrap();
6045 assert_eq!(result_b.status, AgentStatus::Failed);
6046
6047 // Now make BOTH pass: only then is the outcome Success.
6048 std::fs::write(dir.path().join("still-missing"), "done").unwrap();
6049 let result_c = evaluate_agent_result_inner(
6050 dir.path(),
6051 &state,
6052 &GitFlowConfig::default(),
6053 Some(&approval_swapped),
6054 )
6055 .unwrap();
6056 assert_eq!(result_c.status, AgentStatus::Success);
6057 assert_eq!(result_c.decided_by_layer, Some(0));
6058 }
6059
6060 #[test]
6061 fn archive_moves_captures_into_history_and_removes_pid_file() {
6062 // 16b: prior-stage captures must survive a simulated next-launch by
6063 // appearing under .devflow/history/phase-NN/, not be wiped outright.
6064 let dir = tempfile::tempdir().unwrap();
6065 let root = dir.path();
6066 std::fs::create_dir_all(root.join(".devflow")).unwrap();
6067 std::fs::write(root.join(".devflow/phase-01-stdout"), "prior stdout").unwrap();
6068 std::fs::write(root.join(".devflow/phase-01-exit"), "0").unwrap();
6069 std::fs::write(root.join(".devflow/phase-01-agent-pid"), "1234").unwrap();
6070
6071 archive_phase_files(root, root, 1, 5).unwrap();
6072
6073 // The live capture paths are gone (moved, not merely deleted).
6074 assert!(!root.join(".devflow/phase-01-stdout").exists());
6075 assert!(!root.join(".devflow/phase-01-exit").exists());
6076 // Agent-pid is bookkeeping, not diagnostic — still removed outright.
6077 assert!(!root.join(".devflow/phase-01-agent-pid").exists());
6078
6079 let history = history_dir(root, 1);
6080 let archived: Vec<_> = std::fs::read_dir(&history)
6081 .unwrap()
6082 .flatten()
6083 .map(|e| e.file_name().to_string_lossy().into_owned())
6084 .collect();
6085 let archived_stdout = archived
6086 .iter()
6087 .find(|name| name.ends_with("-stdout"))
6088 .expect("stdout capture should be archived into history");
6089 assert!(archived.iter().any(|name| name.ends_with("-exit")));
6090 let contents = std::fs::read_to_string(history.join(archived_stdout)).unwrap();
6091 assert_eq!(contents, "prior stdout");
6092 }
6093
6094 #[test]
6095 fn archive_is_noop_when_nothing_to_archive() {
6096 let dir = tempfile::tempdir().unwrap();
6097 let root = dir.path();
6098 // Should not panic when there is nothing to archive (first launch).
6099 archive_phase_files(root, root, 1, 5).unwrap();
6100 assert!(!history_dir(root, 1).exists());
6101 }
6102
6103 #[test]
6104 fn archive_handles_missing_devflow_dir() {
6105 let dir = tempfile::tempdir().unwrap();
6106 let root = dir.path();
6107 // No .devflow dir at all — should not panic.
6108 archive_phase_files(root, root, 1, 5).unwrap();
6109 }
6110
6111 #[test]
6112 fn archive_failure_preserves_live_capture_for_retry() {
6113 let dir = tempfile::tempdir().unwrap();
6114 let root = dir.path();
6115 std::fs::create_dir_all(root.join(".devflow")).unwrap();
6116 std::fs::write(stdout_path(root, 1), "evidence").unwrap();
6117 // A file where the history directory must be forces create_dir_all
6118 // to fail before the live capture is moved or a monitor can truncate it.
6119 std::fs::write(root.join(".devflow/history"), "blocked").unwrap();
6120
6121 assert!(archive_phase_files(root, root, 1, 5).is_err());
6122 assert_eq!(
6123 std::fs::read_to_string(stdout_path(root, 1)).unwrap(),
6124 "evidence"
6125 );
6126 }
6127
6128 #[test]
6129 fn archive_second_publish_failure_rolls_back_complete_live_pair() {
6130 let dir = tempfile::tempdir().unwrap();
6131 let root = dir.path();
6132 std::fs::create_dir_all(root.join(".devflow")).unwrap();
6133 std::fs::write(stdout_path(root, 1), "stdout evidence").unwrap();
6134 std::fs::write(exit_code_path(root, 1), "17").unwrap();
6135 let history = history_dir(root, 1);
6136 std::fs::create_dir_all(history.join("fixed-exit/blocker")).unwrap();
6137
6138 assert!(archive_phase_files_with_stamp(root, root, 1, 5, "fixed").is_err());
6139
6140 assert_eq!(
6141 std::fs::read_to_string(stdout_path(root, 1)).unwrap(),
6142 "stdout evidence"
6143 );
6144 assert_eq!(
6145 std::fs::read_to_string(exit_code_path(root, 1)).unwrap(),
6146 "17"
6147 );
6148 assert!(!history.join("fixed-stdout").exists());
6149 assert!(!history.join(".pending-fixed").exists());
6150 }
6151
6152 #[test]
6153 fn archive_review_copy_failure_rolls_back_complete_live_pair() {
6154 let dir = tempfile::tempdir().unwrap();
6155 let root = dir.path();
6156 let evidence_root = root.join("phase-worktree");
6157 std::fs::create_dir_all(root.join(".devflow")).unwrap();
6158 std::fs::write(stdout_path(root, 1), "stdout evidence").unwrap();
6159 std::fs::write(exit_code_path(root, 1), "23").unwrap();
6160 let review = evidence_root.join(".planning/phases/01-example/01-REVIEW.md");
6161 std::fs::create_dir_all(&review).unwrap();
6162
6163 assert!(archive_phase_files_with_stamp(root, &evidence_root, 1, 5, "review-copy").is_err());
6164
6165 assert_eq!(
6166 std::fs::read_to_string(stdout_path(root, 1)).unwrap(),
6167 "stdout evidence"
6168 );
6169 assert_eq!(
6170 std::fs::read_to_string(exit_code_path(root, 1)).unwrap(),
6171 "23"
6172 );
6173 let history = history_dir(root, 1);
6174 assert!(!history.join("review-copy-stdout").exists());
6175 assert!(!history.join("review-copy-exit").exists());
6176 assert!(!history.join(".pending-review-copy").exists());
6177 }
6178
6179 #[test]
6180 fn archive_snapshots_current_review_into_same_generation() {
6181 let dir = tempfile::tempdir().unwrap();
6182 let root = dir.path();
6183 let evidence_root = root.join("phase-worktree");
6184 std::fs::create_dir_all(root.join(".devflow")).unwrap();
6185 std::fs::write(stdout_path(root, 1), "attempt").unwrap();
6186 let phase_dir = evidence_root.join(".planning/phases/01-example");
6187 std::fs::create_dir_all(&phase_dir).unwrap();
6188 std::fs::write(phase_dir.join("01-REVIEW.md"), "review one").unwrap();
6189
6190 let stamp = archive_phase_files(root, &evidence_root, 1, 5)
6191 .unwrap()
6192 .unwrap();
6193
6194 assert_eq!(
6195 std::fs::read_to_string(history_dir(root, 1).join(format!("{stamp}-REVIEW.md")))
6196 .unwrap(),
6197 "review one"
6198 );
6199 }
6200
6201 #[test]
6202 fn archive_prunes_history_to_retain_count() {
6203 let dir = tempfile::tempdir().unwrap();
6204 let root = dir.path();
6205 std::fs::create_dir_all(root.join(".devflow")).unwrap();
6206
6207 for i in 0..7 {
6208 std::fs::write(root.join(".devflow/phase-01-stdout"), format!("gen {i}")).unwrap();
6209 std::fs::write(root.join(".devflow/phase-01-exit"), "0").unwrap();
6210 archive_phase_files(root, root, 1, 3).unwrap();
6211 }
6212
6213 let history = history_dir(root, 1);
6214 let stdout_count = std::fs::read_dir(&history)
6215 .unwrap()
6216 .flatten()
6217 .filter(|e| e.file_name().to_string_lossy().ends_with("-stdout"))
6218 .count();
6219 let exit_count = std::fs::read_dir(&history)
6220 .unwrap()
6221 .flatten()
6222 .filter(|e| e.file_name().to_string_lossy().ends_with("-exit"))
6223 .count();
6224 assert_eq!(stdout_count, 3, "expected at most 3 retained generations");
6225 assert_eq!(exit_count, 3, "expected at most 3 retained generations");
6226 }
6227
6228 /// The set of stamp groups currently surviving in a history directory,
6229 /// derived the same way `prune_history` derives them (`rsplit_once('-')`,
6230 /// keep the left part) so the assertion measures grouping rather than a
6231 /// listing length.
6232 fn surviving_stamps(history: &Path) -> std::collections::BTreeSet<String> {
6233 std::fs::read_dir(history)
6234 .unwrap()
6235 .flatten()
6236 .filter_map(|entry| {
6237 let name = entry.file_name().to_str()?.to_string();
6238 name.rsplit_once('-')
6239 .map(|(stamp, _suffix)| stamp.to_string())
6240 })
6241 .collect()
6242 }
6243
6244 /// ROADMAP criterion 7's retention half. `DEFAULT_CAPTURE_RETENTION` was
6245 /// `5`, and `archive_phase_files` runs once per launch: a clean five-stage
6246 /// Define→Plan→Code→Validate→Ship run produces 4 archive events and each
6247 /// Validate→Code loop-back adds 2. At `5`, the first loop-back's sixth
6248 /// event evicted Define's capture — silently, with no error and no log.
6249 ///
6250 /// What a regression here costs: a stage capture destroyed before the
6251 /// phase that requested it has read it, which is unrecoverable after the
6252 /// fact because `.devflow/` is the only copy until it is deliberately
6253 /// copied out.
6254 #[test]
6255 fn prune_history_retains_a_full_five_stage_run_with_loop_backs() {
6256 let dir = tempfile::tempdir().unwrap();
6257 let root = dir.path();
6258 let history = history_dir(root, 1);
6259 std::fs::create_dir_all(&history).unwrap();
6260
6261 let retain = crate::config::DEFAULT_CAPTURE_RETENTION;
6262
6263 // Twelve generations, strictly increasing. The suffix is load-bearing:
6264 // `prune_history` derives a stamp with `rsplit_once('-')` and keeps the
6265 // LEFT part, so a bare `{nanos}-{seq}` name would yield the stamp
6266 // `{nanos}` and then delete `{nanos}-stdout`, which never exists — the
6267 // retain half would false-pass via the `stamps.len() <= retain` early
6268 // return while the evict half could never pass at all.
6269 let stamps: Vec<String> = (0..12)
6270 .map(|i| format!("{}-0", 1_700_000_000_000_000_000u128 + i))
6271 .collect();
6272 for stamp in &stamps {
6273 std::fs::write(history.join(format!("{stamp}-stdout")), "capture").unwrap();
6274 }
6275 // The oldest generation gets a second suffix so eviction-by-stamp-group
6276 // is actually exercised rather than assumed: one evicted stamp must
6277 // take BOTH its files.
6278 std::fs::write(history.join(format!("{}-exit", stamps[0])), "0").unwrap();
6279
6280 prune_history(&history, retain);
6281
6282 let survivors = surviving_stamps(&history);
6283 assert_eq!(
6284 survivors.len(),
6285 12,
6286 "a five-stage run with loop-backs must not lose a capture at the default \
6287 retention; found {survivors:?}"
6288 );
6289 for stamp in &stamps {
6290 assert!(
6291 survivors.contains(stamp),
6292 "generation {stamp} was evicted at exactly the retention boundary"
6293 );
6294 }
6295
6296 // Opposite-result control. Without this half the test would be
6297 // measuring a directory listing, not pruning: `prune_history` returns
6298 // early whenever `stamps.len() <= retain`, so a fixture that never
6299 // crosses the boundary passes identically against a `prune_history`
6300 // that does nothing at all.
6301 let thirteenth = format!("{}-0", 1_700_000_000_000_000_000u128 + 12);
6302 std::fs::write(history.join(format!("{thirteenth}-stdout")), "capture").unwrap();
6303
6304 prune_history(&history, retain);
6305
6306 let after = surviving_stamps(&history);
6307 assert_eq!(
6308 after.len(),
6309 12,
6310 "crossing the boundary by one must evict exactly one stamp group, not zero \
6311 and not several; found {after:?}"
6312 );
6313 assert!(
6314 !after.contains(&stamps[0]),
6315 "the evicted generation must be the OLDEST by stamp order"
6316 );
6317 assert!(
6318 !history.join(format!("{}-exit", stamps[0])).exists(),
6319 "eviction operates on the stamp GROUP: the oldest generation's -exit file must \
6320 go with its -stdout, or pruning is leaking partial generations"
6321 );
6322 assert!(
6323 after.contains(&thirteenth),
6324 "the newest generation must survive its own arrival"
6325 );
6326 }
6327
6328 #[test]
6329 fn evaluate_agent_result_reads_files_end_to_end() {
6330 let dir = tempfile::tempdir().unwrap();
6331 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6332 std::fs::write(
6333 stdout_path(dir.path(), 6),
6334 "done\nDEVFLOW_RESULT: {\"status\":\"success\",\"commits\":2,\"summary\":\"ok\"}\n",
6335 )
6336 .unwrap();
6337 std::fs::write(exit_code_path(dir.path(), 6), "0").unwrap();
6338 let state = state_in(dir.path(), 6);
6339
6340 let result = evaluate_agent_result(dir.path(), &state, &GitFlowConfig::default()).unwrap();
6341
6342 assert_eq!(result.status, AgentStatus::Success);
6343 assert_eq!(result.commits, Some(2));
6344 assert_eq!(result.summary.as_deref(), Some("ok"));
6345 }
6346
6347 // ---- exit-code arbitration on a claimed success (31-04, T-31-15) -----
6348 //
6349 // Every test below drives the FULL cascade through
6350 // `evaluate_agent_result_inner`, never the parser's own return value.
6351 // 31-RESEARCH.md § Pitfall 4 records why: a truncation-boundary test that
6352 // checks only `parse_claude_event_result` exercises constraint 9's items 1
6353 // and 2, which the `a557805` root-cause refactor already closed. The
6354 // residual this arbitration exists for lives in the WIRING — Layer 1
6355 // returning before Layer 2 is ever consulted — and only the cascade
6356 // exercises it.
6357
6358 /// A success marker that also claims `verdict: pass` — the shape a naive
6359 /// "carry every other field over" downgrade would have preserved. Used to
6360 /// prove `verdict` is dropped.
6361 ///
6362 /// Correction (34-01, D-15): an earlier version of this comment asserted
6363 /// that keeping the field would classify Validate as Passed because
6364 /// `classify_validate_outcome` matches `Some(Verdict::Pass)` first with the
6365 /// status discarded. That overstated the reachability — `decide_action`
6366 /// intercepts a non-`Success` status before the classifier runs. The
6367 /// corrected record of how the inversion is actually reached lives on
6368 /// [`super::reconcile_layer0_verdict`].
6369 const MARKER_SUCCESS_CLAIMING_PASS: &str =
6370 r#"Done.\nDEVFLOW_RESULT: {\"status\":\"success\",\"verdict\":\"pass\"}"#;
6371
6372 /// The residual of constraint 9 that no parser assertion can reach.
6373 ///
6374 /// A capture cut at an exact line boundary is byte-identical to a healthy
6375 /// shorter run, so the stream itself carries no evidence of the tear. The
6376 /// writer that died between flushing turn N and turn N+1 also died
6377 /// non-zero, and that exit code is the only signal left.
6378 #[test]
6379 fn stream_success_cannot_stand_against_nonzero_exit_code() {
6380 let dir = tempfile::tempdir().unwrap();
6381 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6382 std::fs::write(
6383 stdout_path(dir.path(), 31),
6384 v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS_CLAIMING_PASS),
6385 )
6386 .unwrap();
6387
6388 // NEGATIVE CONTROL, encoded in the test rather than described in prose:
6389 // Layer 1 on its own decides Success here AND reports `verdict: Pass`.
6390 // Without this the assertions below cannot distinguish "the arbitration
6391 // downgraded a success" from "nothing ever claimed success", nor
6392 // "`verdict` was dropped" from "`verdict` was never set".
6393 let layer1 = evaluate_layer1(dir.path(), 31).unwrap();
6394 assert_eq!(layer1.status, AgentStatus::Success);
6395 assert_eq!(layer1.verdict, Some(Verdict::Pass));
6396
6397 std::fs::write(exit_code_path(dir.path(), 31), "1\n").unwrap();
6398 let state = state_in(dir.path(), 31);
6399
6400 let result =
6401 evaluate_agent_result_inner(dir.path(), &state, &GitFlowConfig::default(), None)
6402 .unwrap();
6403
6404 assert_eq!(result.status, AgentStatus::Failed);
6405 assert_eq!(result.exit_code, Some(1));
6406 assert!(
6407 result.reason.as_deref().is_some_and(|r| r.contains("1")),
6408 "the reason must name the exit code: {:?}",
6409 result.reason
6410 );
6411 // Layer 1 still decided this — the arbitration corrects its verdict, it
6412 // does not hand the decision to Layer 2.
6413 assert_eq!(result.decided_by_layer, Some(1));
6414 // Load-bearing: a downgraded result has no verdict to offer. Carrying
6415 // `Some(Verdict::Pass)` over would leave Validate classified Passed and
6416 // make this whole test's premise false at the stage that matters most
6417 // (999.74 / DEN-95).
6418 assert_eq!(result.verdict, None);
6419 }
6420
6421 /// The matched negative control for the test above. Without it, that test
6422 /// cannot tell "the arbitration works" from "the arbitration fires on
6423 /// everything".
6424 #[test]
6425 fn stream_success_stands_when_the_exit_code_is_zero() {
6426 let dir = tempfile::tempdir().unwrap();
6427 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6428 std::fs::write(
6429 stdout_path(dir.path(), 32),
6430 v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS_CLAIMING_PASS),
6431 )
6432 .unwrap();
6433 std::fs::write(exit_code_path(dir.path(), 32), "0\n").unwrap();
6434 let state = state_in(dir.path(), 32);
6435
6436 let result =
6437 evaluate_agent_result_inner(dir.path(), &state, &GitFlowConfig::default(), None)
6438 .unwrap();
6439
6440 assert_eq!(result.status, AgentStatus::Success);
6441 assert_eq!(result.decided_by_layer, Some(1));
6442 // The verdict survives an untouched result — proof that the `None`
6443 // asserted in the downgrade test is the arbitration's doing and not a
6444 // property of the fixture.
6445 assert_eq!(result.verdict, Some(Verdict::Pass));
6446 }
6447
6448 /// A missing exit file is not evidence of failure. This matches
6449 /// `evaluate_layer2`'s own tolerance (`Err(_) => return Ok(None)`); a
6450 /// stricter reading here would fail every stage whose monitor had not yet
6451 /// written the file.
6452 #[test]
6453 fn stream_success_stands_when_no_exit_file_exists() {
6454 let dir = tempfile::tempdir().unwrap();
6455 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6456 std::fs::write(
6457 stdout_path(dir.path(), 33),
6458 v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS),
6459 )
6460 .unwrap();
6461 assert!(
6462 !exit_code_path(dir.path(), 33).exists(),
6463 "fixture precondition: there must be no exit file"
6464 );
6465 let state = state_in(dir.path(), 33);
6466
6467 let result =
6468 evaluate_agent_result_inner(dir.path(), &state, &GitFlowConfig::default(), None)
6469 .unwrap();
6470
6471 assert_eq!(result.status, AgentStatus::Success);
6472 assert_eq!(result.decided_by_layer, Some(1));
6473 }
6474
6475 /// Only a *claimed success* is arbitrated. Downgrading a rate limit to a
6476 /// generic failure would route the run to a human gate instead of the
6477 /// auto-resume cron it needs — the exact harm `rate_limited_result`'s
6478 /// precedence over `detect_claude_envelope_failure` exists to prevent.
6479 #[test]
6480 fn rate_limited_verdict_is_not_arbitrated_by_exit_code() {
6481 let dir = tempfile::tempdir().unwrap();
6482 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6483 std::fs::write(
6484 stdout_path(dir.path(), 34),
6485 r#"{"type":"result","subtype":"error_rate_limit","is_error":true,"retry_after":"2026-06-18T15:45:30Z"}"#,
6486 )
6487 .unwrap();
6488 std::fs::write(exit_code_path(dir.path(), 34), "1\n").unwrap();
6489 let state = state_in(dir.path(), 34);
6490
6491 let result =
6492 evaluate_agent_result_inner(dir.path(), &state, &GitFlowConfig::default(), None)
6493 .unwrap();
6494
6495 assert_eq!(result.status, AgentStatus::RateLimited);
6496 assert_eq!(
6497 result.reason.as_deref(),
6498 Some("rate limited until 2026-06-18T15:45:30Z"),
6499 "the rate-limit reason must survive verbatim — the resume cron reads it"
6500 );
6501 }
6502
6503 /// Plan 31-02's side-channel verdict survives arbitration unchanged. An
6504 /// `IdleTimeout` collapsed into `Failed` would lose exactly the distinction
6505 /// 31-02 exists to create, and the monitor writes a NON-zero exit for a
6506 /// child it killed, so this is not a hypothetical pairing.
6507 #[test]
6508 fn idle_timeout_verdict_is_not_arbitrated_by_exit_code() {
6509 let dir = tempfile::tempdir().unwrap();
6510 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6511 std::fs::write(
6512 stdout_path(dir.path(), 35),
6513 v3_stream_capture(MARKER_SUCCESS, MARKER_SUCCESS, MARKER_SUCCESS),
6514 )
6515 .unwrap();
6516 write_idle_timeout_record(
6517 dir.path(),
6518 35,
6519 &[("aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa", "feat: partial")],
6520 );
6521 std::fs::write(exit_code_path(dir.path(), 35), "143\n").unwrap();
6522 let state = state_in(dir.path(), 35);
6523
6524 let result =
6525 evaluate_agent_result_inner(dir.path(), &state, &GitFlowConfig::default(), None)
6526 .unwrap();
6527
6528 assert_eq!(result.status, AgentStatus::IdleTimeout);
6529 assert_eq!(
6530 result.exit_code, None,
6531 "the arbitration must not graft an exit code onto a timeout verdict"
6532 );
6533 }
6534
6535 /// Exit-code fidelity (adversarial review of 31-04, W1). A blanket `Failed`
6536 /// would flatten the two codes `evaluate_layer2` classifies specially, and
6537 /// `outcome_policy::decide_action` routes those to `GateInfra` rather than
6538 /// `GateReview`. The same exit code must not reach two different operator
6539 /// gates depending on whether a stale Layer 1 success happened to be there.
6540 #[test]
6541 fn arbitration_preserves_layer2s_resource_and_unavailable_codes() {
6542 for (code, expected) in [
6543 (137, AgentStatus::ResourceKilled),
6544 (127, AgentStatus::AgentUnavailable),
6545 (2, AgentStatus::Failed),
6546 ] {
6547 let dir = tempfile::tempdir().unwrap();
6548 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6549 std::fs::write(
6550 stdout_path(dir.path(), 36),
6551 v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS),
6552 )
6553 .unwrap();
6554 std::fs::write(exit_code_path(dir.path(), 36), format!("{code}\n")).unwrap();
6555 let state = state_in(dir.path(), 36);
6556
6557 let arbitrated =
6558 evaluate_agent_result_inner(dir.path(), &state, &GitFlowConfig::default(), None)
6559 .unwrap();
6560
6561 assert_eq!(
6562 arbitrated.status, expected,
6563 "exit {code} must arbitrate to {expected:?}, matching evaluate_layer2"
6564 );
6565 assert_eq!(arbitrated.exit_code, Some(code));
6566 }
6567 }
6568
6569 /// D-12's inverse assertion, and the mirror of
6570 /// [`single_doc_envelope_not_consumed_by_claude_stream_parser`].
6571 ///
6572 /// That test pins one direction: today's shipped `--output-format json`
6573 /// envelope must NOT be consumed by the stream parser. This pins the other:
6574 /// a capture produced by plan 31-01's new `stream-json` argv classifies as
6575 /// [`CaptureKind::ClaudeStream`] and is NOT consumed by the
6576 /// single-document envelope path. Without both directions, widening either
6577 /// gate is only half-guarded.
6578 ///
6579 /// Cites `classify()` / `CaptureKind::ClaudeStream` deliberately: the gate
6580 /// predicate `31-CONTEXT.md` and `30-VERIFICATION.md` W-02 still name is no
6581 /// longer a live function — the `a557805` refactor replaced it.
6582 #[test]
6583 fn stream_json_capture_is_not_consumed_by_the_single_document_path() {
6584 let capture = v3_stream_capture(NO_MARKER, NO_MARKER, MARKER_SUCCESS);
6585
6586 // The classifier owns it.
6587 assert!(capture_is_claude_stream(&capture));
6588
6589 // Every single-document reader declines it...
6590 assert!(claude_session_id(&capture).is_none());
6591 assert!(detect_claude_envelope_failure(&capture).is_none());
6592 assert!(detect_claude_rate_limit(&capture).is_none());
6593
6594 // ...and the stream parser still owns it, so declining costs no verdict.
6595 assert_eq!(
6596 parse_claude_event_result(&capture).unwrap().status,
6597 AgentStatus::Success
6598 );
6599
6600 // Non-vacuity: the single-document readers are not simply broken — the
6601 // same three answer a real envelope. Without this, the `is_none()`
6602 // assertions above would pass against a reader that returned `None` for
6603 // everything.
6604 let envelope = r#"{"type":"result","subtype":"error_rate_limit","is_error":true,"retry_after":"2026-06-18T15:45:30Z","session_id":"abc"}"#;
6605 assert_eq!(claude_session_id(envelope).as_deref(), Some("abc"));
6606 assert!(detect_claude_envelope_failure(envelope).is_some());
6607 assert!(detect_claude_rate_limit(envelope).is_some());
6608 assert!(!capture_is_claude_stream(envelope));
6609 }
6610
6611 #[test]
6612 fn evaluate_layer1_finds_devflow_result_in_file() {
6613 let dir = tempfile::tempdir().unwrap();
6614 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6615 std::fs::write(
6616 stdout_path(dir.path(), 3),
6617 "output\ndevflow_result: {\"status\":\"failed\",\"reason\":\"bad output\"}\n",
6618 )
6619 .unwrap();
6620
6621 let result = evaluate_layer1(dir.path(), 3).unwrap();
6622
6623 assert_eq!(result.status, AgentStatus::Failed);
6624 assert_eq!(result.reason.as_deref(), Some("bad output"));
6625 }
6626
6627 /// The case `consecutive_failures_reaches_ceiling_across_cycles`
6628 /// (`pipeline_outcomes.rs`) silently depends on: a repository with no
6629 /// `feature/phase-NN` branch at all must report 0, not error or panic.
6630 #[test]
6631 fn phase_commit_count_reports_zero_without_a_branch() {
6632 let dir = tempfile::tempdir().unwrap();
6633 git(dir.path(), &["init"]);
6634 git(dir.path(), &["config", "user.email", "devflow@example.com"]);
6635 git(dir.path(), &["config", "user.name", "DevFlow Tests"]);
6636 git(dir.path(), &["config", "commit.gpgsign", "false"]);
6637 git(dir.path(), &["config", "tag.gpgsign", "false"]);
6638 git(dir.path(), &["config", "core.hooksPath", "/dev/null"]);
6639 git(dir.path(), &["checkout", "-b", "develop"]);
6640 std::fs::write(dir.path().join("README.md"), "base\n").unwrap();
6641 git(dir.path(), &["add", "README.md"]);
6642 git(dir.path(), &["commit", "-m", "base"]);
6643
6644 let count = phase_commit_count(dir.path(), &GitFlowConfig::default(), 999);
6645
6646 assert_eq!(count, 0, "no feature/phase-99 branch exists in this repo");
6647 }
6648
6649 #[test]
6650 fn evaluate_layer2_falls_back_to_exit_code_and_commit_count() {
6651 let dir = tempfile::tempdir().unwrap();
6652 init_repo_with_feature_commit(dir.path(), 4);
6653 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6654 std::fs::write(exit_code_path(dir.path(), 4), "0").unwrap();
6655 let state = state_in(dir.path(), 4);
6656
6657 let result = evaluate_layer2(dir.path(), 4, &GitFlowConfig::default(), state.stage)
6658 .unwrap()
6659 .unwrap();
6660
6661 assert_eq!(result.status, AgentStatus::Success);
6662 assert_eq!(result.exit_code, Some(0));
6663 assert_eq!(result.commits, Some(1));
6664 assert!(result.reason.unwrap().contains("1 commits"));
6665 }
6666
6667 #[test]
6668 fn evaluate_layer2_exit_zero_no_commits_is_failed() {
6669 // exit=0 but the feature branch has 0 commits ahead of develop →
6670 // "no work done" failure (the Layer 2 middle branch).
6671 let dir = tempfile::tempdir().unwrap();
6672 init_repo_with_feature_no_commit(dir.path(), 4);
6673 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6674 std::fs::write(exit_code_path(dir.path(), 4), "0").unwrap();
6675 let state = state_in(dir.path(), 4);
6676
6677 let result = evaluate_layer2(dir.path(), 4, &GitFlowConfig::default(), state.stage)
6678 .unwrap()
6679 .unwrap();
6680
6681 assert_eq!(result.status, AgentStatus::Failed);
6682 assert_eq!(result.exit_code, Some(0));
6683 assert_eq!(result.commits, Some(0));
6684 assert!(result.reason.unwrap().contains("no commits"));
6685 }
6686
6687 #[test]
6688 fn evaluate_layer2_nonzero_exit_is_failed() {
6689 // Non-zero exit code → failure regardless of commit count.
6690 let dir = tempfile::tempdir().unwrap();
6691 init_repo_with_feature_commit(dir.path(), 4);
6692 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6693 std::fs::write(exit_code_path(dir.path(), 4), "1").unwrap();
6694 let state = state_in(dir.path(), 4);
6695
6696 let result = evaluate_layer2(dir.path(), 4, &GitFlowConfig::default(), state.stage)
6697 .unwrap()
6698 .unwrap();
6699
6700 assert_eq!(result.status, AgentStatus::Failed);
6701 assert_eq!(result.exit_code, Some(1));
6702 assert!(result.reason.unwrap().contains("exited with code 1"));
6703 }
6704
6705 #[test]
6706 fn layer2_nonzero_exit_is_failed_all_stages() {
6707 // Non-zero exit is Failed regardless of stage — including Define and
6708 // Validate, which are exempt from the zero-commit gate but NOT from
6709 // the exit-code check.
6710 let dir = tempfile::tempdir().unwrap();
6711 init_repo_with_feature_no_commit(dir.path(), 10);
6712 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6713 std::fs::write(exit_code_path(dir.path(), 10), "1").unwrap();
6714
6715 for stage in [
6716 Stage::Define,
6717 Stage::Plan,
6718 Stage::Code,
6719 Stage::Validate,
6720 Stage::Ship,
6721 ] {
6722 let result = evaluate_layer2(dir.path(), 10, &GitFlowConfig::default(), stage)
6723 .unwrap()
6724 .unwrap();
6725 assert_eq!(
6726 result.status,
6727 AgentStatus::Failed,
6728 "stage {stage:?} should be Failed on nonzero exit"
6729 );
6730 }
6731 }
6732
6733 #[test]
6734 fn layer2_skips_commit_gate_for_define_and_validate() {
6735 let dir = tempfile::tempdir().unwrap();
6736 init_repo_with_feature_no_commit(dir.path(), 11);
6737 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6738 std::fs::write(exit_code_path(dir.path(), 11), "0").unwrap();
6739
6740 for stage in [Stage::Define, Stage::Validate] {
6741 let result = evaluate_layer2(dir.path(), 11, &GitFlowConfig::default(), stage)
6742 .unwrap()
6743 .unwrap();
6744 assert_ne!(
6745 result.status,
6746 AgentStatus::Failed,
6747 "stage {stage:?} should not be Failed for zero commits"
6748 );
6749 }
6750
6751 // Code stage with the same zero-commit inputs is still Failed
6752 // (existing behavior preserved).
6753 let result = evaluate_layer2(dir.path(), 11, &GitFlowConfig::default(), Stage::Code)
6754 .unwrap()
6755 .unwrap();
6756 assert_eq!(result.status, AgentStatus::Failed);
6757 }
6758
6759 #[test]
6760 fn evaluate_layer3_falls_back_to_commit_count() {
6761 let dir = tempfile::tempdir().unwrap();
6762 init_repo_with_feature_commit(dir.path(), 5);
6763
6764 let result = evaluate_layer3(dir.path(), 5, &GitFlowConfig::default()).unwrap();
6765
6766 assert_eq!(result.status, AgentStatus::Unknown);
6767 assert_eq!(result.exit_code, None);
6768 assert_eq!(result.commits, Some(1));
6769 assert!(result.reason.unwrap().contains("1 commits"));
6770 assert_eq!(result.decided_by_layer, Some(3));
6771 }
6772
6773 /// D-02/D-03 case 3 (17-03): "process gone, nothing accounted for" — zero
6774 /// commits and no declared external post-condition — is a fail-closed
6775 /// `Failed` outcome that flags human review, not a blanket advanceable
6776 /// `Unknown`. The commits-present case above stays `Unknown` (gated
6777 /// downstream by Plan 04's never-advance dispatch, D-04) — only the
6778 /// zero-commit sub-case is reclassified here.
6779 #[test]
6780 fn evaluate_layer3_zero_commits_is_failed_and_flags_human_review() {
6781 let dir = tempfile::tempdir().unwrap();
6782 init_repo_with_feature_no_commit(dir.path(), 5);
6783
6784 let result = evaluate_layer3(dir.path(), 5, &GitFlowConfig::default()).unwrap();
6785
6786 assert_eq!(result.status, AgentStatus::Failed);
6787 assert_eq!(result.exit_code, None);
6788 assert_eq!(result.commits, Some(0));
6789 assert_eq!(result.decided_by_layer, Some(3));
6790 let reason = result.reason.unwrap();
6791 assert!(reason.contains("no work"), "reason was: {reason}");
6792 assert!(
6793 reason.to_ascii_lowercase().contains("human review"),
6794 "reason was: {reason}"
6795 );
6796 }
6797
6798 #[test]
6799 fn parse_devflow_result_reads_verdict() {
6800 let stdout = r#"DEVFLOW_RESULT: {"status":"success","verdict":"gaps"}"#;
6801 let result = parse_devflow_result(stdout).unwrap();
6802 assert_eq!(result.status, AgentStatus::Success);
6803 assert_eq!(result.verdict, Some(Verdict::Gaps));
6804 }
6805
6806 #[test]
6807 fn parse_devflow_result_reads_verdict_pass() {
6808 let stdout = r#"DEVFLOW_RESULT: {"status":"success","verdict":"pass"}"#;
6809 let result = parse_devflow_result(stdout).unwrap();
6810 assert_eq!(result.status, AgentStatus::Success);
6811 assert_eq!(result.verdict, Some(Verdict::Pass));
6812 }
6813
6814 #[test]
6815 fn parse_devflow_result_verdict_absent_is_none() {
6816 let stdout = r#"DEVFLOW_RESULT: {"status":"success"}"#;
6817 let result = parse_devflow_result(stdout).unwrap();
6818 assert_eq!(result.status, AgentStatus::Success);
6819 assert_eq!(result.verdict, None);
6820 }
6821
6822 #[test]
6823 fn parse_devflow_result_malformed_verdict_is_none_not_parse_error() {
6824 // An unknown verdict string must not fail the whole marker parse —
6825 // status must still come through as Success with verdict None (T-13-14).
6826 let unknown = r#"DEVFLOW_RESULT: {"status":"success","verdict":"wat"}"#;
6827 let result = parse_devflow_result(unknown).unwrap();
6828 assert_eq!(result.status, AgentStatus::Success);
6829 assert_eq!(result.verdict, None);
6830
6831 // Mis-cased ("Pass" instead of "pass") must also be lenient, not an error.
6832 let miscased = r#"DEVFLOW_RESULT: {"status":"success","verdict":"Pass"}"#;
6833 let result = parse_devflow_result(miscased).unwrap();
6834 assert_eq!(result.status, AgentStatus::Success);
6835 assert_eq!(result.verdict, None);
6836 }
6837
6838 /// WR-09 (13-REVIEW.md): a `verdict` field present with a non-string
6839 /// JSON *type* (bool, number, object) must be just as lenient as a
6840 /// malformed string value — before the fix, deserializing straight to
6841 /// `Option<String>` errored out the entire `AgentResult` parse for a
6842 /// type mismatch, defeating the doc comment's "a malformed verdict must
6843 /// never silently drop a valid status" guarantee for this specific case.
6844 #[test]
6845 fn parse_devflow_result_non_string_verdict_type_is_none_not_parse_error() {
6846 let bool_verdict = r#"DEVFLOW_RESULT: {"status":"success","verdict":true}"#;
6847 let result = parse_devflow_result(bool_verdict).unwrap();
6848 assert_eq!(result.status, AgentStatus::Success);
6849 assert_eq!(result.verdict, None);
6850
6851 let numeric_verdict = r#"DEVFLOW_RESULT: {"status":"success","verdict":123}"#;
6852 let result = parse_devflow_result(numeric_verdict).unwrap();
6853 assert_eq!(result.status, AgentStatus::Success);
6854 assert_eq!(result.verdict, None);
6855
6856 let object_verdict = r#"DEVFLOW_RESULT: {"status":"success","verdict":{"x":1}}"#;
6857 let result = parse_devflow_result(object_verdict).unwrap();
6858 assert_eq!(result.status, AgentStatus::Success);
6859 assert_eq!(result.verdict, None);
6860 }
6861
6862 /// D-07 (17-01): the two new multi-word variants must serialize with
6863 /// their word boundary preserved — `#[serde(rename_all = "lowercase")]`
6864 /// alone would collapse `ResourceKilled` to `"resourcekilled"` (Pitfall 1).
6865 #[test]
6866 fn multi_word_variants_serialize_with_word_boundary() {
6867 assert_eq!(
6868 serde_json::to_string(&AgentStatus::ResourceKilled).unwrap(),
6869 "\"resource_killed\""
6870 );
6871 assert_eq!(
6872 serde_json::to_string(&AgentStatus::AgentUnavailable).unwrap(),
6873 "\"agent_unavailable\""
6874 );
6875 assert_eq!(
6876 serde_json::from_str::<AgentStatus>("\"resource_killed\"").unwrap(),
6877 AgentStatus::ResourceKilled
6878 );
6879 assert_eq!(
6880 serde_json::from_str::<AgentStatus>("\"agent_unavailable\"").unwrap(),
6881 AgentStatus::AgentUnavailable
6882 );
6883 }
6884
6885 /// Existing variants must keep their pre-existing lowercase wire form
6886 /// unchanged by the two new variants' additions.
6887 #[test]
6888 fn existing_variants_keep_wire_form() {
6889 assert_eq!(
6890 serde_json::to_string(&AgentStatus::Success).unwrap(),
6891 "\"success\""
6892 );
6893 assert_eq!(
6894 serde_json::to_string(&AgentStatus::Failed).unwrap(),
6895 "\"failed\""
6896 );
6897 assert_eq!(
6898 serde_json::to_string(&AgentStatus::RateLimited).unwrap(),
6899 "\"ratelimited\""
6900 );
6901 assert_eq!(
6902 serde_json::to_string(&AgentStatus::Unknown).unwrap(),
6903 "\"unknown\""
6904 );
6905 }
6906
6907 /// review consensus #1: `as_wire_str()` must never diverge from the serde
6908 /// form for ANY variant — pin it for all seven via a single round-trip
6909 /// assertion (quotes stripped).
6910 ///
6911 /// 31-02: `IdleTimeout` is enumerated here explicitly rather than left to
6912 /// the compiler. `as_wire_str`'s wildcard-free match makes a MISSING arm a
6913 /// compile error, but it cannot catch a WRONG one — an arm returning
6914 /// `"idletimeout"` compiles happily and diverges from the serde form the
6915 /// `#[serde(rename)]` produces. Only enumerating the variant here pins that.
6916 #[test]
6917 fn as_wire_str_matches_serde_form_for_every_variant() {
6918 for variant in [
6919 AgentStatus::Success,
6920 AgentStatus::Failed,
6921 AgentStatus::RateLimited,
6922 AgentStatus::Unknown,
6923 AgentStatus::ResourceKilled,
6924 AgentStatus::AgentUnavailable,
6925 AgentStatus::IdleTimeout,
6926 ] {
6927 let serde_form = serde_json::to_string(&variant).unwrap();
6928 let stripped = serde_form.trim_matches('"');
6929 assert_eq!(
6930 variant.as_wire_str(),
6931 stripped,
6932 "as_wire_str() diverged from serde form for {variant:?}"
6933 );
6934 }
6935 }
6936
6937 #[test]
6938 fn evaluate_layer2_exit_137_is_resource_killed() {
6939 let dir = tempfile::tempdir().unwrap();
6940 init_repo_with_feature_commit(dir.path(), 20);
6941 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6942 std::fs::write(exit_code_path(dir.path(), 20), "137").unwrap();
6943 let state = state_in(dir.path(), 20);
6944
6945 let result = evaluate_layer2(dir.path(), 20, &GitFlowConfig::default(), state.stage)
6946 .unwrap()
6947 .unwrap();
6948
6949 assert_eq!(result.status, AgentStatus::ResourceKilled);
6950 assert_eq!(result.exit_code, Some(137));
6951 }
6952
6953 #[test]
6954 fn evaluate_layer2_exit_127_is_agent_unavailable() {
6955 let dir = tempfile::tempdir().unwrap();
6956 init_repo_with_feature_commit(dir.path(), 21);
6957 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
6958 std::fs::write(exit_code_path(dir.path(), 21), "127").unwrap();
6959 let state = state_in(dir.path(), 21);
6960
6961 let result = evaluate_layer2(dir.path(), 21, &GitFlowConfig::default(), state.stage)
6962 .unwrap()
6963 .unwrap();
6964
6965 assert_eq!(result.status, AgentStatus::AgentUnavailable);
6966 assert_eq!(result.exit_code, Some(127));
6967 }
6968
6969 // -----------------------------------------------------------------
6970 // 27-03 (D-01/D-03): branch-exists + commit-count evidence resolves
6971 // the caller's own repository under a hostile GIT_DIR, not an
6972 // unrelated one.
6973 // -----------------------------------------------------------------
6974
6975 /// D-03/T-27-08: `evaluate_layer2`'s branch-exists and commit-count
6976 /// evidence (the two production sites at what were base-commit lines
6977 /// 574/583) resolves `project_root`'s own repository even when the
6978 /// process inherited a hostile `GIT_DIR` pointed at an unrelated
6979 /// repository — proven with a real spawned `git` process, not by
6980 /// inspecting a `Command` object alone. Mirrors
6981 /// `version::tests::tag_reads_resolve_caller_root_under_a_hostile_git_dir`
6982 /// (27-03) and `origin_main_ancestor_status_holds_under_a_hostile_git_dir`
6983 /// (`git.rs`, 27-01): the hostile `GIT_DIR` this test's own `<verify>`
6984 /// entries exercise (`GIT_DIR=<hostile>/.git cargo test ... this test`)
6985 /// is injected the same way any inherited-env attack reaches
6986 /// `evaluate_layer2` in production — via the whole process's
6987 /// environment, then down into the spawned child unless the
6988 /// constructor scrubs it.
6989 ///
6990 /// Deliberately tests the mirror direction from the plan's literal
6991 /// framing (real repo HAS the feature branch with a real commit;
6992 /// the standard hostile-`GIT_DIR` harness's throwaway repository does
6993 /// NOT), because the standard harness (`git init -q "$HOSTILE"`, no
6994 /// `feature/phase-NN` branch) cannot itself manufacture a false
6995 /// *positive* — an empty repository has no branch to spuriously
6996 /// report as present. It can, however, still prove the scrub's
6997 /// necessity by manufacturing a false *negative*: before this plan's
6998 /// migration, the two unmigrated `Command::new("git")` sites inherit
6999 /// the poisoned `GIT_DIR` and silently read the hostile repository
7000 /// instead of `project_root` — `rev-parse --verify` reports the real
7001 /// branch absent, the commit count is undercounted to zero, and a
7002 /// real agent's completed work is wrongly classified `Failed`. This
7003 /// is the same trust-boundary violation T-27-08 names (a foreign
7004 /// repository's state substituting for the real one), reached from
7005 /// the opposite direction; the scrub this plan adds removes `GIT_DIR`'s
7006 /// ability to redirect the spawned child at all, closing both
7007 /// directions identically.
7008 /// 27-REVIEW WR-01: this test previously set no hostile environment at
7009 /// all — it asserted ordinary-path behavior and claimed a hostile-
7010 /// `GIT_DIR` proof, so it passed identically with or without the scrub
7011 /// and could never have caught a regression back to a bare
7012 /// `Command::new("git")`. It now uses the spawned-child shape this
7013 /// phase established in `staleness.rs`
7014 /// (`embedded_commit_is_stale_resolves_execution_root_under_a_hostile_git_dir`):
7015 /// `GIT_DIR` is never set on this process (Rust 2024 `unsafe`, unsound
7016 /// under threaded tests — Phase 25 D-14), only on one freshly spawned
7017 /// child that re-invokes this same binary filtered to this one test.
7018 #[test]
7019 fn branch_evidence_resolves_caller_root_under_a_hostile_git_dir() {
7020 const INNER_ROOT: &str = "DEVFLOW_27_03_BRANCH_EVIDENCE_INNER_ROOT";
7021
7022 if let Ok(root) = std::env::var(INNER_ROOT) {
7023 // Inner mode: spawned by the outer half below with GIT_DIR
7024 // pointed at an unrelated foreign repository, scoped to this
7025 // child process only.
7026 let root = std::path::PathBuf::from(root);
7027 let phase = 27;
7028 let state = state_in(&root, phase);
7029
7030 let result = evaluate_layer2(&root, phase, &GitFlowConfig::default(), state.stage)
7031 .unwrap()
7032 .unwrap();
7033
7034 assert_eq!(
7035 result.status,
7036 AgentStatus::Success,
7037 "evaluate_layer2 must see project_root's own branch/commits, \
7038 not a hostile GIT_DIR's repository: {result:?}"
7039 );
7040 assert_eq!(result.commits, Some(1));
7041 return;
7042 }
7043
7044 // Outer mode: build the real repository (which HAS the feature
7045 // branch and its commit) plus a second, unrelated foreign
7046 // repository that has neither. Unscrubbed, the child would read the
7047 // foreign repo, find no branch, count zero commits, and misreport a
7048 // real agent's completed work as Failed.
7049 let dir = tempfile::tempdir().unwrap();
7050 let phase = 27;
7051 init_repo_with_feature_commit(dir.path(), phase);
7052 std::fs::create_dir_all(dir.path().join(".devflow")).unwrap();
7053 std::fs::write(exit_code_path(dir.path(), phase), "0").unwrap();
7054
7055 let foreign = tempfile::tempdir().unwrap();
7056 git(foreign.path(), &["init", "-q"]);
7057
7058 let exe = std::env::current_exe().expect("current_exe for child re-invocation");
7059 let out = std::process::Command::new(&exe)
7060 // Substring filter, NOT `--exact`: the binary's real test name is
7061 // module-qualified (`agent_result::tests::branch_evidence_...`),
7062 // so `--exact` against the bare name matches nothing, runs zero
7063 // tests, and still exits 0 — a false green.
7064 .arg("branch_evidence_resolves_caller_root_under_a_hostile_git_dir")
7065 .arg("--test-threads=1")
7066 .env(INNER_ROOT, dir.path().to_str().unwrap())
7067 .env("GIT_DIR", foreign.path().join(".git"))
7068 .output()
7069 .expect("spawn hostile child test process");
7070
7071 let stdout = String::from_utf8_lossy(&out.stdout);
7072 // Assert the child actually RAN the test, not merely that it exited
7073 // 0. A filter matching nothing exits 0 with "0 passed".
7074 assert!(
7075 stdout.contains("1 passed"),
7076 "child test process must have run exactly the inner test; \
7077 stdout:\n{stdout}"
7078 );
7079 assert!(
7080 out.status.success(),
7081 "child test process (hostile GIT_DIR pointed at an unrelated \
7082 foreign repository) must still resolve project_root's own \
7083 branch and commits; child exit status {:?}\nstdout:\n{stdout}",
7084 out.status
7085 );
7086 }
7087
7088 /// D-01 (33-CONTEXT.md): `phase_verification_exists` is the sole signal
7089 /// a Validate→Code loop-back consults to tell a mid-arc phase apart from
7090 /// a genuinely gap-flagged one. Covers all three states: no
7091 /// `.planning/phases` directory at all, a phase directory with no
7092 /// verification artifact, and a phase directory that has one — mirroring
7093 /// `phase_review_path`'s directory-prefix-scan idiom.
7094 #[test]
7095 fn phase_verification_exists_finds_the_artifact_by_prefix() {
7096 let dir = tempfile::tempdir().unwrap();
7097 let root = dir.path();
7098
7099 assert!(
7100 !phase_verification_exists(root, 82),
7101 "no .planning/phases directory at all must return false, not panic"
7102 );
7103
7104 let phase_dir = root.join(".planning/phases/82-loop-back-fix");
7105 std::fs::create_dir_all(&phase_dir).unwrap();
7106 assert!(
7107 !phase_verification_exists(root, 82),
7108 "a phase directory with no {{N}}-VERIFICATION.md must return false"
7109 );
7110
7111 std::fs::write(phase_dir.join("82-VERIFICATION.md"), "verified\n").unwrap();
7112 assert!(
7113 phase_verification_exists(root, 82),
7114 "a phase directory holding {{N}}-VERIFICATION.md must return true"
7115 );
7116 }
7117}