magi/config.rs
1//! Run configuration: the agent roster, the shape of the graph, and the
2//! blindness / verification policy.
3//!
4//! Layers, lowest precedence first: `<config_dir>/magi/config.toml` (machine),
5//! `.magi/config.toml`, `magi.toml`. With none, built-in defaults apply, the
6//! agent roster derived from the agent CLIs actually installed on this
7//! machine. `--config <path>` replaces the whole stack with that one file.
8//!
9//! **The repository layers are read from `<remote>/<base>`, never from the
10//! working tree.** A primary checkout is often stale or on a detached HEAD, so
11//! a gate read from it can be older than the one the branch under review is
12//! cut from (run ce6e failed on a gate step `origin/main` had already
13//! dropped). `git show <sha>:magi.toml` of the remote-tracking ref's commit is
14//! written to a content-addressed directory under the git common dir
15//! (`<common>/magi/config/<sha>-<hash of repo>/`, temp dir when that is not
16//! writable) and loaded from there.
17//!
18//! `remote` and `base` are themselves in the file, so they are resolved in a
19//! fixed order, first hit wins:
20//!
21//! 1. `[merge] base` / `remote` of the machine layer, rendered as a real load
22//! would (env templates, includes);
23//! 2. the same keys, read minimally (only `[merge]`, nothing deserialized into
24//! `Config`) from the checkout's `magi.toml`, then `.magi/config.toml` -
25//! only to *locate* the ref; an unreadable file is ignored;
26//! 3. `remote` defaults to `origin`, `base` to what
27//! `refs/remotes/<remote>/HEAD` points at.
28//!
29//! [`Config::discover`] is synchronous and reads the remote-tracking ref as it
30//! is (`daemon::fetch_loop` keeps it fresh); [`Config::discover_fetched`] runs
31//! `git fetch` first and is what every path that starts work uses. A failed
32//! fetch, a missing ref or an unresolvable base is an **error**, never a
33//! silent fall back to the working tree or a local branch. Two cases are not
34//! that: an explicit `--config`, and a `repo` that is not a git working tree
35//! at all (no branch exists, so nothing can be stale).
36//!
37//! A repository layer that declares `include = [...]` gets the whole tree of
38//! that commit extracted, so relative includes resolve against the same
39//! commit. The directory is built beside its name and renamed in, so a
40//! concurrent `discover` never reads a half-written layer.
41use std::collections::BTreeMap;
42use std::path::{Path, PathBuf};
43
44use anyhow::{Context as _, Result, bail};
45use serde::{Deserialize, Serialize};
46
47/// Which CLI drives an agent.
48#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize, Serialize)]
49#[serde(rename_all = "lowercase")]
50pub enum AgentKind {
51 /// Anthropic Claude Code (`claude -p`).
52 Claude,
53 /// opencode (`opencode run`).
54 Opencode,
55 /// Antigravity CLI (`agy -p`). Gemini CLI is deliberately absent: Google
56 /// retired the standalone client for individual accounts in favour of this
57 /// one, so an adapter for it would be dead code on a live machine.
58 Antigravity,
59 /// OpenAI Codex CLI (`codex exec`). The one roster member with a real
60 /// read-only mode: `--sandbox read-only` is enforced by the CLI, not by
61 /// the prompt.
62 Codex,
63 /// oh-my-pi (`omp -p --mode=json`). Another CLI with no read-only mode of
64 /// its own - `--auto-approve` gates reads and writes together - so a
65 /// read-only seat rests on the prompt and on the worktree discipline the
66 /// other non-codex seats already rely on. It is also the roster member that
67 /// reaches DeepSeek, whose models ship in `omp`'s own catalog.
68 Omp,
69 /// Arbitrary command. The escape hatch, and what the test suite drives.
70 Command,
71}
72
73impl AgentKind {
74 /// Executable that must be on `PATH` for this kind, if any.
75 pub fn program(self) -> Option<&'static str> {
76 match self {
77 Self::Claude => Some("claude"),
78 Self::Opencode => Some("opencode"),
79 Self::Antigravity => Some("agy"),
80 Self::Codex => Some("codex"),
81 Self::Omp => Some("omp"),
82 Self::Command => None,
83 }
84 }
85
86 /// Lowercase name as written in the config file.
87 pub fn as_str(self) -> &'static str {
88 match self {
89 Self::Claude => "claude",
90 Self::Opencode => "opencode",
91 Self::Antigravity => "antigravity",
92 Self::Codex => "codex",
93 Self::Omp => "omp",
94 Self::Command => "command",
95 }
96 }
97
98 /// Every kind the roster can name, in display order.
99 ///
100 /// This is what `magi doctor` lists, and it has to be one list rather than
101 /// the same set typed out again wherever a kind is enumerated. The last
102 /// time it was typed out twice, `omp` was added as a roster member and the
103 /// doctor output went on saying the machine had four CLIs - which reads as
104 /// "that agent is not installed" to the person the command exists for.
105 pub const ALL: [Self; 6] = [
106 Self::Claude,
107 Self::Opencode,
108 Self::Antigravity,
109 Self::Codex,
110 Self::Omp,
111 Self::Command,
112 ];
113}
114
115/// How the prompt reaches the agent process.
116#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize, Serialize)]
117#[serde(rename_all = "lowercase")]
118pub enum Delivery {
119 /// Piped on stdin.
120 Stdin,
121 /// Passed as a positional argument. Beware OS command-line limits.
122 Argv,
123 /// Written to a file; the agent is told to read it. No length limit.
124 File,
125}
126
127/// One addressable agent in the roster.
128#[derive(Debug, Clone, Deserialize, Serialize)]
129#[serde(deny_unknown_fields)]
130pub struct AgentSpec {
131 /// Stable identifier used by `[roles]` and by the stats tables.
132 pub id: String,
133 /// Which CLI to drive.
134 pub kind: AgentKind,
135 /// Model passed through to the CLI (`--model` / `-m`). CLI default if unset.
136 #[serde(default)]
137 pub model: Option<String>,
138 /// `kind = "command"` only: argv. Supports `{prompt_file}`, `{cwd}`,
139 /// `{label}`, `{session}` placeholders.
140 #[serde(default)]
141 pub command: Vec<String>,
142 /// Extra arguments appended to the built command line.
143 #[serde(default)]
144 pub extra_args: Vec<String>,
145 /// Extra environment variables for the child process.
146 #[serde(default)]
147 pub env: BTreeMap<String, String>,
148 /// Override the per-kind prompt delivery default.
149 #[serde(default)]
150 pub prompt_delivery: Option<Delivery>,
151}
152
153impl AgentSpec {
154 /// Default prompt delivery for this agent.
155 ///
156 /// `opencode` and `agy` take the prompt as an argument, which on Windows
157 /// caps out around 32 KiB — well under a judging prompt carrying three
158 /// patches — so both get a file instead.
159 pub fn delivery(&self) -> Delivery {
160 self.prompt_delivery.unwrap_or(match self.kind {
161 AgentKind::Claude | AgentKind::Command => Delivery::Stdin,
162 // `codex exec -` reads the prompt from stdin, so the whole
163 // instruction arrives without an argv length limit and without a
164 // tool round-trip to open a file.
165 AgentKind::Codex => Delivery::Stdin,
166 // `omp -p` reads the prompt from stdin too, and a judging prompt
167 // carrying three patches is well past the Windows argv cap, so this
168 // is the only delivery that works for every node.
169 AgentKind::Omp => Delivery::Stdin,
170 AgentKind::Opencode | AgentKind::Antigravity => Delivery::File,
171 })
172 }
173
174 /// Human-facing label, e.g. `opus (claude:opus)`.
175 pub fn display(&self) -> String {
176 match &self.model {
177 Some(m) => format!("{} ({}:{m})", self.id, self.kind.as_str()),
178 None => format!("{} ({})", self.id, self.kind.as_str()),
179 }
180 }
181}
182
183/// A standalone seat's agent: one roster id, or an ordered fallback chain.
184///
185/// The string form is what every existing config uses and serializes back as
186/// a string; the array form tries each id at most once, in order, moving on
187/// when a call ends in quota or fails (see [`crate::agent::pick_chain`]).
188/// An empty string or array is the same as unset.
189#[derive(Debug, Clone, PartialEq, Eq, Deserialize, Serialize)]
190#[serde(untagged)]
191pub enum AgentChoice {
192 /// One agent id.
193 One(String),
194 /// Ordered fallback chain of agent ids.
195 Chain(Vec<String>),
196}
197
198impl AgentChoice {
199 /// The ids named, in order, with blanks dropped. Empty means unset.
200 pub fn ids(&self) -> Vec<&str> {
201 let all: Vec<&str> = match self {
202 Self::One(id) => vec![id.as_str()],
203 Self::Chain(ids) => ids.iter().map(String::as_str).collect(),
204 };
205 all.into_iter()
206 .map(str::trim)
207 .filter(|id| !id.is_empty())
208 .collect()
209 }
210
211 /// The first id named, if any.
212 pub fn primary(&self) -> Option<&str> {
213 self.ids().into_iter().next()
214 }
215}
216
217impl From<&str> for AgentChoice {
218 fn from(id: &str) -> Self {
219 Self::One(id.to_owned())
220 }
221}
222
223/// The first id of an optional choice: what a single-seat caller resolves.
224pub fn primary(choice: Option<&AgentChoice>) -> Option<&str> {
225 choice.and_then(AgentChoice::primary)
226}
227
228/// Explicit role assignment. Empty lists are filled in by
229/// [`Config::resolve_roles`] by rotating the roster.
230#[derive(Debug, Clone, Default, Deserialize, Serialize)]
231#[serde(deny_unknown_fields, default)]
232pub struct Roles {
233 /// Agents that implement the task, one worktree each.
234 pub implementers: Vec<String>,
235 /// Agents that rank the candidates blind.
236 pub judges: Vec<String>,
237 /// Agents that review the winning patch.
238 pub reviewers: Vec<String>,
239 /// Agent that applies review findings. Defaults to the winner's author.
240 ///
241 /// Accepts one id or an ordered array of ids: a fallback chain, see
242 /// [`AgentChoice`]. The review loop, the gate-fix round, `land`'s fix
243 /// rounds and the rebase conflict round move to the next id when one
244 /// errors, hits its quota or answers nothing usable; each id is tried at
245 /// most once per call, and once a fallback answered, the rest of the run
246 /// starts from it (see `crate::fixer`). Unset keeps the winner's author.
247 pub fixer: Option<AgentChoice>,
248 /// Agent that answers the standing chat's turns (`src/talk.rs`).
249 ///
250 /// Unset picks a `claude` seat, else the first runnable agent in roster
251 /// order (see [`crate::agent::pick`]) - which is roster *order*, not a
252 /// judgement about who converses well. Naming one here matters once an
253 /// agent is also sitting as a judge: the chat is opened far more often
254 /// than any single competition, and every open competes with that judge
255 /// seat for the same account's concurrency. A timeout on an ordinary chat
256 /// turn traced to exactly this - `opus` triple-booked as chatter and judge
257 /// - is what this field exists to let an operator break apart.
258 ///
259 /// Accepts one id or an ordered array of ids: a fallback chain, see
260 /// [`AgentChoice`].
261 pub chatter: Option<AgentChoice>,
262 /// Agent that judges duplicate-work claims in [`crate::dupes`].
263 /// Unset inherits [`Self::chatter`], including its default agent selection.
264 /// Accepts a string or an ordered fallback chain, like [`Self::chatter`].
265 pub dupes_judge: Option<AgentChoice>,
266 /// Agent that arranges the queue between polls: `crate::conduct`'s single
267 /// seat, called once per cycle to decide a runnable task's `blocked_by`
268 /// and how a stalled or finished task recovers.
269 ///
270 /// The same precedent as [`Self::chatter`] for a seat that stands alone
271 /// rather than rotating through the roster - resolved through
272 /// [`crate::agent::pick`], so unset falls back to its own default order
273 /// (a claude seat, else the first runnable agent) rather than reusing a
274 /// judge or reviewer seat that the conductor's own poll-cycle cadence
275 /// would otherwise compete with for the same account's concurrency.
276 ///
277 /// Accepts a string or a fallback chain, like [`Self::chatter`].
278 pub conductor: Option<AgentChoice>,
279 /// Seats for the design-deliberation stage `graph::Runner::advise` runs
280 /// before `implement`: independent, read-only design proposals gathered
281 /// once a run has a settled task instruction and before any implementer
282 /// touches the repository.
283 ///
284 /// Empty falls back to `judges` rather than to the whole roster: a panel
285 /// trusted to rank patches independently is exactly the panel worth
286 /// asking to sketch a design independently, and an operator who has
287 /// already thought about judge diversity gets advisor diversity for free
288 /// instead of a fourth roster to maintain.
289 pub advisors: Vec<String>,
290 /// Agent that blends [`Self::advisors`]' proposals into the one design
291 /// brief `graph::Runner::synthesize_brief` carries into every
292 /// implementer's prompt.
293 ///
294 /// The same precedent as [`Self::chatter`] and [`Self::conductor`] for a
295 /// seat that stands alone rather than rotating through the roster -
296 /// resolved through [`crate::agent::pick`], so unset falls back to its
297 /// own default order (a claude seat, else the first runnable agent in
298 /// roster order) - exactly today's behavior, unchanged by leaving this
299 /// field out.
300 ///
301 /// Accepts a string or a fallback chain, like [`Self::chatter`].
302 pub synthesizer: Option<AgentChoice>,
303 /// Agent (normally a `kind = "command"` wrapper around Jev) that decides
304 /// whether a pull request title and prose are English, for
305 /// [`crate::github_text`]'s posting gate.
306 ///
307 /// Unset spawns nothing: the gate keeps its built-in heuristics. Set, a
308 /// failed, timed-out or unparsable call also falls back to them, so the
309 /// gate never depends on this seat being reachable. Accepts a string or a
310 /// fallback chain, like [`Self::chatter`].
311 pub language_judge: Option<AgentChoice>,
312}
313
314/// Graph shape and limits.
315#[derive(Debug, Clone, Deserialize, Serialize)]
316#[serde(deny_unknown_fields, default)]
317pub struct Graph {
318 /// Gate GitHub prose before posting. Secret redaction remains mandatory when disabled.
319 pub github_text_guard: bool,
320 /// Parallel implementations of the same task (the number of implementer
321 /// seats; pairs with [`Roles::implementers`]). **One by default.**
322 ///
323 /// Formerly `candidates`, still read as a deprecated alias.
324 ///
325 /// Competition is the thing magi is for, and it is still here - it is just
326 /// no longer what every task buys without being asked. Three days and 13
327 /// runs on this repository, which is the workload these numbers are drawn
328 /// from:
329 ///
330 /// - **0 of 13** competed runs reached a merge. Everything that landed in
331 /// that window went through `magi review` - the cheap half, no
332 /// competition - and passed on the first try.
333 /// - The judges' first choices **split 73% of the time** (8 of 11
334 /// tallies). Candidates that close together make the ranking a weak
335 /// signal for what it costs to produce.
336 /// - One run's own breakdown: implement 60min, judge 40min, fix 40min,
337 /// review 28min, **verify 5min** - and verify is the node that caught a
338 /// defect every reviewer had passed as clean. The cheapest step is the
339 /// one that earns its place every time.
340 ///
341 /// It is not worthless: `oc` won 3 of those tallies against `sonnet`, so a
342 /// single-seat default would have shipped the worse implementation in
343 /// roughly a quarter of them. That is exactly why this is a *default* and
344 /// not a removal - `magi run --candidates N` and a per-task seat count are
345 /// how a task that deserves a competition gets one.
346 ///
347 /// A single-candidate run needs no special case: `Runner::review`'s doc
348 /// records that `execute` already degrades to implement -> review -> gate
349 /// -> merge, because `judge` skips a one-candidate field, `deliberate` has
350 /// no two first choices to reconcile and `vote` returns early.
351 #[serde(alias = "candidates")]
352 pub implementers: usize,
353 /// Independent judges.
354 pub judges: usize,
355 /// Deliberation rounds when the judges' first choices disagree.
356 pub deliberate_rounds: usize,
357 /// Reviewers per review round. **Three by default** - the smallest panel
358 /// a lens cycle (see [`crate::prompt::Lens`]) covers exactly once, so the
359 /// default panel reads the patch for spec compliance, regressions, and
360 /// simplicity without repeating an angle. Review is also the one stage
361 /// [`Self::implementers`]'s doc describes as running on every task
362 /// regardless of competition, which is what makes a panel worth its cost
363 /// here even though `implementers` itself defaults to one.
364 pub reviewers: usize,
365 /// Maximum review+fix rounds before the run is declared blocked.
366 pub review_rounds: usize,
367 /// Fix rounds a failing `verify.gate` gets before the run is blocked. The
368 /// fixer is handed the gate's own output; `0` keeps the old behaviour of
369 /// blocking on the first red gate. Independent of [`Self::review_rounds`]
370 /// so a run that spent its review budget can still repair a one-line gate
371 /// failure.
372 pub gate_fix_rounds: usize,
373 /// Maximum agent processes running at once.
374 pub max_parallel: usize,
375 /// Language for the prose the agents write (`en` / `ja` / any language name).
376 pub language: String,
377 /// Keep one CLI conversation per seat, so a judge remembers its own
378 /// argument across deliberation rounds and the fixer remembers its own
379 /// implementation across review rounds.
380 ///
381 /// Sessions are scoped to a *seat*, never to an agent id: the same model
382 /// sitting as implementer and as judge gets two unrelated conversations,
383 /// which is what keeps blind judging blind.
384 pub sessions: bool,
385 /// Per-node timeouts, seconds.
386 pub timeout_implement: u64,
387 /// Per-node timeouts, seconds.
388 pub timeout_judge: u64,
389 /// Per-node timeouts, seconds.
390 pub timeout_review: u64,
391 /// Timeout for `verify.e2e` and `verify.gate`, seconds. Separate from
392 /// [`Self::timeout_review`] so shrinking a reviewer's budget cannot
393 /// silently shrink a real-machine command's budget too — the two used to
394 /// share `timeout_review`, and turning a slow reviewer down cut the
395 /// timeout `cargo test --all-targets` runs under along with it. When
396 /// omitted, preserves legacy configurations by using
397 /// [`Self::timeout_review`]. Set an explicit value to make verification
398 /// independent of later review-seat budget changes.
399 pub timeout_verify: Option<u64>,
400 /// Per-node timeouts, seconds.
401 pub timeout_fix: u64,
402 /// Wall-clock limit for one turn of [`crate::talk`]'s standing
403 /// conversation, seconds.
404 ///
405 /// An hour: the operator is not watching this turn resolve in real time,
406 /// so the budget can match what the work - reading files, running
407 /// commands, checking their output - actually needs rather than what a
408 /// person waiting on a phone can tolerate.
409 pub timeout_talk: u64,
410 /// Longest an agent seat may wait, before it is spawned, for something
411 /// that is not the agent itself (today: the shared build cache's lease),
412 /// seconds.
413 ///
414 /// A separate clock from every `timeout_*` above: those start when the CLI
415 /// is spawned and never include this wait. Expiry is a resource block (a
416 /// resumable hold that spends no attempt and rotates no agent), not an
417 /// agent timeout.
418 pub timeout_resource_wait: u64,
419 /// Retries for an agent invocation that fails or returns nothing usable.
420 pub retries: usize,
421 /// Root for candidate / judge worktrees. Defaults to `~/wt/magi`.
422 pub worktree_root: Option<PathBuf>,
423 /// After the pull request is open, keep going: watch its checks and
424 /// reviews, run a fix round when they are unhappy, and ask to merge.
425 ///
426 /// On, because stopping at an open pull request left the operator doing
427 /// the watching by hand - six times in the session this was built in - and
428 /// that is the work the loop exists to take. It only engages for
429 /// `merge = "pr"`; every other merge mode ends the run as before.
430 ///
431 /// Turning this on does **not** hand magi the merge button:
432 /// [`Graph::land_approval`] is on too, and nothing merges without an
433 /// explicit answer. Setting both to their non-defaults is the only way to
434 /// get an unattended merge, and it has to be chosen twice.
435 pub land: bool,
436 /// Land rounds - watch, fix, push - before the run is left for a human.
437 pub land_rounds: usize,
438 /// When `land` sees a required check fail on the head it is bound to, rerun
439 /// that check's workflow run once (`gh run rerun <id> --failed`) before
440 /// spending a fix round on it. A run id is rerun at most once, recorded
441 /// before the call, and forgotten when the head moves; the rerun costs no
442 /// fix round and no attempt. A failure with no workflow run (an external
443 /// status) goes straight to the fixer. On; off sends every red check to
444 /// the fixer at once.
445 pub rerun_flaky_checks: bool,
446 /// Ask the owner before merging, showing what is about to land.
447 ///
448 /// On, and it is what makes `land` safe to have on: the question carries a
449 /// rendered panel - the diffstat, the patch, the checks, the review
450 /// comments that were addressed, and the subject the squash will use - so
451 /// the decision is made on evidence rather than on trust, from wherever
452 /// the operator happens to be.
453 ///
454 /// Silence is a hold. An unanswered approval never merges, and neither
455 /// does any answer other than the word `merge`.
456 pub land_approval: bool,
457 /// Ask the owner before merging a run whose review loop handed off over a
458 /// contested finding: the last round ended with a Major-or-above finding
459 /// still open *and* at least one reviewer's final vote (after
460 /// reconsideration) was reject. Findings raised in the last round are
461 /// never fixed or re-reviewed, so this is the one case an unattended
462 /// merge would land over a reviewer's explicit objection.
463 ///
464 /// On. Interaction with [`Self::land_approval`]: when that is on, every
465 /// merge is asked about and this key changes nothing. When it is off,
466 /// this key alone decides whether a contested hand-off is asked about
467 /// (the question says why); every other run still merges unattended.
468 /// Read at land time, so turning it off restores the unattended merge at
469 /// once. The pull request is opened either way; only the merge waits.
470 pub hold_contested_merge: bool,
471 /// After a run's pull request merges, file the findings its last review
472 /// round left open as follow-up tasks (`crate::followup`) instead of
473 /// letting them vanish with the merge: every Major-or-above finding, and
474 /// every finding of a seat whose final vote was reject. Other Minor/Nit
475 /// findings are only listed in the pull-request comment. Does not affect
476 /// whether the merge happens. On; best-effort, a failure is a run event.
477 pub file_followups: bool,
478 /// After a run's pull request merges, release the follow-up tasks the
479 /// merge approval's deputy filed held for that pull request
480 /// (`crate::followup::settle_deputy_tasks`): only a task the deputy filed
481 /// for this run whose hold reason names this pull request, exactly once,
482 /// and never one that duplicates an automatic follow-up (that one is left
483 /// held with the reason). The generation cap applies. Independent of
484 /// `file_followups`. On; off leaves them held until `magi task release`.
485 pub release_deputy_followups: bool,
486 /// How long to wait for an owner to answer a question before the run is
487 /// abandoned, seconds. A parked run costs nothing, so this is generous;
488 /// it exists so a forgotten question cannot pin a worktree forever.
489 pub answer_timeout: u64,
490 /// What a round does when one or more reviewer seats never answered
491 /// (timeout, crash, unparsable output).
492 pub incomplete_review: IncompleteReviewPolicy,
493 /// Run `verify.e2e` on every round, even one that already has blocking
494 /// findings and another round left to try.
495 ///
496 /// Off by default: a round with a blocking finding and rounds still left
497 /// is going back to the fixer regardless of what `verify.e2e` says, so
498 /// running it first only spends the round's slowest step (minutes, on a
499 /// Rust repo's `cargo test --all-targets`) on a head about to be
500 /// rewritten anyway. `verify.e2e` still runs once a round has no
501 /// blocking findings left (a round cannot go `clean` without it) and the
502 /// final `verify.gate` always runs on the actual tree that would land —
503 /// deferring is about *when* e2e runs mid-loop, never about skipping it.
504 ///
505 /// Set this to restore the old every-round diagnostic behaviour: e2e
506 /// output from a round that still has blocking findings is occasionally
507 /// useful on its own (a runtime failure a reviewer's panel would not
508 /// have caught by reading), and this is the way back to seeing it every
509 /// round instead of only once the panel has nothing left to flag.
510 pub e2e_every_round: bool,
511 /// Run the design-deliberation stage before `implement`: independent
512 /// advisor seats each sketch a design, and (when at least one produced a
513 /// usable proposal) a synthesis blends them into a brief carried in the
514 /// implementer's prompt. See `graph::Runner::advise`.
515 ///
516 /// On by default. A design sketch is a few paragraphs an agent can write
517 /// without touching the repository, where a full implementation is a
518 /// tool loop that re-reads the codebase on every turn - so three
519 /// sketches, gathered once before `implement` starts, cost a fraction of
520 /// a fourth candidate and buy back a form of the same disagreement
521 /// [`Self::implementers`]'s doc describes moving away from being the
522 /// default, on every run rather than only the ones an operator remembers
523 /// to ask for with `--implementers`.
524 pub advise: bool,
525 /// How many independent design proposals the deliberation stage gathers.
526 /// **Three by default** - the same number [`Self::implementers`]'s doc
527 /// names as the point where a fourth judge's first choice stopped
528 /// changing the tally.
529 pub advisors: usize,
530 /// After the tally picks a winner and before the review loop, a fixer
531 /// round ports the strengths the judges explicitly attributed to the
532 /// losing candidates onto the winner (`graph::Runner::graft`). Skipped
533 /// for a single candidate, a run with no winner or no quorum, and when
534 /// no judge recorded a strength for a loser. Best-effort: a failure, a
535 /// quota hit, no change or a diff over the size limit restores the
536 /// winner's pre-graft commit and the run carries on as if this were off.
537 /// On by default.
538 pub graft_runner_up: bool,
539}
540
541impl Default for Graph {
542 fn default() -> Self {
543 Self {
544 github_text_guard: true,
545 implementers: 1,
546 judges: 3,
547 deliberate_rounds: 1,
548 reviewers: 3,
549 review_rounds: 6,
550 gate_fix_rounds: 1,
551 max_parallel: 4,
552 language: "en".to_owned(),
553 sessions: true,
554 timeout_implement: 3600,
555 timeout_judge: 1200,
556 timeout_review: 1200,
557 timeout_verify: None,
558 timeout_fix: 1800,
559 timeout_talk: 3600,
560 timeout_resource_wait: 1800,
561 retries: 1,
562 worktree_root: None,
563 land: true,
564 land_rounds: 4,
565 rerun_flaky_checks: true,
566 land_approval: true,
567 hold_contested_merge: true,
568 file_followups: true,
569 release_deputy_followups: true,
570 answer_timeout: 86_400,
571 incomplete_review: IncompleteReviewPolicy::Block,
572 e2e_every_round: false,
573 advise: true,
574 advisors: 3,
575 graft_runner_up: true,
576 }
577 }
578}
579
580impl Graph {
581 /// Effective machine-command budget. Older configuration files had only
582 /// `timeout_review`, which also governed verification, so absence is a
583 /// compatibility fallback rather than a new 1200-second default.
584 pub fn verify_timeout(&self) -> u64 {
585 self.timeout_verify.unwrap_or(self.timeout_review)
586 }
587}
588
589/// What a review round does when a reviewer seat never answered.
590///
591/// A round where half the panel timed out is not evidence of a clean patch —
592/// it is evidence of nothing. The default refuses to call that clean; `warn`
593/// exists for an operator who would rather keep a flaky seat from stalling
594/// every run, and accepts that the gap is on them to read in the report.
595#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize, Serialize)]
596#[serde(rename_all = "lowercase")]
597pub enum IncompleteReviewPolicy {
598 /// A round with a missing seat is never `clean`: with nothing raised to
599 /// fix, the round is re-reviewed instead of gating; with the max rounds
600 /// exhausted, the run is left `Blocked` rather than declared ready.
601 Block,
602 /// A round with a missing seat can still gate as clean, once every seat
603 /// that *did* answer raised nothing blocking and verification is green.
604 /// The record keeps the gap visible (`magi show`, `magi stats`) even
605 /// though the run does not wait on it.
606 Warn,
607}
608
609/// What to do when vendor-identifying text is found in material shown to judges.
610#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize, Serialize)]
611#[serde(rename_all = "lowercase")]
612pub enum LeakPolicy {
613 /// Record the leak, show the patch unmodified.
614 Warn,
615 /// Replace the token with `[REDACTED]` in the presented patch.
616 Redact,
617 /// Abort the run.
618 Fail,
619}
620
621/// Blindness policy.
622///
623/// Commit messages and candidate summaries are *always* stripped of
624/// attribution trailers and redacted — that is where signatures actually
625/// appear. [`Blind::on_leak`] governs the patch body only, where blanket
626/// redaction would corrupt the artifact under judgement.
627#[derive(Debug, Clone, Deserialize, Serialize)]
628#[serde(deny_unknown_fields, default)]
629pub struct Blind {
630 /// Install a per-worktree `commit-msg` hook that deletes attribution
631 /// trailers before they can land in a candidate's history.
632 pub commit_msg_hook: bool,
633 /// Literal, case-insensitive substrings. A line containing any of them is
634 /// dropped from commit messages and summaries; the `commit-msg` hook is
635 /// generated from the same list.
636 pub strip_lines: Vec<String>,
637 /// Case-insensitive substrings that identify a vendor or model.
638 pub vendor_tokens: Vec<String>,
639 /// Policy for vendor tokens found in the patch body.
640 pub on_leak: LeakPolicy,
641 /// Seed for label assignment and per-judge presentation order. Derived from
642 /// the run id when unset; set it to make a run reproducible.
643 pub seed: Option<u64>,
644}
645
646impl Default for Blind {
647 fn default() -> Self {
648 Self {
649 commit_msg_hook: true,
650 strip_lines: [
651 "Co-Authored-By:",
652 "Signed-off-by:",
653 "Assisted-by:",
654 "Generated-by:",
655 "Generated with",
656 "\u{1f916}",
657 ]
658 .iter()
659 .map(|s| (*s).to_owned())
660 .collect(),
661 vendor_tokens: [
662 "claude",
663 "anthropic",
664 "codex",
665 "openai",
666 "chatgpt",
667 "gemini",
668 "grok",
669 "xai",
670 "copilot",
671 "opencode",
672 "qoder",
673 "cursor",
674 "\u{1f916}",
675 ]
676 .iter()
677 .map(|s| (*s).to_owned())
678 .collect(),
679 on_leak: LeakPolicy::Warn,
680 seed: None,
681 }
682 }
683}
684
685/// Shell commands that gate the winner.
686#[derive(Debug, Clone, Default, Deserialize, Serialize)]
687#[serde(deny_unknown_fields, default)]
688pub struct Verify {
689 /// Run in the winner's worktree once per review round. Its output is fed
690 /// back to the fixer. This is the "real machine" leg of the review.
691 pub e2e: Vec<String>,
692 /// Final gate. Must all exit 0 before a merge is attempted.
693 pub gate: Vec<String>,
694 /// Mechanical fixers (formatters) run in the winner's worktree just before
695 /// `gate`, after review is clean. Anything they change is committed as one
696 /// neutral-identity commit so the gate and the PR see it. A command that
697 /// fails or times out is a warning, never a failed run: the gate stays the
698 /// single arbiter. Empty (the default) turns the feature off. Runs under
699 /// `graph.verify_timeout()` per command, like the gate. Meant for light
700 /// tools: it is not scanned for `CARGO_TARGET_DIR`, and it should be
701 /// idempotent since a resumed run may repeat it.
702 #[serde(default)]
703 pub pre_gate: Vec<String>,
704 /// Shell used to run the commands above. Defaults to `sh -c`, or
705 /// `cmd /C` when `sh` is not on `PATH`.
706 pub shell: Option<Vec<String>>,
707}
708
709impl Verify {
710 /// The `CARGO_TARGET_DIR=` value of the first rendered command that sets
711 /// one, if any. See [`crate::disk::extract_cargo_target_dir`] for the shape
712 /// this reads back. One rendering is enough - they all set the same
713 /// rendered `{{ vars.cache }}` path via the same shell - and the first e2e
714 /// command is checked before the gate because the e2e rebuilds the crate.
715 pub fn cache_dir(&self) -> Option<PathBuf> {
716 self.e2e
717 .iter()
718 .chain(self.gate.iter())
719 .find_map(|cmd| crate::disk::extract_cargo_target_dir(cmd))
720 }
721
722 /// Every `--manifest-path` the e2e and gate commands pass to cargo, for the
723 /// shared cache's freshness check in repositories with no root manifest.
724 pub fn cargo_manifests(&self) -> Vec<PathBuf> {
725 let all: Vec<String> = self.e2e.iter().chain(self.gate.iter()).cloned().collect();
726 crate::cache::extract_manifest_paths(&all)
727 }
728}
729
730/// Per-repository worktree preparation, run right after magi creates a worktree.
731#[derive(Debug, Clone, Default, PartialEq, Deserialize, Serialize)]
732#[serde(deny_unknown_fields, default)]
733pub struct Worktree {
734 /// Ordered steps. Declare in one layer only. Empty (the default) changes
735 /// nothing. See [`crate::worktree_setup`].
736 pub setup: Vec<SetupStep>,
737}
738
739/// One setup step: exactly one of `copy` / `run`.
740#[derive(Debug, Clone, Default, PartialEq, Deserialize, Serialize)]
741#[serde(deny_unknown_fields, default)]
742pub struct SetupStep {
743 /// `"src -> dst"` (or just `"path"`): copy from the primary checkout (an
744 /// absolute source is allowed) into the worktree. Never overwrites.
745 pub copy: Option<String>,
746 /// With `copy`: a missing source is skipped instead of an error.
747 pub optional: bool,
748 /// Shell command, run with the worktree as cwd.
749 pub run: Option<String>,
750 /// With `run`: seconds before the command is killed. Defaults to
751 /// `graph.verify_timeout()`.
752 pub timeout_secs: Option<u64>,
753}
754
755impl Worktree {
756 /// Refuse a malformed step at load time rather than in the middle of a run.
757 pub fn validate(&self) -> Result<()> {
758 for (i, step) in self.setup.iter().enumerate() {
759 let n = i + 1;
760 match (&step.copy, &step.run) {
761 (Some(c), None) => {
762 if step.timeout_secs.is_some() {
763 bail!("[worktree] setup step {n}: `timeout_secs` applies to `run` only");
764 }
765 crate::worktree_setup::validate_copy(c)
766 .with_context(|| format!("[worktree] setup step {n}"))?;
767 }
768 (None, Some(r)) => {
769 if r.trim().is_empty() {
770 bail!("[worktree] setup step {n}: `run` is empty");
771 }
772 if step.optional {
773 bail!("[worktree] setup step {n}: `optional` applies to `copy` only");
774 }
775 }
776 _ => bail!("[worktree] setup step {n} must set exactly one of `copy` and `run`"),
777 }
778 }
779 Ok(())
780 }
781}
782
783/// Disk hygiene: how hard magi is allowed to press on the machine's free space.
784///
785/// The numbers below come from one incident, not from theory: a machine with
786/// 951.8 GB free ran a few competitions and plans and best read 6.7 GB free.
787/// Three multi-gigabyte classes of junk accumulated side by side - per-run
788/// worktrees that end as `Merged`/`Ready`/`Failed`, a shared build cache whose
789/// each verify round and each implementation wave recompiles the derived
790/// section of the project into, and the outputs of runs that were removed but
791/// whose folders nobody deleted.
792#[derive(Debug, Clone, PartialEq, Deserialize, Serialize)]
793#[serde(deny_unknown_fields, default)]
794pub struct Disk {
795 /// Free space, in bytes, below which no new run may start: the daemon and
796 /// the `magi run` gate answer with "the disk is full" instead of letting
797 /// the graph fill it the rest of the way. `0` turns the gate off.
798 ///
799 /// Default 8 GiB. The incident ran down to 6.7 GB free of 951.8 GB total
800 /// before anybody noticed; 8 GiB is enough headroom for the compile a fresh
801 /// competition triggers and small enough that a 1 TB disk with 100 GB free
802 /// is nowhere near the threshold.
803 pub min_free_bytes: u64,
804 /// Fold finished runs without being asked. `Merged`, `Ready` and `Failed`
805 /// runs older than [`fold_grace_secs`](Self::fold_grace_secs) have their
806 /// worktrees removed. `0` turns the janitor off.
807 ///
808 /// Default true.
809 pub auto_fold: bool,
810 /// How old a finished run must be before the janitor folds it, seconds.
811 ///
812 /// Default 6 hours. A run that `Ready` at 8am is the operator's answer; a
813 /// run that `Ready` a week ago is worktrees holding a compile each. Six
814 /// hours is long enough that nobody loses an answer in the gap between
815 /// reading the report and starting from it, and short enough that a backlog
816 /// cannot pile up across two nights.
817 pub fold_grace_secs: u64,
818 /// Ceiling for the shared build cache (`CARGO_TARGET_DIR` in the rendered
819 /// verify commands), in bytes. When the janitor runs and the cache is over
820 /// it, files are dropped oldest-first until it is not. `0` turns pruning
821 /// off - the cache then only ever grows, which is the operator's call.
822 ///
823 /// Default 10 GiB. This is what the incident measured: 30.61 GB sat in the
824 /// shared cache on top of ~16 GB in the primary target directory and 6.7-
825 /// 11.15 GB in each of four per-worktree targets. 10 GiB holds a healthy
826 /// stack of prebuilt dependencies (cargo's per-file fingerprinting means
827 /// pruning only costs the rebuild of the dropped files, not of the world)
828 /// without letting one addled cache swallow the machine.
829 pub cache_limit_bytes: u64,
830}
831
832impl Default for Disk {
833 fn default() -> Self {
834 Self {
835 min_free_bytes: 8 * 1024 * 1024 * 1024,
836 auto_fold: true,
837 fold_grace_secs: 6 * 60 * 60,
838 cache_limit_bytes: 10 * 1024 * 1024 * 1024,
839 }
840 }
841}
842
843/// What to do with the winning branch.
844#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize, Serialize)]
845#[serde(rename_all = "lowercase")]
846pub enum MergeMode {
847 /// Leave the branch alone and print the merge command matching
848 /// [`Merge::style`].
849 None,
850 /// Merge into the base branch in the primary worktree, using
851 /// [`Merge::style`].
852 Local,
853 /// Push the branch and open a PR with `gh pr create`. Landing this PR
854 /// (`[graph] land`) always squashes — see `land`'s module doc — so
855 /// [`Merge::style`] does not apply here.
856 Pr,
857}
858
859/// How the winning branch is attached to the base branch: what
860/// `mode = "local"` runs, and what `mode = "none"`'s printed guidance tells
861/// the operator to run by hand.
862///
863/// Read from configuration rather than asked of the repository at run time
864/// (e.g. `gh api repos/{owner}/{repo}/rulesets`) for two reasons: it keeps
865/// `mode = "none"`'s guidance a pure function of `RunState`, assertable in a
866/// unit test the same way `land::decide` is kept pure (see that module's
867/// doc), and it works for a base branch that is not hosted on GitHub, or not
868/// reachable at all, at the moment the report is rendered. An operator whose
869/// base branch enforces a ruleset already knows what it allows; declaring it
870/// once here is cheaper than magi re-discovering it, with a network call, on
871/// every render.
872#[derive(Debug, Clone, Copy, PartialEq, Eq, Default, Deserialize, Serialize)]
873#[serde(rename_all = "lowercase")]
874pub enum MergeStyle {
875 /// `git merge --no-ff`: every candidate commit lands, plus a merge
876 /// commit that records the merge as its own event in history. Rejected
877 /// by a base branch whose ruleset requires linear history or forbids
878 /// merge commits outright.
879 #[default]
880 Merge,
881 /// `git merge --squash` followed by a commit under an explicit message:
882 /// every candidate commit folds into one, and none of the candidate's
883 /// own placeholder subjects (`magi: candidate A (uncommitted work)`)
884 /// reach the base branch. No merge commit, so this satisfies a linear-
885 /// history ruleset.
886 Squash,
887 /// A fast-forward-only merge: every candidate commit lands verbatim, in
888 /// order, with no merge commit. Only succeeds because the winner was
889 /// already rebased onto the tracked base tip before this runs (see
890 /// `Runner::sync_to_base`) — equivalent to GitHub's "rebase and merge"
891 /// once that has happened.
892 Rebase,
893}
894
895/// Merge policy.
896#[derive(Debug, Clone, Deserialize, Serialize)]
897#[serde(deny_unknown_fields, default)]
898pub struct Merge {
899 /// Default is [`MergeMode::None`]: magi never touches your base branch
900 /// unless you ask it to.
901 pub mode: MergeMode,
902 /// Base branch. Defaults to the branch checked out when the run started.
903 pub base: Option<String>,
904 /// How the winner is attached to `base`; see [`MergeStyle`]. Ignored by
905 /// `mode = "pr"`.
906 pub style: MergeStyle,
907 /// Remote for `mode = "pr"`.
908 pub remote: String,
909 /// After a `mode = "pr"` run lands, open a `chore/release-vX.Y.Z` pull
910 /// request sized to the change by an agent's own judgement, so a version
911 /// bump does not depend on a human remembering to cut one.
912 ///
913 /// Only acts on Rust repositories (a `Cargo.toml` at the root of the base
914 /// branch); on any other repository it is a no-op that records one event.
915 ///
916 /// On by default. **Turning this off means a merge landed from the phone
917 /// never becomes a release**, so `POST /api/upgrade` keeps reporting
918 /// "already on the newest release" against a `main` that has moved past
919 /// it - the same gap this feature exists to close. Only for a repository
920 /// that wants to keep cutting releases by hand.
921 pub release_bump: bool,
922}
923
924impl Default for Merge {
925 fn default() -> Self {
926 Self {
927 mode: MergeMode::None,
928 base: None,
929 style: MergeStyle::default(),
930 remote: "origin".to_owned(),
931 release_bump: true,
932 }
933 }
934}
935
936/// How magi keeps itself current.
937#[derive(Debug, Clone, Copy, PartialEq, Eq, Deserialize, Serialize)]
938#[serde(rename_all = "lowercase")]
939pub enum UpdateMode {
940 /// Never check.
941 Off,
942 /// Check in the background and print a one-line banner when a newer
943 /// release exists.
944 Notify,
945 /// Check and install silently.
946 Install,
947}
948
949/// Self-update policy.
950#[derive(Debug, Clone, Deserialize, Serialize)]
951#[serde(deny_unknown_fields, default)]
952pub struct Update {
953 /// Default is [`UpdateMode::Notify`]: magi tells you, and lets you decide.
954 pub mode: UpdateMode,
955 /// Minimum time between checks, e.g. `24h`. kaishin's default when unset.
956 pub interval: Option<String>,
957}
958
959impl Default for Update {
960 fn default() -> Self {
961 Self {
962 mode: UpdateMode::Notify,
963 interval: None,
964 }
965 }
966}
967
968/// Who performs the release after a `chore/release-vX.Y.Z` pull request merges.
969#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Deserialize, Serialize)]
970#[serde(rename_all = "lowercase")]
971pub enum ReleaseMode {
972 /// The repository's own workflows do (`auto-tag.yml` then `release.yml`),
973 /// and the release pull request merges on its checks.
974 #[default]
975 Actions,
976 /// magi does, from this machine: no CI is awaited, the owner approves the
977 /// merge, and magi then tags and runs [`Release::commands`]. See
978 /// [`crate::release_local`].
979 Local,
980}
981
982/// `[release]`: releasing without GitHub Actions. Opt-in per repository.
983#[derive(Debug, Clone, Deserialize, Serialize)]
984#[serde(deny_unknown_fields, default)]
985pub struct Release {
986 /// [`ReleaseMode::Actions`] (default: nothing changes) or
987 /// [`ReleaseMode::Local`].
988 pub mode: ReleaseMode,
989 /// Shell commands run in order, from a clean checkout of the merge commit,
990 /// once the tag is pushed. The first failure stops the list. They see
991 /// `MAGI_RELEASE_VERSION`, `MAGI_RELEASE_TAG`, `MAGI_RELEASE_COMMIT` and
992 /// `MAGI_RELEASE_PR` in the environment; `{{ ... }}` cannot carry the
993 /// version because teravars renders the whole file before any release
994 /// exists. Only read when `mode = "local"`.
995 pub commands: Vec<String>,
996 /// Minutes each command may run before it is killed and counted as failed.
997 pub timeout_minutes: u64,
998}
999
1000impl Default for Release {
1001 fn default() -> Self {
1002 Self {
1003 mode: ReleaseMode::Actions,
1004 commands: Vec::new(),
1005 timeout_minutes: 30,
1006 }
1007 }
1008}
1009
1010impl Release {
1011 /// Is this repository released by magi rather than by its workflows?
1012 pub fn is_local(&self) -> bool {
1013 self.mode == ReleaseMode::Local
1014 }
1015}
1016
1017/// Top-level configuration.
1018#[derive(Debug, Clone, Default, Deserialize, Serialize)]
1019#[serde(deny_unknown_fields, default)]
1020pub struct Config {
1021 /// Agent roster.
1022 pub agents: Vec<AgentSpec>,
1023 /// Role assignment.
1024 pub roles: Roles,
1025 /// Graph shape.
1026 pub graph: Graph,
1027 /// Blindness policy.
1028 pub blind: Blind,
1029 /// Verification commands.
1030 pub verify: Verify,
1031 /// Worktree preparation.
1032 pub worktree: Worktree,
1033 /// Disk hygiene.
1034 pub disk: Disk,
1035 /// Merge policy.
1036 pub merge: Merge,
1037 /// Self-update policy.
1038 pub update: Update,
1039 /// Project-specific text appended to the node prompts.
1040 pub prompts: Prompts,
1041 /// How the operator is told a run is waiting on them.
1042 pub notify: Notify,
1043 /// Local repositories the plan surface can start or derive a conversation
1044 /// against.
1045 pub repos: Repos,
1046 /// Policy for the standing conversation ([`crate::talk`]).
1047 pub talk: Talk,
1048 /// How many runs `magi serve`'s own loop drives at once.
1049 pub daemon: Daemon,
1050 /// Who tags and publishes after a release pull request merges.
1051 pub release: Release,
1052 /// Context-window sizes in tokens, keyed by model name, merged over
1053 /// [`BUILTIN_CONTEXT_WINDOWS`]. See [`Config::context_window`].
1054 pub context_windows: BTreeMap<String, u64>,
1055}
1056
1057/// Context windows of models magi commonly drives, in tokens.
1058///
1059/// A convenience, not a source of truth: vendors resize windows and ship new
1060/// models faster than this table is edited, so an entry that is stale or
1061/// missing is fixed by `[context_windows]` in `magi.toml`, which wins. Matched
1062/// by [`Config::context_window`]: exact name first, then the longest key that
1063/// prefixes the model name (so a dated or suffixed id such as
1064/// `claude-sonnet-4-5-20250929` finds `claude-sonnet-4-5`). A model matching
1065/// nothing has no percentage - the UI shows the raw count.
1066const BUILTIN_CONTEXT_WINDOWS: &[(&str, u64)] = &[
1067 ("claude-opus-4", 200_000),
1068 ("claude-sonnet-4", 200_000),
1069 ("claude-haiku-4", 200_000),
1070 ("claude-3", 200_000),
1071 ("opus", 200_000),
1072 ("sonnet", 200_000),
1073 ("haiku", 200_000),
1074 ("gpt-5", 400_000),
1075 ("gpt-4.1", 1_000_000),
1076 ("gpt-4o", 128_000),
1077 ("o3", 200_000),
1078 ("o4-mini", 200_000),
1079 ("gemini-2.5", 1_000_000),
1080 ("deepseek", 128_000),
1081];
1082
1083impl Config {
1084 /// Context window in tokens for `model`: `[context_windows]` first, then
1085 /// the built-in table; within each, an exact name beats the longest
1086 /// prefix. `None` for an unknown model, which is a deliberate answer -
1087 /// guessing a denominator would print a confident wrong percentage.
1088 pub fn context_window(&self, model: &str) -> Option<u64> {
1089 fn lookup<'a>(model: &str, entries: impl Iterator<Item = (&'a str, u64)>) -> Option<u64> {
1090 let mut best: Option<(usize, u64)> = None;
1091 for (key, window) in entries {
1092 if key == model {
1093 return Some(window);
1094 }
1095 if model.starts_with(key) && best.is_none_or(|(len, _)| key.len() > len) {
1096 best = Some((key.len(), window));
1097 }
1098 }
1099 best.map(|(_, w)| w)
1100 }
1101 lookup(
1102 model,
1103 self.context_windows.iter().map(|(k, w)| (k.as_str(), *w)),
1104 )
1105 .or_else(|| lookup(model, BUILTIN_CONTEXT_WINDOWS.iter().copied()))
1106 }
1107}
1108
1109/// How the daemon loop itself behaves, as opposed to what one run does.
1110///
1111/// Machine-layer material in the same sense [`Repos::roots`] is: how many
1112/// competitions this machine's own loop is willing to babysit at once is a
1113/// fact about the machine running `magi serve`, not about any one
1114/// repository's task, so it belongs in `<config_dir>/magi/config.toml`
1115/// rather than a repository's own `magi.toml` - though, like `Repos::roots`,
1116/// nothing stops a repository from setting it too, since a scalar field
1117/// takes whichever layer has the highest precedence.
1118#[derive(Debug, Clone, Deserialize, Serialize)]
1119#[serde(deny_unknown_fields, default)]
1120pub struct Daemon {
1121 /// How many runs `magi serve` may have actively in flight at once.
1122 /// **One by default** - today's behaviour, one run at a time.
1123 ///
1124 /// This is a different knob from [`Graph::max_parallel`], and the two
1125 /// must not be confused: `max_parallel` bounds how many agent *processes*
1126 /// one run starts inside itself (implementers, judges, reviewers -
1127 /// candidates competing on a single task); this field bounds how many
1128 /// *runs* - whole competitions, each with its own `max_parallel` budget -
1129 /// the loop drives side by side, possibly across different tasks and
1130 /// different repositories. Raising `max_parallel` buys a bigger panel for
1131 /// one task; raising this buys more tasks worked at once. A config file
1132 /// that meant one and wrote the other would either starve a competition
1133 /// of judges or leave the rest of the backlog waiting for no reason.
1134 ///
1135 /// A run parked waiting on the operator's land-merge approval - see
1136 /// [`crate::land`] - does not hold one of these slots while it waits: the
1137 /// whole point of parking there is to let the loop spend the slot on
1138 /// something runnable instead of sitting on a decision only a human can
1139 /// make. So even at the default of `1`, an approval that comes back does
1140 /// not queue behind whatever else the loop happens to be running.
1141 pub max_concurrent_runs: usize,
1142 /// Let a task marked [`crate::queue::Task::interrupt`] (`magi task
1143 /// interrupt`) cut ahead of whatever `magi serve` already has in flight,
1144 /// instead of waiting for it to finish.
1145 ///
1146 /// **Off by default.** When enabled, a run that is in flight and is not
1147 /// itself the interrupt candidate may be paused at its next safe node
1148 /// boundary - see [`crate::graph::Runner::park_here`] - so the marked
1149 /// task can run alone; the paused run resumes automatically, through the
1150 /// same path any other parked run does, the moment the interrupting
1151 /// task's own run reaches a terminal status. This is opt-in because it
1152 /// bends the loop's own "one run at a time" principle (see this repository's
1153 /// `magi serve` help) for a specific, explicit operator request, and a
1154 /// repository that never files an interrupt task pays nothing for having
1155 /// it on - but an operator who does not want any run of theirs preempted,
1156 /// ever, should leave this `false`.
1157 pub pause_for_interrupts: bool,
1158 /// How many deputies `magi serve` runs at once. A deputy is the short-lived
1159 /// seat that waits on one open conductor question (see [`crate::deputy`]);
1160 /// each is a live agent conversation, so this is bounded - **two by
1161 /// default**, because the agent CLIs' quota is the scarce resource. A
1162 /// question over the limit keeps its place and starts when a deputy
1163 /// finishes; `0` runs none.
1164 pub max_deputies: usize,
1165 /// Minutes a release-bump pull request may make no progress before
1166 /// `magi serve` raises a notice and asks the owner (see
1167 /// [`crate::release_watch`]). Progress is a change of head commit or of any
1168 /// check's verdict. A required check that stays red after the one rerun the
1169 /// watcher is allowed escalates at once, whatever this says. **Sixty by
1170 /// default**; `0` turns the watcher off.
1171 pub release_stall_minutes: u64,
1172}
1173
1174impl Default for Daemon {
1175 fn default() -> Self {
1176 Self {
1177 max_concurrent_runs: 1,
1178 pause_for_interrupts: false,
1179 max_deputies: 2,
1180 release_stall_minutes: 60,
1181 }
1182 }
1183}
1184
1185/// Where `magi repos` and `GET /api/repos` look for local checkouts.
1186///
1187/// `roots` is one of the array keys [`array_merge_policy`] marks as
1188/// append-across-layers: which checkouts exist in general is a *machine*
1189/// fact in the same way the agent roster is - a repository's own `magi.toml`
1190/// cannot state where its siblings live before magi has resolved which
1191/// repository to read that file from in the first place - but a repository
1192/// that genuinely has an extra root worth scanning is not forced to choose
1193/// between an error and losing the machine's roots outright. Both layers'
1194/// roots are scanned; see [`Config::refuse_split_arrays`] for the keys that
1195/// are still refused.
1196#[derive(Debug, Clone, Deserialize, Serialize)]
1197#[serde(deny_unknown_fields, default)]
1198pub struct Repos {
1199 /// Roots to scan for a ghq-layout checkout: `<root>/<host>/<owner>/<repo>`
1200 /// with a `.git` directory. Empty by default - nothing is scanned unless
1201 /// asked to be.
1202 pub roots: Vec<PathBuf>,
1203 /// How long a scan is trusted before the next request re-scans it,
1204 /// seconds. `0` means never trust it: scan on every request. Defaults to
1205 /// a day, the same order of magnitude as [`Graph::answer_timeout`] for
1206 /// the same reason - a checkout does not usually appear or vanish inside
1207 /// a session, so there is little to gain from scanning more often than
1208 /// that, and an explicit refresh exists for the moment one does.
1209 pub scan_ttl: u64,
1210 /// How often `magi serve` runs `git fetch origin` in every checkout under
1211 /// `roots`, seconds, so `origin/main` stays fresh whatever state the
1212 /// working tree is in. Fetch only: no branch, HEAD or working tree is
1213 /// touched. `0` disables it.
1214 pub fetch_interval: u64,
1215}
1216
1217impl Default for Repos {
1218 fn default() -> Self {
1219 Self {
1220 roots: Vec::new(),
1221 scan_ttl: 86_400,
1222 fetch_interval: 600,
1223 }
1224 }
1225}
1226
1227/// Project-specific text appended to each node's prompt.
1228///
1229/// **Additive by construction.** These fields cannot replace magi's prompts,
1230/// only extend them, and that restriction is the whole design. The built-in
1231/// prompts carry the invariants the competition rests on: a judging prompt
1232/// names no authors, every structured answer must arrive as one fenced `json`
1233/// block, and a judge is told not to speculate about who wrote what. A config
1234/// that could overwrite them would let a typo silently un-blind the panel or
1235/// break the parser, and the symptom would be "the judges got worse" rather
1236/// than an error.
1237///
1238/// Repository-wide context belongs in `AGENTS.md`, which every agent already
1239/// reads from the checkout. Use these fields for the things a *magi node*
1240/// needs to know and a repository file cannot say - for instance that
1241/// reviewers here should ignore formatting because a hook owns it.
1242#[derive(Debug, Clone, Default, Deserialize, Serialize)]
1243#[serde(deny_unknown_fields, default)]
1244pub struct Prompts {
1245 /// Appended to every node's prompt.
1246 pub all: String,
1247 /// Appended for implementers.
1248 pub implement: String,
1249 /// Appended for judges, both ranking and voting.
1250 pub judge: String,
1251 /// Appended for reviewers.
1252 pub review: String,
1253 /// Appended for the fixer.
1254 pub fix: String,
1255}
1256
1257impl Prompts {
1258 /// The overlay for one node, or `None` when nothing is configured.
1259 ///
1260 /// `node` is the graph's own node name, so a new node gets no overlay
1261 /// rather than the wrong one.
1262 pub fn overlay(&self, node: &str) -> Option<String> {
1263 let specific = match node {
1264 "implement" => &self.implement,
1265 "judge" | "vote" | "deliberate" => &self.judge,
1266 "review" => &self.review,
1267 "fix" => &self.fix,
1268 _ => "",
1269 };
1270 let mut parts: Vec<&str> = Vec::new();
1271 for p in [self.all.trim(), specific.trim()] {
1272 if !p.is_empty() {
1273 parts.push(p);
1274 }
1275 }
1276 if parts.is_empty() {
1277 return None;
1278 }
1279 Some(parts.join("\n\n"))
1280 }
1281}
1282
1283/// How the operator is told that a run is waiting on them.
1284///
1285/// A command rather than a built-in integration: magi is one binary with no
1286/// network dependencies, and every operator's notification path is different -
1287/// ntfy, a Slack webhook, a Windows toast, an SSH to a machine that beeps.
1288/// Shelling out keeps all of them possible and none of them magi's problem.
1289#[derive(Debug, Clone, Default, Deserialize, Serialize)]
1290#[serde(deny_unknown_fields, default)]
1291pub struct Notify {
1292 /// Command and arguments. `{summary}`, `{run}` and `{url}` are replaced.
1293 /// Empty means no notification - the web UI is then the only surface.
1294 pub command: Vec<String>,
1295}
1296
1297/// Policy for [`crate::talk`], the standing conversation.
1298#[derive(Debug, Clone, Default, Deserialize, Serialize)]
1299#[serde(deny_unknown_fields, default)]
1300pub struct Talk {
1301 /// Let the conversation's agent edit files in the repository instead of
1302 /// filing a task for one. **Off by default** - see
1303 /// [`crate::talk`]'s module doc for why an edit made mid-conversation is
1304 /// an edit no run and no review can be attributed to, which is exactly
1305 /// the property a repository entered in a competition depends on. A
1306 /// repository that is never judged - dotfiles, a personal config
1307 /// checkout - has nothing to lose by turning this on in its own
1308 /// `magi.toml`, and a one-line fix stops costing a queued task to get.
1309 pub allow_write: bool,
1310 /// Extra chat personas, `[[talk.personas]]`. They add to the built-ins in
1311 /// [`crate::persona`]; an entry with a built-in's id replaces it. Like every
1312 /// array outside [`array_merge_policy`]'s append list, declaring it in two
1313 /// layers is refused.
1314 pub personas: Vec<PersonaSpec>,
1315 /// What the chat calls the operator: the persona addresses them by this
1316 /// name and the web Chat labels their turns with it instead of `YOU`.
1317 /// Blank is unset; at most [`OPERATOR_NAME_MAX`] characters, one line.
1318 pub operator_name: Option<String>,
1319}
1320
1321/// Longest accepted `[talk] operator_name`, in characters.
1322pub const OPERATOR_NAME_MAX: usize = 64;
1323
1324impl Talk {
1325 /// The trimmed operator name, `None` when unset or blank.
1326 pub fn operator_name(&self) -> Option<&str> {
1327 self.operator_name
1328 .as_deref()
1329 .map(str::trim)
1330 .filter(|n| !n.is_empty())
1331 }
1332}
1333
1334/// One `[[talk.personas]]` entry: a tone for the standing chat, nothing more.
1335#[derive(Debug, Clone, Default, PartialEq, Eq, Deserialize, Serialize)]
1336#[serde(deny_unknown_fields)]
1337pub struct PersonaSpec {
1338 /// Stable id the conversation stores. Not `default`.
1339 pub id: String,
1340 /// What the selector shows. May be omitted when `id` is a built-in's: the
1341 /// built-in's name is kept.
1342 #[serde(default)]
1343 pub name: String,
1344 /// The voice instruction appended to the chat's system prompt. May be
1345 /// omitted when `id` is a built-in's.
1346 #[serde(default)]
1347 pub prompt: String,
1348 /// Optional avatar, display only: a `data:image/(png|jpeg|gif|webp);base64,`
1349 /// URI or an absolute (or `~/`) path to a raster image. See
1350 /// [`crate::persona_icon`]. Never part of the briefing.
1351 #[serde(default, skip_serializing_if = "Option::is_none")]
1352 pub icon: Option<String>,
1353}
1354
1355/// Roles resolved to concrete agent specs for one run.
1356#[derive(Debug, Clone)]
1357pub struct ResolvedRoles {
1358 /// One per candidate.
1359 pub implementers: Vec<AgentSpec>,
1360 /// One per judge.
1361 pub judges: Vec<AgentSpec>,
1362 /// One per reviewer slot.
1363 pub reviewers: Vec<AgentSpec>,
1364 /// The configured fixer chain in order, if `[roles] fixer` is set
1365 /// (resolved like [`crate::agent::pick_chain`]: unknown or uninstalled ids
1366 /// skipped, duplicates keep their first place). `None` means the winner's
1367 /// own author.
1368 pub fixer: Option<Vec<AgentSpec>>,
1369 /// Queue conductor, explicitly selected or resolved by the standalone-seat fallback.
1370 pub conductor: AgentSpec,
1371 /// The full ordered implementer roster, in [`Roles::implementers`]'s own
1372 /// order — or `[[agents]]` in file order, when that list is empty. Unlike
1373 /// [`Self::implementers`], never truncated to `graph.implementers` and
1374 /// never `rotate`d/wrapped: a solo run (`implementers = 1`) resolves
1375 /// `implementers` down to a single slot, but `graph::Runner`'s per-seat
1376 /// quota fallback needs the *whole* list to walk forward through when
1377 /// that one slot's agent runs out of quota mid-run.
1378 pub implementer_roster: Vec<AgentSpec>,
1379 /// [`Self::implementer_roster`]'s counterpart for judge seats: the whole
1380 /// `[roles] judges` list (or every agent), untruncated and unrotated, for
1381 /// the per-seat handover when a judge's agent fails.
1382 pub judge_roster: Vec<AgentSpec>,
1383 /// Likewise for reviewer seats.
1384 pub reviewer_roster: Vec<AgentSpec>,
1385}
1386
1387/// Every array-valued key in a config table, as a dotted path.
1388///
1389/// Dotted so the error names `roles.implementers` rather than `implementers`:
1390/// an operator with three config files needs to know which key, not just that
1391/// there was one. `vars` is skipped because it is teravars' own input, merged
1392/// on purpose and never deserialised into `Config`.
1393fn array_keys(table: &toml::value::Table, prefix: &str) -> Vec<String> {
1394 let mut out = Vec::new();
1395 for (k, v) in table {
1396 if prefix.is_empty() && k == "vars" {
1397 continue;
1398 }
1399 let path = if prefix.is_empty() {
1400 k.clone()
1401 } else {
1402 format!("{prefix}.{k}")
1403 };
1404 match v {
1405 toml::Value::Array(_) => out.push(path),
1406 toml::Value::Table(t) => out.extend(array_keys(t, &path)),
1407 _ => {}
1408 }
1409 }
1410 out
1411}
1412
1413/// How an array key behaves when two config layers both declare it.
1414#[derive(Debug, Clone, Copy, PartialEq, Eq)]
1415enum ArrayMerge {
1416 /// Two layers may both declare it; the composed value is the
1417 /// low-to-high-priority concatenation teravars already produces (see
1418 /// [`Config::load_layers`]'s doc for why that order and no dedup).
1419 Append,
1420 /// Two layers declaring it is refused; see
1421 /// [`Config::refuse_split_arrays`].
1422 Replace,
1423}
1424
1425/// The single place that decides, for a dotted array key (as returned by
1426/// [`array_keys`]), whether declaring it in two config layers is a
1427/// concatenation the operator asked for or a silent accident.
1428///
1429/// Kept as one match so the whole policy is visible in one place - the same
1430/// reason `claude_quota` and `dropped_stream` close their own classification
1431/// in one spot elsewhere in this codebase. Anything not listed defaults to
1432/// [`ArrayMerge::Replace`]: refusing is the safe default for a key nobody has
1433/// reasoned about yet, and a new array key added later has to be added here
1434/// deliberately to become appendable.
1435///
1436/// - `verify.e2e` / `verify.gate` — a "run all of these, all must exit 0"
1437/// gate. Concatenating two of them is exactly the checks both layers
1438/// wanted, which is what lets a common gate (e.g. `editorconfig-checker`)
1439/// live in a shared layer while a repository's own layer adds its own
1440/// command, instead of every repository copying the shared command into
1441/// its own file.
1442/// - `repos.roots` — a set of directories to scan for checkouts. A
1443/// repository adding its own root on top of the machine's is additive by
1444/// nature, not a replacement of where the machine looks; see
1445/// [`Repos::roots`].
1446///
1447/// Left on the refuse side, and why:
1448/// - `roles.implementers` / `roles.judges` / `roles.reviewers` — an ordered
1449/// list of *seats*, not a set. A machine's two implementers plus a
1450/// repository's one is three seats nobody asked for and nobody is paying
1451/// for on purpose.
1452/// - `notify.command` — an argv. Concatenating two argvs does not produce a
1453/// program that runs; it produces `["ntfy", "publish", "curl", "-X"]`.
1454/// - `blind.strip_lines` — technically safe to concatenate (each entry is
1455/// matched as an independent substring, so a longer list only strips
1456/// *more*), but left on the refuse side anyway: the same list also drives
1457/// `commit_msg_hook`'s generated `sed` addresses, where position matters,
1458/// and a silent three-layer merge is exactly the kind of surprise
1459/// `refuse_split_arrays` exists to catch rather than to reason about
1460/// case-by-case. A repository that wants one more stripped phrase restates
1461/// the whole list; that restatement is visible in review, an accidental
1462/// concatenation would not be.
1463fn array_merge_policy(key: &str) -> ArrayMerge {
1464 match key {
1465 "verify.e2e" | "verify.gate" | "verify.pre_gate" | "repos.roots" => ArrayMerge::Append,
1466 _ => ArrayMerge::Replace,
1467 }
1468}
1469
1470impl Config {
1471 /// Load one file through teravars: Tera rendering, `[vars]` resolution,
1472 /// and the `include = [...]` directive.
1473 pub fn load(path: &Path) -> Result<Self> {
1474 Self::load_layers(&[path.to_path_buf()])
1475 }
1476
1477 /// The Tera render context shared by every layer: `system.*` (from
1478 /// teravars), `env` (magi's own addition - a config that names a shared
1479 /// build-cache directory or a machine-specific path needs
1480 /// `{{ env.NAME | default(value='...') }}`), and `repo` / `repo_name`
1481 /// derived from the last (highest-priority) path's parent directory.
1482 ///
1483 /// Factored out so [`Config::array_provenance`] can re-render a single
1484 /// layer under the exact same context [`Config::load_layers`] uses for
1485 /// the joint render, rather than drifting from it by accident.
1486 fn render_ctx(paths: &[PathBuf]) -> teravars::Context {
1487 let mut ctx = teravars::system_context();
1488 let env: std::collections::BTreeMap<String, String> = std::env::vars().collect();
1489 ctx.insert("env", &env);
1490 if let Some(last) = paths.last()
1491 && let Some(dir) = last.parent()
1492 {
1493 // A layer extracted from `<remote>/<base>` records the checkout it
1494 // stands for beside itself; its own directory is a cache path.
1495 let dir = std::fs::read_to_string(dir.join(REPO_MARKER))
1496 .map(|s| PathBuf::from(s.trim()))
1497 .unwrap_or_else(|_| dir.to_path_buf());
1498 ctx.insert("repo", &dir.to_string_lossy());
1499 ctx.insert(
1500 "repo_name",
1501 &dir.file_name().unwrap_or_default().to_string_lossy(),
1502 );
1503 }
1504 ctx
1505 }
1506
1507 /// Load and deep-merge a stack of config files, later files winning.
1508 ///
1509 /// This is why the config is TOML-through-teravars rather than plain serde:
1510 /// the roster is a *machine* fact (which CLIs and plans you pay for) while
1511 /// the gate is a *repository* fact (`cargo make check` here, `pnpm test`
1512 /// there). Picking one file and ignoring the other would force every repo
1513 /// to restate the roster.
1514 pub fn load_layers(paths: &[PathBuf]) -> Result<Self> {
1515 let mut engine = teravars::Engine::default();
1516 let ctx = Self::render_ctx(paths);
1517 if paths.len() > 1 {
1518 Self::refuse_split_arrays(paths, &mut engine, &ctx)?;
1519 }
1520 let merged = teravars::load_merged(paths, &mut engine, &ctx).with_context(|| {
1521 format!(
1522 "rendering config via teravars: {}",
1523 paths
1524 .iter()
1525 .map(|p| p.display().to_string())
1526 .collect::<Vec<_>>()
1527 .join(", ")
1528 )
1529 })?;
1530 let mut table = merged.config;
1531 Self::normalize_renamed_keys(paths, &mut engine, &ctx, &mut table)?;
1532 // `[vars]` is teravars' own input, already resolved into the render
1533 // context; `deny_unknown_fields` must not trip over it.
1534 table.remove("vars");
1535 // No layer says anything about `agents`: the roster is whatever is on
1536 // `PATH` now, exactly as when no layer exists at all. An explicit
1537 // `agents = []` is a statement and is kept.
1538 let detect_agents = !table.contains_key("agents");
1539 let mut cfg: Self = toml::Value::Table(table)
1540 .try_into()
1541 .context("deserializing magi config")?;
1542 crate::persona::validate(&cfg.talk.personas)?;
1543 cfg.worktree.validate()?;
1544 if let Some(name) = cfg.talk.operator_name() {
1545 if name.chars().count() > OPERATOR_NAME_MAX || name.chars().any(char::is_control) {
1546 bail!(
1547 "[talk] operator_name must be one line of at most {OPERATOR_NAME_MAX} characters"
1548 );
1549 }
1550 }
1551 if detect_agents {
1552 cfg.agents = Self::autodetected().agents;
1553 }
1554 Ok(cfg)
1555 }
1556
1557 /// `graph.candidates` was renamed `graph.implementers`. Both spellings are
1558 /// read, but the layers are merged into one table before serde sees it, so
1559 /// an old name in a lower layer and the new name in a higher one would
1560 /// arrive together and fail as a duplicate field instead of letting the
1561 /// higher layer win. Resolve it here: a single layer that writes both is
1562 /// refused (naming both keys and the file); otherwise the value from the
1563 /// last layer that wrote either key becomes `implementers`.
1564 fn normalize_renamed_keys(
1565 paths: &[PathBuf],
1566 engine: &mut teravars::Engine,
1567 ctx: &teravars::Context,
1568 table: &mut toml::map::Map<String, toml::Value>,
1569 ) -> Result<()> {
1570 let both = matches!(
1571 table.get("graph"),
1572 Some(toml::Value::Table(g)) if g.contains_key("candidates") && g.contains_key("implementers")
1573 );
1574 if !both {
1575 return Ok(());
1576 }
1577 let mut winner: Option<toml::Value> = None;
1578 for path in paths {
1579 let one = teravars::load_merged([path], engine, ctx)
1580 .with_context(|| format!("rendering {}", path.display()))?;
1581 let Some(toml::Value::Table(g)) = one.config.get("graph") else {
1582 continue;
1583 };
1584 if g.contains_key("candidates") && g.contains_key("implementers") {
1585 bail!(
1586 "`[graph]` sets both `implementers` and its deprecated alias \n `candidates` in {}; keep only `implementers`.",
1587 path.display()
1588 );
1589 }
1590 if let Some(v) = g.get("implementers").or_else(|| g.get("candidates")) {
1591 winner = Some(v.clone());
1592 }
1593 }
1594 if let Some(toml::Value::Table(g)) = table.get_mut("graph") {
1595 g.remove("candidates");
1596 if let Some(v) = winner {
1597 g.insert("implementers".to_owned(), v);
1598 }
1599 }
1600 Ok(())
1601 }
1602
1603 /// Refuse an array that two layers both declare, unless
1604 /// [`array_merge_policy`] says that key is meant to accumulate.
1605 ///
1606 /// teravars **appends** arrays when it merges layers, and that is wrong for
1607 /// most arrays magi has: `implementers` is an ordered list of seats,
1608 /// `notify.command` is an argv. Concatenating two of them yields something
1609 /// nobody wrote - three implementers out of a machine's two and a
1610 /// repository's one, or an argv of `["ntfy", "publish", "curl", "-X"]`.
1611 ///
1612 /// Replacing instead would be the right merge rule for those keys, but the
1613 /// rule lives in teravars, which several other projects depend on;
1614 /// changing it there is a decision for that crate, not something to fake
1615 /// here by re-reading the files with different semantics and hoping the
1616 /// two paths agree.
1617 ///
1618 /// So magi refuses the ambiguity rather than resolving it silently, for
1619 /// every array key except the short, deliberate list
1620 /// [`array_merge_policy`] marks [`ArrayMerge::Append`] - for those, the
1621 /// concatenation teravars already produces *is* what both files say, so
1622 /// there is nothing to refuse. The cost of guessing wrong on the refused
1623 /// keys is a roster the operator did not ask for and is paying for by the
1624 /// token; the append keys carry no such risk because every element runs
1625 /// (or every directory is scanned) regardless of order.
1626 fn refuse_split_arrays(
1627 paths: &[PathBuf],
1628 engine: &mut teravars::Engine,
1629 ctx: &teravars::Context,
1630 ) -> Result<()> {
1631 let mut seen: std::collections::BTreeMap<String, PathBuf> = Default::default();
1632 for path in paths {
1633 let one = teravars::load_merged([path], engine, ctx)
1634 .with_context(|| format!("rendering {}", path.display()))?;
1635 for key in array_keys(&one.config, "") {
1636 if array_merge_policy(&key) == ArrayMerge::Append {
1637 continue;
1638 }
1639 if let Some(first) = seen.get(&key) {
1640 bail!(
1641 "`{key}` is an array declared in two config layers:\n \
1642 {}\n {}\nteravars appends arrays when it merges, so \
1643 magi would run the concatenation of both - which is \
1644 not what either file says. Declare `{key}` in exactly \
1645 one of them.",
1646 first.display(),
1647 path.display()
1648 );
1649 }
1650 seen.insert(key, path.clone());
1651 }
1652 }
1653 Ok(())
1654 }
1655
1656 /// Which layers contributed to a composed, appendable array key (e.g.
1657 /// `"verify.gate"`), in the same low-to-high-priority order
1658 /// [`Config::load_layers`] concatenates them in. Layers that do not
1659 /// declare `key` at all are omitted.
1660 ///
1661 /// This is a **display aid for `magi doctor` only.** The command list
1662 /// that actually runs always comes from the one joint
1663 /// [`teravars::load_merged`] call in `load_layers`, never from this
1664 /// function - the exact hazard [`Config::refuse_split_arrays`] warns
1665 /// about is two merge paths that might disagree, so this function must
1666 /// never become a second source of the *composed* value, only of which
1667 /// file wrote which line in it.
1668 ///
1669 /// Re-rendering each layer alone can, in principle, resolve a
1670 /// `{{ vars.x }}` differently than the joint render would, if `x` is
1671 /// defined in one layer and referenced in another - the same caveat
1672 /// `refuse_split_arrays`'s structural, key-only check already lives with.
1673 /// None of magi's own gate commands cross that line, and a doctor listing
1674 /// is read by a human who can compare it against the joint one printed
1675 /// alongside it, so this is judged worth the simplicity of not
1676 /// threading provenance through the real load path.
1677 pub fn array_provenance(paths: &[PathBuf], key: &str) -> Vec<(PathBuf, Vec<String>)> {
1678 let mut engine = teravars::Engine::default();
1679 let ctx = Self::render_ctx(paths);
1680 let mut out = Vec::new();
1681 for path in paths {
1682 let Ok(one) = teravars::load_merged([path], &mut engine, &ctx) else {
1683 continue;
1684 };
1685 let mut cur = &one.config;
1686 let mut found = None;
1687 let parts: Vec<&str> = key.split('.').collect();
1688 for (i, part) in parts.iter().enumerate() {
1689 match cur.get(*part) {
1690 Some(toml::Value::Array(a)) if i == parts.len() - 1 => {
1691 found = Some(a);
1692 break;
1693 }
1694 Some(toml::Value::Table(t)) => cur = t,
1695 _ => break,
1696 }
1697 }
1698 let Some(values) = found else { continue };
1699 let strings: Vec<String> = values
1700 .iter()
1701 .filter_map(|v| v.as_str().map(str::to_owned))
1702 .collect();
1703 if !strings.is_empty() {
1704 out.push((path.clone(), strings));
1705 }
1706 }
1707 out
1708 }
1709
1710 /// Each layer rendered on its own, with the outcome per file: the
1711 /// settings screen's way of saying which layer a value came from and which
1712 /// layer is the one that does not parse. A **display aid only**, for the
1713 /// same reason as [`Config::array_provenance`]: the effective config
1714 /// always comes from the one joint [`Config::load_layers`].
1715 pub(crate) fn layer_tables(paths: &[PathBuf]) -> Vec<(PathBuf, Result<toml::Table>)> {
1716 let mut engine = teravars::Engine::default();
1717 let ctx = Self::render_ctx(paths);
1718 paths
1719 .iter()
1720 .map(|path| {
1721 let one = teravars::load_merged([path], &mut engine, &ctx)
1722 .map(|m| m.config)
1723 .with_context(|| format!("rendering {}", path.display()));
1724 (path.clone(), one)
1725 })
1726 .collect()
1727 }
1728
1729 /// Render a composed command list for `magi doctor`: the joined command
1730 /// line the run actually uses, plus - only when more than one layer
1731 /// contributed - which layer wrote which line.
1732 ///
1733 /// A single contributing layer (the common case today) stays the plain
1734 /// one-line summary magi has always printed, `empty` included: that
1735 /// honest "(none — ...)" is what caught a real gate-composition gap
1736 /// before this array could compose at all, and composition should not
1737 /// make the common case noisier.
1738 pub fn describe_composed(
1739 paths: &[PathBuf],
1740 commands: &[String],
1741 key: &str,
1742 empty: &str,
1743 ) -> String {
1744 if commands.is_empty() {
1745 return empty.to_owned();
1746 }
1747 let joined = commands.join(" && ");
1748 let provenance = Self::array_provenance(paths, key);
1749 if provenance.len() <= 1 {
1750 return joined;
1751 }
1752 let mut out = joined;
1753 for (path, cmds) in &provenance {
1754 out.push_str(&format!("\n [{}] {}", path.display(), cmds.join(" && ")));
1755 }
1756 out
1757 }
1758
1759 /// Resolve the config for `repo`, honouring an explicit `--config` path.
1760 ///
1761 /// Returns the config and the layers it came from, empty for built-in
1762 /// defaults. The repository layers come from the remote-tracking ref as it
1763 /// stands (see the module doc); use [`Config::discover_fetched`] where a
1764 /// stale ref would matter.
1765 pub fn discover(repo: &Path, explicit: Option<&Path>) -> Result<(Self, Vec<PathBuf>)> {
1766 #[cfg(test)]
1767 crate::web::request_cost::discovery();
1768 if let Some(p) = explicit {
1769 let paths = vec![p.to_path_buf()];
1770 return Ok((Self::load_layers(&paths)?, paths));
1771 }
1772 let paths = Self::layers_checked(repo)?;
1773 if paths.is_empty() {
1774 return Ok((Self::autodetected(), paths));
1775 }
1776 Ok((Self::load_layers(&paths)?, paths))
1777 }
1778
1779 /// [`Config::discover`] after `git fetch <remote> <base>`: for every path
1780 /// that is about to start work. A failed fetch is an error.
1781 pub async fn discover_fetched(
1782 repo: &Path,
1783 explicit: Option<&Path>,
1784 ) -> Result<(Self, Vec<PathBuf>)> {
1785 if explicit.is_none() && is_git_worktree(repo) {
1786 // A checkout that was never cloned from the remote has no
1787 // `<remote>/HEAD`; learn it here (refs/remotes only) so an
1788 // ordinary start recovers instead of failing in `discover`.
1789 let (remote, cfg_base) = probe_remote_base(repo);
1790 let base = crate::git::merge_base_branch(repo, &remote, cfg_base.as_deref()).await?;
1791 let out = crate::git::fetch(repo, &remote, &base)
1792 .await
1793 .with_context(|| format!("fetching {remote}/{base} to read magi.toml"))?;
1794 if !out.ok() {
1795 bail!(
1796 "cannot read magi.toml: `git fetch {remote} {base}` failed ({}); refusing to use the local checkout's copy, which may be stale",
1797 out.stderr.lines().next().unwrap_or("").trim()
1798 );
1799 }
1800 if cfg_base.is_none() {
1801 crate::git::ensure_remote_head(repo, &remote, &base).await;
1802 }
1803 }
1804 Self::discover(repo, explicit)
1805 }
1806
1807 /// [`Config::layers`], but an unreadable remote ref is an error.
1808 fn layers_checked(repo: &Path) -> Result<Vec<PathBuf>> {
1809 if !is_git_worktree(repo) {
1810 return Ok(Self::layers(repo));
1811 }
1812 let mut paths: Vec<PathBuf> = Self::machine_layer()
1813 .into_iter()
1814 .filter(|p| p.is_file())
1815 .collect();
1816 paths.extend(Self::repo_layers(repo)?);
1817 Ok(paths)
1818 }
1819
1820 /// Describe a repository layer without directing the operator to its cache.
1821 pub(crate) fn repo_layer_description(repo: &Path, path: &Path) -> String {
1822 let cached = path.ancestors().find(|p| p.join(REPO_MARKER).is_file());
1823 if let Some(root) = cached {
1824 let relative = path.strip_prefix(root).unwrap_or(path);
1825 let reference = resolve_remote_base(repo)
1826 .map(|(remote, base)| format!("{remote}/{base}"))
1827 .unwrap_or_else(|_| "remote base".to_owned());
1828 format!(
1829 "{} at {reference} (including its includes)",
1830 repo.join(relative).display()
1831 )
1832 } else {
1833 format!("{} (including its includes)", path.display())
1834 }
1835 }
1836
1837 /// Only the repository layers (lowest first), from `<remote>/<base>` in a
1838 /// git working tree and from disk elsewhere.
1839 pub(crate) fn repo_layers(repo: &Path) -> Result<Vec<PathBuf>> {
1840 if !is_git_worktree(repo) {
1841 return Ok([
1842 repo.join(".magi").join("config.toml"),
1843 repo.join("magi.toml"),
1844 ]
1845 .into_iter()
1846 .filter(|p| p.is_file())
1847 .collect());
1848 }
1849 remote_layers(repo)
1850 }
1851
1852 /// Environment variable that relocates the machine-wide config layer.
1853 ///
1854 /// Set it to a directory and magi reads `<dir>/magi/config.toml` instead
1855 /// of the one under [`dirs::config_dir`]; set it to the empty string and
1856 /// magi reads no machine layer at all.
1857 ///
1858 /// This exists because the machine layer is otherwise unavoidable, and a
1859 /// test that builds a config fixture is not asking for the operator's
1860 /// preferences to be merged into it. Adding `[repos] roots` to the real
1861 /// machine config on a development box turned two passing tests red -
1862 /// `repos_list_returns_name_and_path_for_every_configured_root` and
1863 /// `repos_list_only_rescans_within_the_ttl_when_asked_to`, whose fixtures
1864 /// declare `[repos] roots` of their own, which [`Config::layers`] then
1865 /// found in two layers and [`Config::refuse_split_arrays`] correctly
1866 /// refused. CI never saw it: a runner has no machine config, so the suite
1867 /// was green there and red only where somebody actually uses magi.
1868 ///
1869 /// An operator gets the same escape hatch for free: a second machine
1870 /// config, or none, without moving files about.
1871 pub const CONFIG_DIR_ENV: &str = "MAGI_CONFIG_DIR";
1872
1873 /// Every config layer that applies to `repo`, in increasing precedence.
1874 ///
1875 /// The machine layer is whatever [`Config::machine_layer`] resolves to,
1876 /// which is nothing at all in a test build.
1877 pub fn layers(repo: &Path) -> Vec<PathBuf> {
1878 let mut paths = Vec::new();
1879 paths.extend(Self::machine_layer());
1880 paths.push(repo.join(".magi").join("config.toml"));
1881 paths.push(repo.join("magi.toml"));
1882 paths.retain(|p| p.is_file());
1883 paths
1884 }
1885
1886 /// The machine-wide layer's path, when there is one.
1887 ///
1888 /// **A test build has none unless it names one.** A fixture is a complete
1889 /// statement of the config under test, and the operator's own preferences
1890 /// have no business being merged into it - least of all silently, on one
1891 /// machine, in a suite that is green everywhere else.
1892 #[cfg(test)]
1893 pub(crate) fn machine_layer() -> Option<PathBuf> {
1894 std::env::var(Self::CONFIG_DIR_ENV)
1895 .ok()
1896 .filter(|dir| !dir.trim().is_empty())
1897 .map(|dir| PathBuf::from(dir).join("magi").join("config.toml"))
1898 }
1899
1900 /// The machine-wide layer's path, when there is one.
1901 #[cfg(not(test))]
1902 pub(crate) fn machine_layer() -> Option<PathBuf> {
1903 match std::env::var(Self::CONFIG_DIR_ENV) {
1904 // Named, and empty on purpose: no machine layer.
1905 Ok(dir) if dir.trim().is_empty() => None,
1906 Ok(dir) => Some(PathBuf::from(dir).join("magi").join("config.toml")),
1907 Err(_) => dirs::config_dir().map(|dir| dir.join("magi").join("config.toml")),
1908 }
1909 }
1910
1911 /// Built-in config whose roster is the agent CLIs found on `PATH`.
1912 pub fn autodetected() -> Self {
1913 let mut cfg = Self::default();
1914 for (kind, id, model) in [
1915 (AgentKind::Claude, "opus", Some("opus")),
1916 (AgentKind::Claude, "sonnet", Some("sonnet")),
1917 (AgentKind::Antigravity, "antigravity", None),
1918 (AgentKind::Opencode, "opencode", None),
1919 (AgentKind::Codex, "codex", None),
1920 (AgentKind::Omp, "omp", None),
1921 ] {
1922 if kind.program().is_some_and(which) && !cfg.agents.iter().any(|a| a.id == id) {
1923 cfg.agents.push(AgentSpec {
1924 id: id.to_owned(),
1925 kind,
1926 model: model.map(str::to_owned),
1927 command: Vec::new(),
1928 extra_args: Vec::new(),
1929 env: BTreeMap::new(),
1930 prompt_delivery: None,
1931 });
1932 }
1933 }
1934 cfg
1935 }
1936
1937 /// The shared build cache the verify commands and the agents both build
1938 /// into, when the config declares one. See [`Verify::cache_dir`].
1939 pub fn cache_dir(&self) -> Option<PathBuf> {
1940 self.verify.cache_dir()
1941 }
1942
1943 /// Look an agent up by id.
1944 pub fn agent(&self, id: &str) -> Result<&AgentSpec> {
1945 self.agents
1946 .iter()
1947 .find(|a| a.id == id)
1948 .with_context(|| format!("no agent with id `{id}` in the roster"))
1949 }
1950
1951 /// Rotate `count` seats out of `ids`, or out of the whole roster at
1952 /// `offset` when `ids` is empty.
1953 ///
1954 /// The one rotation rule - explicit ids cycle, an empty list rotates the
1955 /// roster - shared by every seat count [`Config::resolve_roles`] fills in,
1956 /// rather than each seat reimplementing it and drifting apart.
1957 fn rotate(&self, ids: &[String], count: usize, offset: usize) -> Result<Vec<AgentSpec>> {
1958 let mut out = Vec::with_capacity(count);
1959 for i in 0..count {
1960 let spec = if ids.is_empty() {
1961 self.agents[(i + offset) % self.agents.len()].clone()
1962 } else {
1963 self.agent(&ids[i % ids.len()])?.clone()
1964 };
1965 out.push(spec);
1966 }
1967 Ok(out)
1968 }
1969
1970 /// `ids`, resolved to specs in the order named, or the whole `[[agents]]`
1971 /// roster in file order when `ids` is empty — the same "explicit ids
1972 /// stand alone, an empty list means the whole roster" rule [`Self::rotate`]
1973 /// applies, but with no `count` to truncate to and no `offset` to wrap by.
1974 /// See [`ResolvedRoles::implementer_roster`] for why a truncated, rotated
1975 /// list cannot answer the question this exists for.
1976 fn full_roster(&self, ids: &[String]) -> Result<Vec<AgentSpec>> {
1977 if ids.is_empty() {
1978 Ok(self.agents.clone())
1979 } else {
1980 ids.iter().map(|id| self.agent(id).cloned()).collect()
1981 }
1982 }
1983
1984 /// Fill the roles out to the configured widths.
1985 ///
1986 /// An empty role list rotates through the whole roster, so a three-agent
1987 /// roster with `implementers = 3` gives one implementation per agent, and
1988 /// `judges = 3` rotates the judge seats by one so that judge *i* is not the
1989 /// author of candidate *i* whenever the roster has more than one agent.
1990 pub fn resolve_roles(&self) -> Result<ResolvedRoles> {
1991 if self.agents.is_empty() {
1992 bail!(
1993 "agent roster is empty: no agent CLI found on PATH and no \
1994 [[agents]] in the config. Run `magi init` to write a starter \
1995 magi.toml."
1996 );
1997 }
1998 Ok(ResolvedRoles {
1999 implementers: self.rotate(&self.roles.implementers, self.graph.implementers, 0)?,
2000 judges: self.rotate(&self.roles.judges, self.graph.judges, 1)?,
2001 reviewers: self.rotate(&self.roles.reviewers, self.graph.reviewers, 0)?,
2002 fixer: match self.roles.fixer.as_ref() {
2003 // Ids are validated against the roster here; whether the CLI
2004 // is installed is left to invocation, like every other seat.
2005 Some(choice) => Some(crate::agent::pick_chain(
2006 &self.agents,
2007 Some(choice),
2008 &|_| true,
2009 "fixer",
2010 )?),
2011 None => None,
2012 },
2013 // Role resolution validates roster shape, but deliberately does
2014 // not preflight a CLI. The other graph seats have always deferred
2015 // that failure to invocation; doing it only for the conductor
2016 // made otherwise usable graph commands and `doctor` fail as one.
2017 conductor: match self.roles.conductor.as_ref().map(AgentChoice::ids) {
2018 Some(ids) if ids.len() == 1 => self.agent(ids[0])?.clone(),
2019 // A fallback chain: the first id the roster knows. The rest
2020 // are `agent::pick_chain`'s business at invocation.
2021 Some(ids) if !ids.is_empty() => ids
2022 .iter()
2023 .find_map(|id| self.agent(id).ok())
2024 .cloned()
2025 .with_context(|| {
2026 format!("[roles] conductor names no agent in the roster: {ids:?}")
2027 })?,
2028 // Keep the normal standalone-seat preference when something
2029 // is installed, but retain a roster fallback when it is not.
2030 // Invocation then reports the unavailable CLI in the same
2031 // place it does for every other graph role.
2032 _ => crate::agent::pick(&self.agents, None, &crate::agent::installed)
2033 .unwrap_or_else(|_| self.agents[0].clone()),
2034 },
2035 implementer_roster: self.full_roster(&self.roles.implementers)?,
2036 judge_roster: self.full_roster(&self.roles.judges)?,
2037 reviewer_roster: self.full_roster(&self.roles.reviewers)?,
2038 })
2039 }
2040
2041 /// Advisor seats for the design-deliberation stage (see
2042 /// `graph::Runner::advise`): `[roles] advisors` when set, otherwise the
2043 /// judge roster - see [`Roles::advisors`] for why that fallback and not
2044 /// the whole roster.
2045 ///
2046 /// The fallback rotates with `offset = 1`, matching the judges line in
2047 /// [`Config::resolve_roles`] exactly, `ids` and offset both - not just
2048 /// `roles.judges`, which is empty whenever judges themselves are
2049 /// unconfigured and rotating the whole roster. Falling back with
2050 /// `offset = 0` there would silently hand the advisors a *different*
2051 /// agent set than the judges an unconfigured run would actually get,
2052 /// which is the one thing [`Roles::advisors`]'s doc promises will not
2053 /// happen.
2054 ///
2055 /// Called lazily from the graph node itself rather than folded into
2056 /// [`Config::resolve_roles`]: unlike the other roles, a failure here must
2057 /// not stop a run from starting at all - the deliberation stage is an
2058 /// enrichment `[graph] advise` can turn off, not a seat later nodes
2059 /// cannot proceed without - and at the point `resolve_roles` runs (before
2060 /// [`crate::run::RunState`] exists, on `Runner::start`) there would be no
2061 /// run yet for a resolution failure to be reported against.
2062 pub fn advisors(&self) -> Result<Vec<AgentSpec>> {
2063 if self.agents.is_empty() {
2064 bail!(
2065 "agent roster is empty: no agent CLI found on PATH and no \
2066 [[agents]] in the config. Run `magi init` to write a starter \
2067 magi.toml."
2068 );
2069 }
2070 if !self.roles.advisors.is_empty() {
2071 return self.rotate(&self.roles.advisors, self.graph.advisors, 0);
2072 }
2073 self.rotate(&self.roles.judges, self.graph.advisors, 1)
2074 }
2075
2076 /// The untruncated roster advisor seats hand over along: `[roles]
2077 /// advisors` when set, otherwise the judges' roster, exactly as
2078 /// [`Self::advisors`] chooses its own list.
2079 pub fn advisor_roster(&self) -> Result<Vec<AgentSpec>> {
2080 if self.roles.advisors.is_empty() {
2081 self.full_roster(&self.roles.judges)
2082 } else {
2083 self.full_roster(&self.roles.advisors)
2084 }
2085 }
2086
2087 /// Shell prefix for [`Verify`] commands.
2088 pub fn shell(&self) -> Vec<String> {
2089 if let Some(s) = &self.verify.shell {
2090 return s.clone();
2091 }
2092 if which("sh") {
2093 vec!["sh".to_owned(), "-c".to_owned()]
2094 } else {
2095 vec!["cmd".to_owned(), "/C".to_owned()]
2096 }
2097 }
2098
2099 /// Starter config, as written by `magi init`.
2100 pub fn starter_toml() -> String {
2101 let detected = Self::autodetected();
2102 let mut s = String::from(
2103 "# magi — blind multi-agent implementation competition.\n\
2104 # `magi run \"<task>\"` walks: implement (N parallel worktrees)\n\
2105 # -> blind judging -> deliberation -> private final vote\n\
2106 # -> fold losers -> review + E2E loop -> gate -> merge.\n\
2107 #\n\
2108 # Rendered by teravars: a `[vars]` table, env\n\
2109 # and system lookups, and `include = [...]` all work. Tera\n\
2110 # braces are live everywhere in this file, but comments are\n\
2111 # stripped before rendering (teravars >= 0.2.2), so a comment\n\
2112 # may quote `{{ ... }}` freely.\n\
2113 #\n\
2114 # Layers deep-merge in increasing\n\
2115 # precedence, so the roster can live once per machine in\n\
2116 # <config_dir>/magi/config.toml and each repo only states its own\n\
2117 # gate:\n\
2118 # <config_dir>/magi/config.toml < .magi/config.toml < magi.toml\n\n\
2119 [vars]\n\
2120 # Reference it as vars.cache inside Tera braces, anywhere below.\n\
2121 # Single quotes inside the braces: teravars renders the raw file\n\
2122 # text, so TOML's own \\\" escaping never reaches Tera.\n\
2123 cache = \"{{ env.MAGI_CACHE | default(value='/tmp') }}\"\n\n",
2124 );
2125 if detected.agents.is_empty() {
2126 s.push_str(
2127 "# No agent CLI was found on PATH. Fill this in by hand.\n\
2128 # kind = claude | opencode | antigravity | codex | command\n\
2129 [[agents]]\nid = \"opus\"\nkind = \"claude\"\nmodel = \"opus\"\n\n",
2130 );
2131 } else {
2132 for a in &detected.agents {
2133 s.push_str("[[agents]]\n");
2134 s.push_str(&format!("id = {:?}\n", a.id));
2135 s.push_str(&format!("kind = {:?}\n", a.kind.as_str()));
2136 if let Some(m) = &a.model {
2137 s.push_str(&format!("model = {m:?}\n"));
2138 }
2139 s.push('\n');
2140 }
2141 }
2142 s.push_str(
2143 "# Leave a role list empty to rotate through the roster.\n\
2144 # A non-empty list is also an ordered backup list: seats take entries from the\n\
2145 # front (see [graph] counts) and a failed or out-of-quota seat moves on to the\n\
2146 # next agent to its right, each tried once. Same for judges and reviewers.\n\
2147 # implementers = [\"opus\", \"codex\", \"sonnet\"] # with implementers = 1 below: opus sits, codex then sonnet are spares\n\
2148 [roles]\n\
2149 implementers = []\n\
2150 judges = []\n\
2151 reviewers = []\n\
2152 # conductor = \"opus\" # arranges the queue; unset picks a seat like chatter does\n\
2153 # synthesizer = \"opus\" # blends the advisors into one brief; unset picks a seat like chatter does\n\
2154 # synthesizer = [\"opus\", \"codex\"] # array form: fallback chain, each tried once on quota or failure\n\
2155 # language_judge = \"jev\" # kind = \"command\" wrapper deciding if PR text is English; unset keeps the built-in heuristics\n\
2156 # dupes_judge = \"opus\" # judges duplicate-work claims; unset inherits chatter (string or fallback chain)\n\
2157 # fixer = \"opus\" # applies review findings; unset keeps the winner's own author\n\
2158 # fixer = [\"opus\", \"codex\"] # array form: fallback chain, each tried once on quota or failure\n\n\
2159 [graph]\n\
2160 # Gate GitHub prose; secret redaction is always enabled.\n\
2161 github_text_guard = true\n\
2162 implementers = 3\n\
2163 judges = 3\n\
2164 deliberate_rounds = 1\n\
2165 reviewers = 3\n\
2166 review_rounds = 6\n\
2167 # Fix rounds a failing verify.gate gets before the run is blocked (0 = none).\n\
2168 gate_fix_rounds = 1\n\
2169 # Rerun a failed required CI check once before a land fix round (flaky tests).\n\
2170 rerun_flaky_checks = true\n\
2171 # Port the strengths judges named in the losing candidates onto the\n\
2172 # winner before review (best-effort; rolled back on failure or a big diff).\n\
2173 graft_runner_up = true\n\
2174 max_parallel = 4\n\
2175 language = \"en\"\n\
2176 # One CLI conversation per seat: judges keep their own argument\n\
2177 # across deliberation, the fixer keeps its implementation context.\n\
2178 sessions = true\n\
2179 # Reviewer-seat timeout. When timeout_verify is omitted, E2E and\n\
2180 # the final gate inherit this value for compatibility.\n\
2181 timeout_review = 1200\n\
2182 # Optional independent E2E/final-gate timeout; uncomment to keep\n\
2183 # verification independent if timeout_review changes later.\n\
2184 # timeout_verify = 1200\n\
2185 # Longest a seat waits for the shared build cache before it is\n\
2186 # spawned. Never counted against an agent's own timeout; expiry\n\
2187 # holds the run (resumable) instead of failing the agent.\n\
2188 # timeout_resource_wait = 1800\n\n\
2189 # What the chat calls you (persona address and the Chat's turn label).\n\
2190 # [talk]\n\
2191 # operator_name = \"Commander\"\n\n\
2192 # Chat personas (tone only; the web Chat picks one per conversation).\n\
2193 # Built in: magi, rei, misato, ritsuko, shinji, asuka, kaworu. Entries add to\n\
2194 # them, and one with a built-in's id replaces it. `default` is the plain\n\
2195 # voice and cannot be redefined (it may only set `icon`). Declare this in one\n\
2196 # layer only. `icon` is an optional avatar, display only: an absolute or\n\
2197 # ~/ path to a png/jpg/gif/webp, or a data:image/...;base64 URI, up to\n\
2198 # 256 KiB; a missing or bad one falls back to the built-in avatar.\n\
2199 # [[talk.personas]]\n\
2200 # id = \"gendo\"\n\
2201 # name = \"Gendo Ikari\"\n\
2202 # prompt = \"Speak coldly and tersely, hands folded.\"\n\
2203 # icon = \"~/pictures/gendo.png\"\n\n\
2204 [verify]\n\
2205 # Run once per review round in the winner's worktree; failures are\n\
2206 # fed back to the fixer.\n\
2207 e2e = []\n\
2208 # Final gate. Every command must exit 0 before a merge.\n\
2209 gate = []\n\
2210 # Mechanical fixers (formatters) run in the winner's worktree just\n\
2211 # before the gate; whatever they change becomes one commit. A failing\n\
2212 # command only warns - the gate stays the arbiter. Same timeout as the\n\
2213 # gate (timeout_verify). Keep it light and idempotent.\n\
2214 # pre_gate = []\n\n\
2215 # Prepare each worktree magi creates (candidates, judges, advisors,\n\
2216 # reviewers, the operator-fix and review-takeover checkouts), in order,\n\
2217 # before any agent or verify command touches it. `copy` reads from the\n\
2218 # primary checkout and never overwrites; `optional` skips a missing\n\
2219 # source. `run` goes through the shell with the worktree as cwd and\n\
2220 # fails the seat on a non-zero exit or timeout (timeout_secs, default\n\
2221 # the verify timeout). Setup must not edit tracked files, and what it\n\
2222 # leaves untracked is kept out of magi's commits. Declare this in one\n\
2223 # layer only.\n\
2224 # [worktree]\n\
2225 # setup = [\n\
2226 # { copy = \".env.example -> .env\" },\n\
2227 # { copy = \".env.local\", optional = true },\n\
2228 # { run = \"pnpm install --frozen-lockfile\", timeout_secs = 900 },\n\
2229 # ]\n\n\
2230 [merge]\n\
2231 # none | local | pr\n\
2232 mode = \"none\"\n\
2233 # release_bump = true # Rust (Cargo.toml) repositories only; a no-op elsewhere\n\n\
2234 [update]\n\
2235 # off | notify | install — checked in the background, throttled.\n\
2236 mode = \"notify\"\n\
2237 # interval = \"24h\"\n\n\
2238 [daemon]\n\
2239 # Minutes a release-bump pull request (chore/release-v*) may sit with\n\
2240 # no new commit and no check changing state before `magi serve`\n\
2241 # notices and asks you. A failed check is rerun once on its own;\n\
2242 # still red after that escalates at once. 0 turns the watcher off.\n\
2243 # release_stall_minutes = 60\n\n\
2244 [release]\n\
2245 # actions (default) | local. \"local\" is for repositories whose GitHub\n\
2246 # Actions never run (private, no minutes): the release pull request\n\
2247 # does not wait for CI, you approve the merge in magi, and then\n\
2248 # `magi serve` pushes the vX.Y.Z tag (never over an existing one) and\n\
2249 # runs `commands` from a clean checkout of the merge commit. A\n\
2250 # failure holds the release with a notice; it is never retried blind.\n\
2251 # Needs `magi serve` running and [daemon] release_stall_minutes > 0.\n\
2252 # mode = \"actions\"\n\
2253 # Run in order; the first failure stops. They see MAGI_RELEASE_VERSION,\n\
2254 # MAGI_RELEASE_TAG, MAGI_RELEASE_COMMIT and MAGI_RELEASE_PR. Make them\n\
2255 # safe to repeat: a crash mid-command is treated as unfinished.\n\
2256 # commands = [\"cargo build --release\", \"gh release create \\\"$MAGI_RELEASE_TAG\\\" --generate-notes\"]\n\
2257 # timeout_minutes = 30\n",
2258 );
2259 s
2260 }
2261}
2262
2263/// Is `program` on `PATH`?
2264pub fn which(program: &str) -> bool {
2265 find_program(program).is_some()
2266}
2267
2268/// The file a bare `program` name resolves to on `PATH`, the way the OS shell
2269/// would find it.
2270///
2271/// Spawn agents through this, not by bare name: on Windows Rust's std
2272/// only tries `name.exe`, so a CLI installed as an npm `.cmd` shim (codex
2273/// since its 2026-09-23 reinstall) is "not found" even though a shell runs it.
2274/// On Windows an extensionless file is never a match: npm writes an `sh`
2275/// script beside every `.cmd` shim, and spawning that fails with os error 193.
2276pub fn find_program(program: &str) -> Option<PathBuf> {
2277 find_program_on(program, &std::env::var_os("PATH")?)
2278}
2279
2280/// [`find_program`] against an explicit `PATH`, for a child whose environment
2281/// overrides it.
2282pub fn find_program_on(program: &str, paths: &std::ffi::OsStr) -> Option<PathBuf> {
2283 let pathext = std::env::var("PATHEXT").unwrap_or_else(|_| ".EXE;.CMD;.BAT".into());
2284 find_program_in(program, paths, cfg!(windows).then_some(pathext.as_str()))
2285}
2286
2287fn find_program_in(
2288 program: &str,
2289 paths: &std::ffi::OsStr,
2290 pathext: Option<&str>,
2291) -> Option<PathBuf> {
2292 let p = Path::new(program);
2293 if p.components().count() > 1 {
2294 return p.is_file().then(|| p.to_path_buf());
2295 }
2296 std::env::split_paths(paths).find_map(|dir| match pathext {
2297 Some(exts) => {
2298 let bare = dir.join(program);
2299 if p.extension().is_some() && bare.is_file() {
2300 return Some(bare);
2301 }
2302 exts.split(';')
2303 .filter(|e| !e.is_empty())
2304 .map(|e| dir.join(format!("{program}{}", e.to_lowercase())))
2305 .find(|c| c.is_file())
2306 }
2307 None => Some(dir.join(program)).filter(|c| c.is_file()),
2308 })
2309}
2310
2311/// Sidecar naming the checkout an extracted layer stands for.
2312const REPO_MARKER: &str = ".magi-repo";
2313
2314fn git_out(repo: &Path, args: &[&str]) -> Option<String> {
2315 use crate::proc::Quiet as _;
2316 #[cfg(test)]
2317 crate::web::request_cost::git();
2318 let out = std::process::Command::new("git")
2319 .args(args)
2320 .quiet()
2321 .current_dir(repo)
2322 .stdin(std::process::Stdio::null())
2323 .output()
2324 .ok()?;
2325 out.status
2326 .success()
2327 .then(|| String::from_utf8_lossy(&out.stdout).into_owned())
2328}
2329
2330/// Is `repo` inside a git working tree? A directory that is not has no branch
2331/// to be stale against, so its files are read as they are.
2332fn is_git_worktree(repo: &Path) -> bool {
2333 git_out(repo, &["rev-parse", "--git-dir"]).is_some()
2334}
2335
2336#[derive(Deserialize, Default)]
2337struct ProbeMerge {
2338 base: Option<String>,
2339 remote: Option<String>,
2340}
2341
2342#[derive(Deserialize, Default)]
2343struct Probe {
2344 merge: Option<ProbeMerge>,
2345}
2346
2347/// `[merge]` base / remote from one file, nothing else. The file is rendered
2348/// (env templates, includes) exactly as a real load would, so a machine layer
2349/// saying `base = "{{ env.X }}"` locates the same ref; when rendering fails
2350/// (a checkout file that needs its own repo context) the plain TOML is read
2351/// instead. Unreadable is `None`: the real load reports it.
2352fn probe_merge(path: &Path, repo: &Path) -> Option<ProbeMerge> {
2353 if !path.is_file() {
2354 return None;
2355 }
2356 let paths = [path.to_path_buf()];
2357 let mut engine = teravars::Engine::default();
2358 let mut ctx = Config::render_ctx(&paths);
2359 // The layered load renders every layer for the checkout being resolved;
2360 // so does this, or `{{ repo_name }}` would name the file's own directory.
2361 ctx.insert("repo", &repo.to_string_lossy());
2362 ctx.insert(
2363 "repo_name",
2364 &repo.file_name().unwrap_or_default().to_string_lossy(),
2365 );
2366 if let Ok(merged) = teravars::load_merged(&paths, &mut engine, &ctx)
2367 && let Ok(probe) = toml::Value::Table(merged.config).try_into::<Probe>()
2368 {
2369 return probe.merge;
2370 }
2371 let text = std::fs::read_to_string(path).ok()?;
2372 toml::from_str::<Probe>(&text).ok()?.merge
2373}
2374
2375/// The `[merge] remote` / `base` the layers name, before any git lookup.
2376fn probe_remote_base(repo: &Path) -> (String, Option<String>) {
2377 let mut remote = None;
2378 let mut base = None;
2379 let mut sources: Vec<PathBuf> = Config::machine_layer().into_iter().collect();
2380 sources.push(repo.join("magi.toml"));
2381 sources.push(repo.join(".magi").join("config.toml"));
2382 for path in sources {
2383 if let Some(m) = probe_merge(&path, repo) {
2384 remote = remote.or(m.remote.filter(|s| !s.trim().is_empty()));
2385 base = base.or(m.base.filter(|s| !s.trim().is_empty()));
2386 }
2387 }
2388 (remote.unwrap_or_else(|| "origin".to_owned()), base)
2389}
2390
2391/// Where the remote ref is: see the module doc for the order.
2392fn resolve_remote_base(repo: &Path) -> Result<(String, String)> {
2393 let (remote, base) = probe_remote_base(repo);
2394 let base = match base {
2395 Some(b) => b,
2396 None => {
2397 let head = format!("refs/remotes/{remote}/HEAD");
2398 git_out(repo, &["symbolic-ref", "--quiet", &head])
2399 .and_then(|s| crate::git::remote_head_branch(&remote, &s))
2400 .with_context(|| {
2401 format!(
2402 "cannot tell which branch of `{remote}` holds magi.toml: \
2403 set [merge] base, or run `git remote set-head {remote} -a`"
2404 )
2405 })?
2406 }
2407 };
2408 Ok((remote, base))
2409}
2410
2411fn fnv_hex(text: &str) -> String {
2412 let mut hash: u64 = 0xcbf2_9ce4_8422_2325;
2413 for b in text.bytes() {
2414 hash ^= u64::from(b);
2415 hash = hash.wrapping_mul(0x0100_0000_01b3);
2416 }
2417 format!("{hash:016x}")
2418}
2419
2420/// Does the text declare an `include` directive in any form teravars accepts
2421/// (root `include`, bare or quoted, or `[teravars] include`, as a table or an
2422/// inline table)? Parsed as TOML when it parses; a templated file that does
2423/// not parse is judged by a plain search for the word, because over-detecting
2424/// only costs an extra extraction while missing one fails the load.
2425fn declares_include(text: &str) -> bool {
2426 match toml::from_str::<toml::Table>(text) {
2427 Ok(table) => {
2428 table.contains_key("include")
2429 || table
2430 .get("teravars")
2431 .and_then(toml::Value::as_table)
2432 .is_some_and(|t| t.contains_key("include"))
2433 }
2434 Err(_) => text.contains("include"),
2435 }
2436}
2437
2438/// Unpack the whole tree of `sha` into `dest`, so relative `include`s of a
2439/// repository layer resolve against the same commit.
2440fn extract_tree(repo: &Path, sha: &str, dest: &Path) -> Result<()> {
2441 use crate::proc::Quiet as _;
2442 #[cfg(test)]
2443 crate::web::request_cost::git();
2444 let mut archive = std::process::Command::new("git")
2445 .args(["archive", "--format=tar", sha])
2446 .quiet()
2447 .current_dir(repo)
2448 .stdin(std::process::Stdio::null())
2449 .stdout(std::process::Stdio::piped())
2450 .stderr(std::process::Stdio::null())
2451 .spawn()
2452 .context("running git archive")?;
2453 let tar = std::process::Command::new("tar")
2454 .arg("-x")
2455 .arg("-C")
2456 .arg(dest)
2457 .quiet()
2458 .stdin(archive.stdout.take().context("git archive has no stdout")?)
2459 .output()
2460 .context("running tar")?;
2461 let status = archive.wait().context("waiting for git archive")?;
2462 if !status.success() || !tar.status.success() {
2463 bail!(
2464 "cannot unpack {sha} for the config's includes: {}",
2465 String::from_utf8_lossy(&tar.stderr).trim()
2466 );
2467 }
2468 Ok(())
2469}
2470
2471/// The repository layers (`.magi/config.toml`, `magi.toml`; lowest first) of
2472/// `<remote>/<base>`, extracted from one commit so both files agree.
2473fn remote_layers(repo: &Path) -> Result<Vec<PathBuf>> {
2474 let (remote, base) = resolve_remote_base(repo)?;
2475 let tracking = format!("refs/remotes/{remote}/{base}");
2476 let spec = format!("{tracking}^{{commit}}");
2477 let sha = git_out(repo, &["rev-parse", "--verify", "--quiet", &spec])
2478 .map(|s| s.trim().to_owned())
2479 .filter(|s| !s.is_empty())
2480 .with_context(|| {
2481 format!(
2482 "`{remote}/{base}` does not exist in this checkout; fetch it \
2483 (`git fetch {remote} {base}`) or set [merge] base/remote. magi will not read \
2484 a possibly stale local magi.toml instead"
2485 )
2486 })?;
2487 let key = format!("{sha}-{}", fnv_hex(&repo.to_string_lossy()));
2488 let rels = [".magi/config.toml", "magi.toml"];
2489 let mut texts: Vec<(&str, String)> = Vec::new();
2490 for rel in rels {
2491 let object = format!("{sha}:{rel}");
2492 if git_out(repo, &["cat-file", "-e", &object]).is_none() {
2493 continue;
2494 }
2495 let Some(text) = git_out(repo, &["show", &object]) else {
2496 bail!("cannot read {rel} from {remote}/{base} ({sha})");
2497 };
2498 texts.push((rel, text));
2499 }
2500 let wants_tree = texts.iter().any(|(_, t)| declares_include(t));
2501 // Content-addressed and built beside the final name, then renamed in: a
2502 // concurrent `discover` sees the whole directory or none of it, never a
2503 // half-written file.
2504 let build = |root: &Path| -> Result<PathBuf> {
2505 let final_dir = root.join(&key);
2506 if !final_dir.is_dir() {
2507 std::fs::create_dir_all(root)?;
2508 let tmp = root.join(format!(
2509 "{key}.tmp-{}-{}",
2510 std::process::id(),
2511 std::time::SystemTime::now()
2512 .duration_since(std::time::UNIX_EPOCH)
2513 .map(|d| d.as_nanos())
2514 .unwrap_or_default()
2515 ));
2516 let filled = (|| -> Result<()> {
2517 std::fs::create_dir_all(&tmp)?;
2518 if wants_tree {
2519 extract_tree(repo, &sha, &tmp)?;
2520 }
2521 for (rel, text) in &texts {
2522 let file = tmp.join(rel);
2523 let parent = file.parent().unwrap_or(&tmp);
2524 std::fs::create_dir_all(parent)?;
2525 std::fs::write(&file, text)?;
2526 std::fs::write(parent.join(REPO_MARKER), repo.to_string_lossy().as_bytes())?;
2527 }
2528 Ok(())
2529 })();
2530 if let Err(e) = filled {
2531 let _ = std::fs::remove_dir_all(&tmp);
2532 return Err(e);
2533 }
2534 if std::fs::rename(&tmp, &final_dir).is_err() {
2535 // Lost the race to another process building the same commit.
2536 let _ = std::fs::remove_dir_all(&tmp);
2537 if !final_dir.is_dir() {
2538 bail!("cannot place {}", final_dir.display());
2539 }
2540 }
2541 }
2542 Ok(final_dir)
2543 };
2544 let common = git_out(
2545 repo,
2546 &["rev-parse", "--path-format=absolute", "--git-common-dir"],
2547 )
2548 .map(|s| PathBuf::from(s.trim()).join("magi").join("config"));
2549 let dir = match common.map(|c| build(&c)) {
2550 Some(Ok(d)) => d,
2551 _ => build(&std::env::temp_dir().join("magi-config"))
2552 .with_context(|| format!("extracting config from {remote}/{base}"))?,
2553 };
2554 Ok(texts.iter().map(|(rel, _)| dir.join(rel)).collect())
2555}
2556
2557#[cfg(test)]
2558mod tests {
2559 #[test]
2560 fn context_window_prefers_exact_then_longest_prefix_and_user_over_builtin() {
2561 let mut cfg = Config::default();
2562 assert_eq!(
2563 cfg.context_window("claude-sonnet-4-5-20250929"),
2564 Some(200_000)
2565 );
2566 assert_eq!(cfg.context_window("gpt-4.1-mini"), Some(1_000_000));
2567 assert_eq!(cfg.context_window("mystery-model"), None);
2568 cfg.context_windows
2569 .insert("claude-sonnet-4-5".into(), 1_000_000);
2570 cfg.context_windows.insert("claude-sonnet-4".into(), 5);
2571 cfg.context_windows.insert("mystery".into(), 32_768);
2572 // Longest user prefix beats a shorter one and the built-in table.
2573 assert_eq!(
2574 cfg.context_window("claude-sonnet-4-5-20250929"),
2575 Some(1_000_000)
2576 );
2577 // An exact name beats any prefix.
2578 cfg.context_windows
2579 .insert("claude-sonnet-4-5-20250929".into(), 7);
2580 assert_eq!(cfg.context_window("claude-sonnet-4-5-20250929"), Some(7));
2581 assert_eq!(cfg.context_window("mystery-model"), Some(32_768));
2582 }
2583
2584 use super::*;
2585
2586 /// npm writes an extensionless `sh` script beside `codex.cmd`; on Windows
2587 /// the `.cmd` is the spawnable one (os error 193 otherwise, 2026-09-23).
2588 #[test]
2589 fn windows_lookup_picks_the_cmd_shim_over_the_extensionless_script() {
2590 let dir = tempfile::tempdir().unwrap();
2591 std::fs::write(dir.path().join("codex"), "#!/bin/sh\n").unwrap();
2592 std::fs::write(dir.path().join("codex.cmd"), "@echo off\n").unwrap();
2593 let paths = dir.path().as_os_str();
2594 assert_eq!(
2595 find_program_in("codex", paths, Some(".EXE;.CMD")),
2596 Some(dir.path().join("codex.cmd"))
2597 );
2598 assert_eq!(
2599 find_program_in("codex", paths, None),
2600 Some(dir.path().join("codex"))
2601 );
2602 assert_eq!(find_program_in("claude", paths, Some(".EXE;.CMD")), None);
2603 }
2604
2605 fn spec(id: &str) -> AgentSpec {
2606 AgentSpec {
2607 id: id.to_owned(),
2608 kind: AgentKind::Command,
2609 model: None,
2610 command: vec!["true".to_owned()],
2611 extra_args: Vec::new(),
2612 env: BTreeMap::new(),
2613 prompt_delivery: None,
2614 }
2615 }
2616
2617 #[test]
2618 fn github_text_guard_defaults_on_and_can_be_disabled() {
2619 let default: Graph = toml::from_str("").unwrap();
2620 assert!(default.github_text_guard);
2621 let disabled: Graph = toml::from_str("github_text_guard = false").unwrap();
2622 assert!(!disabled.github_text_guard);
2623 let config = layered("", "[graph]\ngithub_text_guard = false\n").unwrap();
2624 assert!(!config.graph.github_text_guard);
2625 assert!(Config::starter_toml().contains("github_text_guard = true"));
2626 }
2627
2628 #[test]
2629 fn rerun_flaky_checks_defaults_on_and_is_in_the_starter() {
2630 assert!(Graph::default().rerun_flaky_checks);
2631 let off: Graph = toml::from_str("rerun_flaky_checks = false").unwrap();
2632 assert!(!off.rerun_flaky_checks);
2633 assert!(Config::starter_toml().contains("rerun_flaky_checks = true"));
2634 }
2635
2636 #[test]
2637 fn timeout_verify_omitted_from_old_toml_inherits_timeout_review() {
2638 // `timeout_verify` used to not exist: `verify.e2e`/`verify.gate` ran
2639 // under `timeout_review`. A config written before this field existed
2640 // must run exactly as before, which means its default has to be the
2641 // same 1200s `timeout_review` has always defaulted to.
2642 let g: Graph = toml::from_str("timeout_review = 3600").expect("parse");
2643 assert_eq!(g.timeout_verify, None);
2644 assert_eq!(g.verify_timeout(), 3600);
2645 }
2646
2647 #[test]
2648 fn shrinking_timeout_review_does_not_shrink_timeout_verify() {
2649 // The bug this field exists to close: `[graph] timeout_review = 45`
2650 // used to shrink the real-machine `verify.e2e`/`verify.gate` budget
2651 // along with the reviewer seats' own timeout, because both read the
2652 // same field.
2653 let g: Graph = toml::from_str("timeout_review = 45").expect("parse");
2654 assert_eq!(g.timeout_review, 45);
2655 assert_eq!(
2656 g.verify_timeout(),
2657 45,
2658 "an omitted legacy value follows review"
2659 );
2660 let explicit: Graph = toml::from_str("timeout_review = 45\ntimeout_verify = 1200")
2661 .expect("parse explicit override");
2662 assert_eq!(explicit.verify_timeout(), 1200);
2663 }
2664
2665 #[test]
2666 fn graph_implementers_new_key_loads_and_old_key_is_an_alias() {
2667 let g: Graph = toml::from_str("implementers = 4\n").unwrap();
2668 assert_eq!(g.implementers, 4);
2669 let g: Graph = toml::from_str("candidates = 5\n").unwrap();
2670 assert_eq!(g.implementers, 5);
2671 }
2672
2673 fn layered(lower: &str, upper: &str) -> Result<Config> {
2674 let dir = tempfile::tempdir().unwrap();
2675 let a = dir.path().join("a.toml");
2676 let b = dir.path().join("b.toml");
2677 std::fs::write(&a, lower).unwrap();
2678 std::fs::write(&b, upper).unwrap();
2679 Config::load_layers(&[a, b])
2680 }
2681
2682 #[test]
2683 fn graph_implementers_higher_layer_wins_across_spellings() {
2684 let c = layered("[graph]\ncandidates = 2\n", "[graph]\nimplementers = 5\n").unwrap();
2685 assert_eq!(c.graph.implementers, 5);
2686 let c = layered("[graph]\nimplementers = 2\n", "[graph]\ncandidates = 5\n").unwrap();
2687 assert_eq!(c.graph.implementers, 5);
2688 let c = layered("[graph]\ncandidates = 2\n", "[graph]\njudges = 3\n").unwrap();
2689 assert_eq!(c.graph.implementers, 2);
2690 }
2691
2692 #[test]
2693 fn graph_implementers_both_keys_in_one_layer_is_refused() {
2694 let err = layered("", "[graph]\ncandidates = 2\nimplementers = 3\n")
2695 .unwrap_err()
2696 .to_string();
2697 assert!(
2698 err.contains("candidates") && err.contains("implementers"),
2699 "{err}"
2700 );
2701 assert!(err.contains("b.toml"), "{err}");
2702 }
2703
2704 #[test]
2705 fn a_toml_layer_written_before_these_fields_existed_still_parses() {
2706 // `deny_unknown_fields` cuts both ways: a config from before
2707 // `timeout_verify`/`e2e_every_round` existed must still parse, with
2708 // both defaulted rather than refused as unknown-in-reverse.
2709 let g: Graph =
2710 toml::from_str("candidates = 1\nreviewers = 3\nreview_rounds = 6\nmax_parallel = 4\n")
2711 .expect("an old-shaped [graph] table must still parse");
2712 assert_eq!(g.verify_timeout(), Graph::default().timeout_review);
2713 assert!(
2714 !g.e2e_every_round,
2715 "off by default, same as before this field existed"
2716 );
2717 }
2718
2719 #[test]
2720 fn empty_roles_rotate_judges_off_their_own_candidate() {
2721 // Three seats, said out loud: this is a test about *rotation*, and it
2722 // has nothing to say about how many candidates a task buys by default.
2723 let cfg = Config {
2724 agents: vec![spec("a"), spec("b"), spec("c")],
2725 graph: Graph {
2726 implementers: 3,
2727 ..Graph::default()
2728 },
2729 ..Config::default()
2730 };
2731 let roles = cfg.resolve_roles().unwrap();
2732 let impls: Vec<&str> = roles.implementers.iter().map(|a| a.id.as_str()).collect();
2733 let judges: Vec<&str> = roles.judges.iter().map(|a| a.id.as_str()).collect();
2734 assert_eq!(impls, ["a", "b", "c"]);
2735 assert_eq!(judges, ["b", "c", "a"]);
2736 for (i, j) in judges.iter().enumerate() {
2737 assert_ne!(*j, impls[i], "judge {i} must not sit on its own candidate");
2738 }
2739 }
2740
2741 #[test]
2742 fn single_agent_roster_fills_every_seat() {
2743 let cfg = Config {
2744 agents: vec![spec("solo")],
2745 graph: Graph {
2746 implementers: 3,
2747 ..Graph::default()
2748 },
2749 ..Config::default()
2750 };
2751 let roles = cfg.resolve_roles().unwrap();
2752 assert_eq!(roles.implementers.len(), 3);
2753 assert!(roles.judges.iter().all(|a| a.id == "solo"));
2754 }
2755
2756 #[test]
2757 fn implementer_roster_is_the_whole_agent_list_untruncated_when_unset() {
2758 // `implementers = 1` (a solo run) still resolves `implementers` down to
2759 // one slot, but `implementer_roster` must carry every agent in the
2760 // roster, in file order, for `graph::Runner`'s quota fallback to walk
2761 // forward through once that one slot's agent runs out of quota.
2762 let cfg = Config {
2763 agents: vec![spec("a"), spec("b"), spec("c")],
2764 graph: Graph {
2765 implementers: 1,
2766 ..Graph::default()
2767 },
2768 ..Config::default()
2769 };
2770 let roles = cfg.resolve_roles().unwrap();
2771 assert_eq!(roles.implementers.len(), 1);
2772 let roster: Vec<&str> = roles
2773 .implementer_roster
2774 .iter()
2775 .map(|a| a.id.as_str())
2776 .collect();
2777 assert_eq!(roster, ["a", "b", "c"]);
2778 }
2779
2780 #[test]
2781 fn implementer_roster_follows_explicit_ids_in_the_order_named() {
2782 let cfg = Config {
2783 agents: vec![spec("a"), spec("b"), spec("c")],
2784 roles: Roles {
2785 implementers: vec!["c".to_owned(), "a".to_owned()],
2786 ..Roles::default()
2787 },
2788 graph: Graph {
2789 implementers: 1,
2790 ..Graph::default()
2791 },
2792 ..Config::default()
2793 };
2794 let roles = cfg.resolve_roles().unwrap();
2795 // Unaffected by the new field: still just the first named id.
2796 assert_eq!(roles.implementers.len(), 1);
2797 assert_eq!(roles.implementers[0].id, "c");
2798 let roster: Vec<&str> = roles
2799 .implementer_roster
2800 .iter()
2801 .map(|a| a.id.as_str())
2802 .collect();
2803 assert_eq!(roster, ["c", "a"]);
2804 }
2805
2806 #[test]
2807 fn explicit_roles_win() {
2808 let cfg = Config {
2809 agents: vec![spec("a"), spec("b")],
2810 roles: Roles {
2811 implementers: vec!["b".to_owned()],
2812 judges: vec!["a".to_owned()],
2813 reviewers: Vec::new(),
2814 fixer: Some("a".into()),
2815 ..Roles::default()
2816 },
2817 ..Config::default()
2818 };
2819 let roles = cfg.resolve_roles().unwrap();
2820 assert!(roles.implementers.iter().all(|a| a.id == "b"));
2821 assert!(roles.judges.iter().all(|a| a.id == "a"));
2822 let fixer = roles.fixer.unwrap();
2823 assert_eq!(fixer.len(), 1);
2824 assert_eq!(fixer[0].id, "a");
2825 assert_eq!(roles.conductor.id, "a");
2826 }
2827
2828 #[test]
2829 fn fixer_takes_a_string_or_an_ordered_chain() {
2830 let one: Roles = toml::from_str("fixer = \"a\"").unwrap();
2831 assert_eq!(one.fixer.as_ref().unwrap().ids(), ["a"]);
2832 let chain: Roles = toml::from_str("fixer = [\"b\", \"a\"]").unwrap();
2833 assert_eq!(chain.fixer.as_ref().unwrap().ids(), ["b", "a"]);
2834
2835 let resolve = |fixer: Option<AgentChoice>| {
2836 Config {
2837 agents: vec![spec("a"), spec("b"), spec("c")],
2838 roles: Roles {
2839 fixer,
2840 ..Roles::default()
2841 },
2842 ..Config::default()
2843 }
2844 .resolve_roles()
2845 };
2846 let ids = |r: &ResolvedRoles| -> Vec<String> {
2847 r.fixer
2848 .as_ref()
2849 .unwrap()
2850 .iter()
2851 .map(|s| s.id.clone())
2852 .collect()
2853 };
2854 // The string form is a chain of one; unset stays unset (the winner's author).
2855 assert_eq!(ids(&resolve(Some("c".into())).unwrap()), ["c"]);
2856 assert!(resolve(None).unwrap().fixer.is_none());
2857 // Order kept, an unknown id skipped, a duplicate keeps its first place.
2858 let chain = AgentChoice::Chain(vec!["b".into(), "zzz".into(), "a".into(), "b".into()]);
2859 assert_eq!(ids(&resolve(Some(chain)).unwrap()), ["b", "a"]);
2860 // Nothing resolving names the role.
2861 let err = resolve(Some(AgentChoice::Chain(vec!["zzz".into()]))).unwrap_err();
2862 assert!(format!("{err:#}").contains("fixer"), "{err:#}");
2863 }
2864
2865 #[test]
2866 fn unknown_agent_id_is_an_error() {
2867 let cfg = Config {
2868 agents: vec![spec("a")],
2869 roles: Roles {
2870 judges: vec!["nope".to_owned()],
2871 ..Roles::default()
2872 },
2873 ..Config::default()
2874 };
2875 assert!(cfg.resolve_roles().is_err());
2876 }
2877
2878 #[test]
2879 fn conductor_role_is_resolved_validated_and_has_a_fallback() {
2880 let mut cfg = Config {
2881 agents: vec![spec("a"), spec("b")],
2882 ..Config::default()
2883 };
2884 assert_eq!(cfg.resolve_roles().unwrap().conductor.id, "a");
2885
2886 cfg.roles.conductor = Some("b".into());
2887 assert_eq!(cfg.resolve_roles().unwrap().conductor.id, "b");
2888
2889 cfg.roles.conductor = Some("missing".into());
2890 assert!(cfg.resolve_roles().is_err());
2891 }
2892
2893 #[test]
2894 fn a_standalone_role_takes_a_string_or_an_array_and_round_trips() {
2895 let one: Roles = toml::from_str("synthesizer = \"opus\"").unwrap();
2896 assert_eq!(one.synthesizer, Some(AgentChoice::One("opus".to_owned())));
2897 assert_eq!(primary(one.synthesizer.as_ref()), Some("opus"));
2898
2899 let chain: Roles = toml::from_str("chatter = [\"opus\", \"codex\"]").unwrap();
2900 assert_eq!(chain.chatter.as_ref().unwrap().ids(), ["opus", "codex"]);
2901 assert_eq!(primary(chain.chatter.as_ref()), Some("opus"));
2902
2903 for roles in [one, chain] {
2904 let text = toml::to_string(&roles).unwrap();
2905 let back: Roles = toml::from_str(&text).unwrap();
2906 assert_eq!(back.synthesizer, roles.synthesizer);
2907 assert_eq!(back.chatter, roles.chatter);
2908 }
2909 let text = toml::to_string(&Roles {
2910 conductor: Some("x".into()),
2911 ..Roles::default()
2912 })
2913 .unwrap();
2914 assert!(
2915 text.contains("conductor = \"x\""),
2916 "string stays a string: {text}"
2917 );
2918 assert!(toml::from_str::<Roles>("synthesizer = 3").is_err());
2919 }
2920
2921 #[test]
2922 fn dupes_judge_role_accepts_strings_and_chains_and_round_trips() {
2923 for (text, ids) in [
2924 (r#"dupes_judge = "dedicated""#, vec!["dedicated"]),
2925 (
2926 r#"dupes_judge = ["first", "second"]"#,
2927 vec!["first", "second"],
2928 ),
2929 ("dupes_judge = []", vec![]),
2930 ] {
2931 let roles: Roles = toml::from_str(text).unwrap();
2932 assert_eq!(roles.dupes_judge.as_ref().unwrap().ids(), ids);
2933 let serialized = toml::to_string(&roles).unwrap();
2934 let back: Roles = toml::from_str(&serialized).unwrap();
2935 assert_eq!(back.dupes_judge, roles.dupes_judge);
2936 assert!(serialized.contains(text));
2937 }
2938 assert!(toml::from_str::<Roles>("dupes_judge = 3").is_err());
2939 assert!(Roles::default().dupes_judge.is_none());
2940 assert!(Config::starter_toml().contains("# dupes_judge ="));
2941
2942 // Standalone roles are resolved lazily, just like chatter.
2943 let mut cfg = Config {
2944 agents: vec![spec("a")],
2945 ..Config::default()
2946 };
2947 cfg.roles.dupes_judge = Some("missing".into());
2948 assert!(cfg.resolve_roles().is_ok());
2949 }
2950
2951 #[test]
2952 fn an_empty_choice_counts_as_unset() {
2953 assert!(AgentChoice::Chain(Vec::new()).ids().is_empty());
2954 assert!(AgentChoice::One(" ".to_owned()).ids().is_empty());
2955 assert_eq!(primary(Some(&AgentChoice::Chain(vec![]))), None);
2956 }
2957
2958 #[test]
2959 fn synthesizer_role_is_a_lazy_lookup_with_the_same_fallback_as_chatter() {
2960 // Unlike `conductor`, `synthesizer` is never validated by
2961 // `resolve_roles` — it is looked up lazily by
2962 // `graph::Runner::synthesize_brief` the same way `[roles] chatter`
2963 // is looked up by `talk::begin`, so this exercises `agent::pick`
2964 // directly instead of going through `resolve_roles`.
2965 let mut cfg = Config {
2966 agents: vec![spec("a"), spec("b")],
2967 ..Config::default()
2968 };
2969 let want = primary(cfg.roles.synthesizer.as_ref());
2970 assert_eq!(
2971 crate::agent::pick(&cfg.agents, want, &crate::agent::installed)
2972 .unwrap()
2973 .id,
2974 "a",
2975 "unset falls back to agent::pick's own default order"
2976 );
2977
2978 cfg.roles.synthesizer = Some("b".into());
2979 let want = primary(cfg.roles.synthesizer.as_ref());
2980 assert_eq!(
2981 crate::agent::pick(&cfg.agents, want, &crate::agent::installed)
2982 .unwrap()
2983 .id,
2984 "b",
2985 "the named agent wins over the default order"
2986 );
2987 }
2988
2989 #[test]
2990 fn empty_roster_is_an_error() {
2991 assert!(Config::default().resolve_roles().is_err());
2992 }
2993
2994 #[test]
2995 fn advise_defaults_to_on_with_three_proposals() {
2996 let g = Graph::default();
2997 assert!(g.advise);
2998 assert!(g.graft_runner_up, "graft is on by default");
2999 assert_eq!(g.advisors, 3);
3000 }
3001
3002 #[test]
3003 fn unset_advisors_falls_back_to_the_judge_roster() {
3004 let cfg = Config {
3005 agents: vec![spec("a"), spec("b")],
3006 roles: Roles {
3007 judges: vec!["b".to_owned()],
3008 ..Roles::default()
3009 },
3010 graph: Graph {
3011 advisors: 2,
3012 ..Graph::default()
3013 },
3014 ..Config::default()
3015 };
3016 let advisors = cfg.advisors().expect("advisors resolve");
3017 assert_eq!(advisors.len(), 2);
3018 assert!(
3019 advisors.iter().all(|a| a.id == "b"),
3020 "an unset [roles] advisors must fall back to [roles] judges: {advisors:?}"
3021 );
3022 }
3023
3024 #[test]
3025 fn an_explicit_advisor_roster_wins_over_the_judge_fallback() {
3026 let cfg = Config {
3027 agents: vec![spec("a"), spec("b")],
3028 roles: Roles {
3029 judges: vec!["b".to_owned()],
3030 advisors: vec!["a".to_owned()],
3031 ..Roles::default()
3032 },
3033 graph: Graph {
3034 advisors: 2,
3035 ..Graph::default()
3036 },
3037 ..Config::default()
3038 };
3039 let advisors = cfg.advisors().expect("advisors resolve");
3040 assert!(advisors.iter().all(|a| a.id == "a"));
3041 }
3042
3043 /// Neither `[roles] advisors` nor `[roles] judges` set: an unconfigured
3044 /// advisor roster must resolve to the exact same agents an unconfigured
3045 /// judge panel would get - same ids, same rotation offset - or the
3046 /// promise in [`Roles::advisors`]'s doc ("advisor diversity for free")
3047 /// does not actually hold.
3048 #[test]
3049 fn an_unconfigured_advisor_and_judge_roster_resolve_to_the_same_agents() {
3050 let cfg = Config {
3051 agents: vec![spec("a"), spec("b"), spec("c")],
3052 graph: Graph {
3053 advisors: 3,
3054 judges: 3,
3055 ..Graph::default()
3056 },
3057 ..Config::default()
3058 };
3059 let advisors = cfg.advisors().expect("advisors resolve");
3060 let judges = cfg.resolve_roles().expect("roles resolve").judges;
3061 let advisor_ids: Vec<&str> = advisors.iter().map(|a| a.id.as_str()).collect();
3062 let judge_ids: Vec<&str> = judges.iter().map(|a| a.id.as_str()).collect();
3063 assert_eq!(
3064 advisor_ids, judge_ids,
3065 "an unconfigured advisor roster must be the same seats an unconfigured judge panel gets"
3066 );
3067 }
3068
3069 #[test]
3070 fn an_unresolvable_advisor_seat_is_an_error_naming_the_id() {
3071 let cfg = Config {
3072 agents: vec![spec("a")],
3073 roles: Roles {
3074 advisors: vec!["nope".to_owned()],
3075 ..Roles::default()
3076 },
3077 graph: Graph {
3078 advisors: 1,
3079 ..Graph::default()
3080 },
3081 ..Config::default()
3082 };
3083 let err = cfg.advisors().expect_err("`nope` is not in the roster");
3084 assert!(format!("{err:#}").contains("nope"));
3085 }
3086
3087 #[test]
3088 fn repos_default_to_no_roots_and_a_day_of_trust() {
3089 assert_eq!(Config::default().repos.roots, Vec::<PathBuf>::new());
3090 assert_eq!(Config::default().repos.scan_ttl, 86_400);
3091 assert_eq!(Config::default().repos.fetch_interval, 600);
3092 }
3093
3094 #[test]
3095 fn a_config_file_with_no_repos_table_still_loads() {
3096 let dir = tempfile::tempdir().unwrap();
3097 let path = dir.path().join("magi.toml");
3098 std::fs::write(&path, "[graph]\ncandidates = 2\n").unwrap();
3099 let cfg = Config::load(&path).expect("must load without [repos]");
3100 assert_eq!(cfg.repos.roots, Vec::<PathBuf>::new());
3101 assert_eq!(cfg.repos.scan_ttl, 86_400);
3102 }
3103
3104 /// A fixture is the whole config under test.
3105 ///
3106 /// `layers` used to reach for `dirs::config_dir()` unconditionally, so on
3107 /// a machine where somebody had written `<config_dir>/magi/config.toml`
3108 /// the suite silently loaded it as the lowest layer. Adding `[repos]
3109 /// roots` there turned two web tests red - their fixtures declare
3110 /// `[repos] roots` too, and `refuse_split_arrays` rightly refuses one
3111 /// array key spread across two layers. CI stayed green throughout,
3112 /// because a runner has no such file: the suite failed only where magi is
3113 /// actually used.
3114 ///
3115 /// So a test build has no machine layer unless it asks for one, and this
3116 /// is that promise. Written against a real file at the real location so
3117 /// it fails if `machine_layer` starts reading it again.
3118 #[test]
3119 fn a_test_build_does_not_read_the_operators_machine_config() {
3120 let repo = tempfile::tempdir().unwrap();
3121 std::fs::write(repo.path().join("magi.toml"), "[graph]\ncandidates = 2\n").unwrap();
3122
3123 let layers = Config::layers(repo.path());
3124 assert_eq!(
3125 layers,
3126 vec![repo.path().join("magi.toml")],
3127 "only the fixture's own file may be a layer"
3128 );
3129 if let Some(real) = dirs::config_dir() {
3130 let machine = real.join("magi").join("config.toml");
3131 assert!(
3132 !layers.contains(&machine),
3133 "the operator's {} must not be a layer in a test build",
3134 machine.display()
3135 );
3136 }
3137 }
3138
3139 #[test]
3140 fn starter_toml_loads_through_teravars() {
3141 let dir = tempfile::tempdir().unwrap();
3142 let path = dir.path().join("magi.toml");
3143 std::fs::write(&path, Config::starter_toml()).unwrap();
3144 let parsed = Config::load(&path).expect("starter config must load");
3145 assert_eq!(parsed.graph.implementers, 3);
3146 assert_eq!(parsed.merge.mode, MergeMode::None);
3147 assert_eq!(parsed.merge.style, MergeStyle::Merge);
3148 assert!(parsed.graph.sessions);
3149 assert!(parsed.graph.github_text_guard);
3150 assert_eq!(parsed.graph.timeout_review, 1200);
3151 assert_eq!(parsed.graph.verify_timeout(), 1200);
3152 assert_eq!(parsed.update.mode, UpdateMode::Notify);
3153 }
3154
3155 #[test]
3156 fn release_defaults_to_actions_and_the_starter_documents_it() {
3157 let dir = tempfile::tempdir().unwrap();
3158 let path = dir.path().join("magi.toml");
3159 std::fs::write(&path, Config::starter_toml()).unwrap();
3160 let r = Config::load(&path).unwrap().release;
3161 assert_eq!(r.mode, ReleaseMode::Actions);
3162 assert!(!r.is_local());
3163 assert!(r.commands.is_empty());
3164 assert_eq!(r.timeout_minutes, 30);
3165 assert!(Config::starter_toml().contains("# mode = \"actions\""));
3166 }
3167
3168 #[test]
3169 fn release_local_parses_and_unknown_keys_are_refused() {
3170 let dir = tempfile::tempdir().unwrap();
3171 let path = dir.path().join("magi.toml");
3172 std::fs::write(
3173 &path,
3174 "[release]\nmode = \"local\"\ncommands = [\"true\", \"echo $MAGI_RELEASE_TAG\"]\ntimeout_minutes = 5\n",
3175 )
3176 .unwrap();
3177 let r = Config::load(&path).unwrap().release;
3178 assert!(r.is_local());
3179 assert_eq!(r.commands.len(), 2);
3180 assert_eq!(r.timeout_minutes, 5);
3181 std::fs::write(&path, "[release]\nmode = \"local\"\nbogus = 1\n").unwrap();
3182 assert!(Config::load(&path).is_err());
3183 std::fs::write(&path, "[release]\nmode = \"gha\"\n").unwrap();
3184 assert!(Config::load(&path).is_err());
3185 }
3186
3187 #[test]
3188 fn pre_gate_defaults_to_empty_and_the_starter_documents_it() {
3189 let dir = tempfile::tempdir().unwrap();
3190 let path = dir.path().join("magi.toml");
3191 std::fs::write(&path, Config::starter_toml()).unwrap();
3192 assert!(Config::load(&path).unwrap().verify.pre_gate.is_empty());
3193 assert!(Config::starter_toml().contains("# pre_gate = []"));
3194 }
3195
3196 #[test]
3197 fn starter_toml_explains_inherited_and_explicit_verify_timeouts() {
3198 let starter = Config::starter_toml();
3199 assert!(starter.contains("When timeout_verify is omitted, E2E and"));
3200 assert!(starter.contains("verification independent if timeout_review changes later"));
3201 assert!(starter.contains("# timeout_verify = 1200"));
3202 }
3203
3204 #[test]
3205 fn starter_toml_documents_chat_personas() {
3206 let starter = Config::starter_toml();
3207 assert!(starter.contains("# [[talk.personas]]"));
3208 assert!(starter.contains("tone only"));
3209 }
3210
3211 #[test]
3212 fn talk_personas_parse_and_are_validated() {
3213 let dir = tempfile::tempdir().unwrap();
3214 let path = dir.path().join("magi.toml");
3215 std::fs::write(
3216 &path,
3217 "[[talk.personas]]\nid = \"rei\"\nname = \"Quiet\"\nprompt = \"Be brief.\"\n\n\
3218 [[talk.personas]]\nid = \"gendo\"\nname = \"Gendo\"\nprompt = \"Be cold.\"\n",
3219 )
3220 .unwrap();
3221 let cfg = Config::load(&path).expect("loads");
3222 assert_eq!(cfg.talk.personas.len(), 2);
3223 let catalog = crate::persona::catalog(&cfg.talk.personas);
3224 let rei = catalog.iter().find(|p| p.id == "rei").unwrap();
3225 assert_eq!(
3226 (rei.name.as_str(), rei.prompt.as_str()),
3227 ("Quiet", "Be brief.")
3228 );
3229 assert_eq!(catalog.last().unwrap().id, "gendo");
3230
3231 for (body, needle) in [
3232 (
3233 "[[talk.personas]]\nid = \"\"\nname = \"n\"\nprompt = \"p\"\n",
3234 "empty `id`",
3235 ),
3236 (
3237 "[[talk.personas]]\nid = \"a\"\nname = \"n\"\nprompt = \" \"\n",
3238 "empty `prompt`",
3239 ),
3240 (
3241 "[[talk.personas]]\nid = \"a\"\nname = \"n\"\nprompt = \"p\"\n\
3242 [[talk.personas]]\nid = \"a\"\nname = \"m\"\nprompt = \"q\"\n",
3243 "more than once",
3244 ),
3245 (
3246 "[[talk.personas]]\nid = \"default\"\nname = \"n\"\nprompt = \"p\"\n",
3247 "cannot be redefined",
3248 ),
3249 ] {
3250 std::fs::write(&path, body).unwrap();
3251 let err = format!("{:#}", Config::load(&path).unwrap_err());
3252 assert!(err.contains(needle), "{needle}: {err}");
3253 }
3254 }
3255
3256 #[test]
3257 fn talk_operator_name_is_trimmed_and_validated() {
3258 let dir = tempfile::tempdir().unwrap();
3259 let path = dir.path().join("magi.toml");
3260 let load = |body: &str| {
3261 std::fs::write(&path, body).unwrap();
3262 Config::load(&path)
3263 };
3264 let ok = load("[talk]\noperator_name = \" Commander \"\n").expect("ok");
3265 assert_eq!(ok.talk.operator_name(), Some("Commander"));
3266 let blank = load("[talk]\noperator_name = \" \"\n").expect("blank");
3267 assert_eq!(blank.talk.operator_name(), None);
3268 assert!(load("[talk]\noperator_name = \"a\\nb\"\n").is_err());
3269 let long = "x".repeat(65);
3270 assert!(load(&format!("[talk]\noperator_name = \"{long}\"\n")).is_err());
3271 }
3272
3273 #[test]
3274 fn talk_persona_icon_parses_and_a_bad_shape_keeps_config_usable() {
3275 let dir = tempfile::tempdir().unwrap();
3276 let load = |body: &str| {
3277 let path = dir.path().join("c.toml");
3278 std::fs::write(&path, body).unwrap();
3279 Config::load(&path)
3280 };
3281 let ok = load(
3282 "[[talk.personas]]\nid = \"rei\"\nicon = \"~/rei.png\"\n\n\
3283 [[talk.personas]]\nid = \"default\"\nicon = \"data:image/png;base64,AAAA\"\n",
3284 )
3285 .unwrap();
3286 let c = crate::persona::catalog(&ok.talk.personas);
3287 assert_eq!(c[0].icon.as_deref(), Some("data:image/png;base64,AAAA"));
3288 assert_eq!(c[2].icon.as_deref(), Some("~/rei.png"));
3289 assert!(c[2].prompt.contains("Rei"), "built-in voice kept");
3290 let oversized = format!(
3291 "data:image/png;base64,{}",
3292 "A".repeat(crate::persona_icon::ICON_MAX_BYTES)
3293 );
3294 for bad in ["", "rei.png", "/a/b.svg", "/a/../b.png", &oversized] {
3295 for (id, fields) in [
3296 ("rei", ""),
3297 ("default", ""),
3298 ("custom", "name = \"Custom\"\nprompt = \"Keep calm\"\n"),
3299 ] {
3300 let body = format!("[[talk.personas]]\nid = \"{id}\"\nicon = \"{bad}\"\n{fields}");
3301 let cfg = load(&body).expect("bad icon must not invalidate config");
3302 let persona = crate::persona::find(&cfg.talk.personas, id).unwrap();
3303 assert_eq!(persona.id, id);
3304 assert!(crate::persona_icon::load(bad).is_err());
3305 }
3306 }
3307 let doc = Config::starter_toml();
3308 assert!(doc.contains("# icon = "));
3309 }
3310
3311 #[test]
3312 fn talk_personas_in_two_layers_are_refused() {
3313 let dir = tempfile::tempdir().unwrap();
3314 let machine = dir.path().join("machine.toml");
3315 let repo = dir.path().join("repo.toml");
3316 let one = "[[talk.personas]]\nid = \"a\"\nname = \"n\"\nprompt = \"p\"\n";
3317 std::fs::write(&machine, one).unwrap();
3318 std::fs::write(&repo, one.replace("\"a\"", "\"b\"")).unwrap();
3319 let err = format!("{:#}", Config::load_layers(&[machine, repo]).unwrap_err());
3320 assert!(err.contains("talk.personas"), "{err}");
3321 }
3322
3323 #[test]
3324 fn starter_toml_documents_role_lists_as_backup_order() {
3325 let starter = Config::starter_toml();
3326 assert!(starter.contains("also an ordered backup list"));
3327 assert!(starter.contains("codex then sonnet are spares"));
3328 }
3329
3330 /// A repository whose ruleset forbids merge commits declares that once,
3331 /// here, rather than magi asking GitHub about it on every render (see
3332 /// [`MergeStyle`]'s own doc for why).
3333 #[test]
3334 fn a_repository_can_declare_a_linear_history_merge_style() {
3335 let dir = tempfile::tempdir().unwrap();
3336 let path = dir.path().join("magi.toml");
3337 std::fs::write(&path, "[merge]\nmode = \"none\"\nstyle = \"squash\"\n").unwrap();
3338 let parsed = Config::load(&path).expect("config must load");
3339 assert_eq!(parsed.merge.style, MergeStyle::Squash);
3340 }
3341
3342 #[test]
3343 fn later_layers_win_and_vars_render() {
3344 let dir = tempfile::tempdir().unwrap();
3345 let machine = dir.path().join("machine.toml");
3346 let project = dir.path().join("magi.toml");
3347 // The machine layer owns the roster...
3348 std::fs::write(
3349 &machine,
3350 "[[agents]]\nid = \"opus\"\nkind = \"claude\"\nmodel = \"opus\"\n\n\
3351 [graph]\ncandidates = 3\nmax_parallel = 8\n",
3352 )
3353 .unwrap();
3354 // ...and the project layer only states what is repo-specific, plus a
3355 // `[vars]` value interpolated into a command.
3356 std::fs::write(
3357 &project,
3358 "[vars]\ncache = \"/shared\"\n\n\
3359 [graph]\ncandidates = 2\n\n\
3360 [verify]\ngate = [\"CARGO_TARGET_DIR={{ vars.cache }}/t cargo test\"]\n",
3361 )
3362 .unwrap();
3363
3364 let cfg = Config::load_layers(&[machine, project]).expect("layered load");
3365 assert_eq!(cfg.agents.len(), 1, "roster comes from the machine layer");
3366 assert_eq!(cfg.graph.implementers, 2, "project layer wins");
3367 assert_eq!(cfg.graph.max_parallel, 8, "machine layer survives");
3368 assert_eq!(
3369 cfg.verify.gate,
3370 ["CARGO_TARGET_DIR=/shared/t cargo test".to_owned()]
3371 );
3372 // The rendered command is where the cache path is read back from.
3373 assert_eq!(cfg.cache_dir(), Some(PathBuf::from("/shared/t")));
3374 }
3375
3376 #[test]
3377 fn talk_defaults_to_an_hour_and_an_unwritten_config_still_gets_it() {
3378 // An operator who writes no `[graph]` timeout keys at all must still
3379 // land on the hour, not on the five/fifteen minutes this turn used
3380 // to hardcode before it read from config.
3381 let g = Graph::default();
3382 assert_eq!(g.timeout_talk, 3600);
3383
3384 let dir = tempfile::tempdir().unwrap();
3385 let path = dir.path().join("magi.toml");
3386 std::fs::write(&path, "[graph]\ncandidates = 2\n").unwrap();
3387 let cfg = Config::load(&path).expect("must load without timeout_talk set");
3388 assert_eq!(cfg.graph.timeout_talk, 3600);
3389 }
3390
3391 #[test]
3392 fn an_overridden_talk_timeout_reaches_the_loaded_config() {
3393 let dir = tempfile::tempdir().unwrap();
3394 let path = dir.path().join("magi.toml");
3395 std::fs::write(&path, "[graph]\ntimeout_talk = 120\n").unwrap();
3396 let cfg = Config::load(&path).expect("must load");
3397 assert_eq!(cfg.graph.timeout_talk, 120);
3398 }
3399
3400 #[test]
3401 fn the_disk_defaults_are_the_measurements_made_up_front() {
3402 let cfg = Config::default();
3403 assert_eq!(cfg.disk.min_free_bytes, 8 * 1024 * 1024 * 1024);
3404 assert!(cfg.disk.auto_fold);
3405 assert_eq!(cfg.disk.fold_grace_secs, 6 * 60 * 60);
3406 assert_eq!(cfg.disk.cache_limit_bytes, 10 * 1024 * 1024 * 1024);
3407 }
3408
3409 #[test]
3410 fn an_unset_disk_section_is_the_safe_default() {
3411 let dir = tempfile::tempdir().unwrap();
3412 std::fs::write(dir.path().join("magi.toml"), "[graph]\ncandidates = 1\n").unwrap();
3413 let cfg = Config::load(&dir.path().join("magi.toml")).expect("load");
3414 assert_eq!(cfg.disk, Disk::default());
3415 }
3416
3417 #[test]
3418 fn env_is_available_to_templates_with_a_default() {
3419 let dir = tempfile::tempdir().unwrap();
3420 let path = dir.path().join("magi.toml");
3421 // teravars ships no `env`; magi adds it, and the `default` filter has
3422 // to cover the unset case or every machine would need the variable.
3423 //
3424 // Deliberately no named variable: `env` is keyed by the exact spelling
3425 // the OS reports, and Windows says `Path` where POSIX says `PATH`, so a
3426 // test asserting `env.PATH` passes on one runner and fails on another.
3427 // The map's non-emptiness is the platform-neutral claim.
3428 std::fs::write(
3429 &path,
3430 "[verify]\n\
3431 gate = [\"cache={{ env.MAGI_TEST_UNSET_XYZ | default(value='fallback') }}\", \
3432 \"populated={{ env | length > 0 }}\"]\n",
3433 )
3434 .unwrap();
3435 let cfg = Config::load(&path).expect("env lookup must render");
3436 assert_eq!(cfg.verify.gate[0], "cache=fallback");
3437 assert_eq!(cfg.verify.gate[1], "populated=true");
3438 }
3439
3440 #[test]
3441 fn a_broken_template_names_the_file() {
3442 let dir = tempfile::tempdir().unwrap();
3443 let path = dir.path().join("magi.toml");
3444 std::fs::write(&path, "[graph]\nlanguage = \"{{ nope.\"\n").unwrap();
3445 let err = Config::load(&path).expect_err("must not silently ignore");
3446 assert!(err.to_string().contains("teravars"), "{err}");
3447 }
3448
3449 #[test]
3450 fn tera_syntax_in_comments_is_inert() {
3451 // teravars >= 0.2.2 strips `#` comments before Tera sees the file, so a
3452 // comment may quote template syntax without rendering. Before 0.2.2 this
3453 // load failed: the commented-out braces reached the template parser.
3454 let dir = tempfile::tempdir().unwrap();
3455 let path = dir.path().join("magi.toml");
3456 std::fs::write(
3457 &path,
3458 "# a comment may quote templates: `{{ env.NOPE | default(value='x') }}` and `{% if %}`\n\
3459 [graph]\ncandidates = 2\n",
3460 )
3461 .unwrap();
3462 let cfg = Config::load(&path).expect("comments must be inert, not rendered");
3463 assert_eq!(cfg.graph.implementers, 2);
3464 }
3465
3466 #[test]
3467 fn opencode_defaults_to_file_delivery() {
3468 let mut s = spec("oc");
3469 s.kind = AgentKind::Opencode;
3470 assert_eq!(s.delivery(), Delivery::File);
3471 s.prompt_delivery = Some(Delivery::Argv);
3472 assert_eq!(s.delivery(), Delivery::Argv);
3473 }
3474 #[test]
3475 fn the_land_loop_is_on_but_it_cannot_merge_without_being_asked() {
3476 // Both default on, and that pair is the safety property: `land` takes
3477 // over the watching an operator was doing by hand, `land_approval`
3478 // keeps the irreversible step a human decision. An unattended merge
3479 // needs BOTH flipped, which has to be chosen deliberately twice.
3480 let g = Graph::default();
3481 assert!(
3482 g.land,
3483 "stopping at an open PR left the watching to a human"
3484 );
3485 assert!(
3486 g.land_approval,
3487 "on-by-default land is only defensible while this is also on"
3488 );
3489 assert!(g.land_rounds > 0, "a loop with no budget never terminates");
3490 assert!(
3491 g.hold_contested_merge,
3492 "a contested hand-off must still reach a human when approvals are off"
3493 );
3494 }
3495 #[test]
3496 fn an_array_declared_in_two_layers_is_refused_instead_of_concatenated() {
3497 // teravars appends arrays. For an ordered list of seats, or an argv,
3498 // the concatenation is something neither file says - and the operator
3499 // pays for the extra seats by the token.
3500 let dir = tempfile::tempdir().unwrap();
3501 let machine = dir.path().join("machine.toml");
3502 let repo = dir.path().join("magi.toml");
3503 std::fs::write(&machine, "[roles]\nimplementers = [\"a\", \"b\"]\n").unwrap();
3504 std::fs::write(&repo, "[roles]\nimplementers = [\"oc\"]\n").unwrap();
3505
3506 let err = Config::load_layers(&[machine.clone(), repo.clone()])
3507 .expect_err("two layers naming one array must not merge silently")
3508 .to_string();
3509 assert!(err.contains("roles.implementers"), "{err}");
3510 // Both files are named: the fix is to delete one of them, and the
3511 // operator has to know which two to choose between.
3512 assert!(err.contains("machine.toml"), "{err}");
3513 assert!(err.contains("magi.toml"), "{err}");
3514 }
3515
3516 #[test]
3517 fn a_scalar_in_one_layer_and_an_array_in_another_still_merges() {
3518 // The split the layering exists for: state a preference machine-wide,
3519 // let the repository own its own lists.
3520 let dir = tempfile::tempdir().unwrap();
3521 let machine = dir.path().join("machine.toml");
3522 let repo = dir.path().join("magi.toml");
3523 std::fs::write(&machine, "[roles]\nchatter = \"opus\"\n").unwrap();
3524 std::fs::write(
3525 &repo,
3526 "[[agents]]\nid = \"oc\"\nkind = \"opencode\"\n\n\
3527 [roles]\nimplementers = [\"oc\"]\n",
3528 )
3529 .unwrap();
3530
3531 let cfg = Config::load_layers(&[machine, repo]).expect("layers merge");
3532 assert_eq!(primary(cfg.roles.chatter.as_ref()), Some("opus"));
3533 assert_eq!(cfg.roles.implementers, ["oc"]);
3534 assert_eq!(cfg.agents.len(), 1, "the roster is not doubled");
3535 }
3536
3537 #[test]
3538 fn two_layers_declaring_verify_gate_run_both_in_priority_order() {
3539 // The `editorconfig-checker` distribution problem: a shared layer
3540 // wants to add a gate command without erasing the repository's own.
3541 let dir = tempfile::tempdir().unwrap();
3542 let machine = dir.path().join("machine.toml");
3543 let repo = dir.path().join("magi.toml");
3544 std::fs::write(&machine, "[verify]\ngate = [\"editorconfig-checker\"]\n").unwrap();
3545 std::fs::write(&repo, "[verify]\ngate = [\"cargo make check\"]\n").unwrap();
3546
3547 let cfg = Config::load_layers(&[machine, repo]).expect("appendable arrays must merge");
3548 assert_eq!(
3549 cfg.verify.gate,
3550 [
3551 "editorconfig-checker".to_owned(),
3552 "cargo make check".to_owned()
3553 ],
3554 "low-priority (machine) command first, high-priority (repo) command after"
3555 );
3556 }
3557
3558 fn load_one(body: &str) -> Result<Config> {
3559 let dir = tempfile::tempdir().unwrap();
3560 let path = dir.path().join("magi.toml");
3561 std::fs::write(&path, body).unwrap();
3562 Config::load_layers(&[path])
3563 }
3564
3565 #[test]
3566 fn worktree_setup_defaults_to_empty_and_parses_steps() {
3567 assert!(Config::default().worktree.setup.is_empty());
3568 let cfg = load_one(
3569 "[worktree]\nsetup = [\n { copy = \".env.example -> .env\" },\n \
3570 { copy = \"x\", optional = true },\n { run = \"true\", timeout_secs = 5 },\n]\n",
3571 )
3572 .unwrap();
3573 assert_eq!(cfg.worktree.setup.len(), 3);
3574 assert!(cfg.worktree.setup[1].optional);
3575 assert_eq!(cfg.worktree.setup[2].timeout_secs, Some(5));
3576 }
3577
3578 #[test]
3579 fn worktree_setup_rejects_malformed_steps() {
3580 for bad in [
3581 "{ copy = \"a\", run = \"b\" }",
3582 "{ optional = true }",
3583 "{ copy = \"a -> ../b\" }",
3584 "{ copy = \"a -> /abs\" }",
3585 "{ copy = \"a -> .git/config\" }",
3586 "{ run = \"b\", optional = true }",
3587 "{ copy = \"a\", timeout_secs = 3 }",
3588 "{ run = \"b\", bogus = 1 }",
3589 ] {
3590 let body = format!("[worktree]\nsetup = [{bad}]\n");
3591 assert!(load_one(&body).is_err(), "{bad}");
3592 }
3593 }
3594
3595 #[test]
3596 fn worktree_setup_in_two_layers_is_refused() {
3597 let dir = tempfile::tempdir().unwrap();
3598 let machine = dir.path().join("machine.toml");
3599 let repo = dir.path().join("magi.toml");
3600 std::fs::write(&machine, "[worktree]\nsetup = [{ run = \"a\" }]\n").unwrap();
3601 std::fs::write(&repo, "[worktree]\nsetup = [{ run = \"b\" }]\n").unwrap();
3602 assert!(Config::load_layers(&[machine, repo]).is_err());
3603 }
3604
3605 #[test]
3606 fn starter_toml_documents_worktree_setup() {
3607 let starter = Config::starter_toml();
3608 assert!(starter.contains("# [worktree]") && starter.contains("setup = ["));
3609 // The commented example must be valid once uncommented.
3610 let body = starter.replace("# [worktree]", "[worktree]");
3611 let mut out = String::new();
3612 let mut live = false;
3613 for line in body.lines() {
3614 if line == "[worktree]" {
3615 live = true;
3616 } else if live && !line.starts_with("# ") {
3617 live = false;
3618 }
3619 out.push_str(&if live {
3620 line.replacen("# ", "", 1)
3621 } else {
3622 line.to_owned()
3623 });
3624 out.push('\n');
3625 }
3626 let cfg = load_one(&out).expect("uncommented example loads");
3627 assert_eq!(cfg.worktree.setup.len(), 3);
3628 }
3629
3630 #[test]
3631 fn two_layers_declaring_verify_e2e_run_both_in_priority_order() {
3632 let dir = tempfile::tempdir().unwrap();
3633 let machine = dir.path().join("machine.toml");
3634 let repo = dir.path().join("magi.toml");
3635 std::fs::write(&machine, "[verify]\ne2e = [\"shared-smoke-test\"]\n").unwrap();
3636 std::fs::write(&repo, "[verify]\ne2e = [\"cargo test\"]\n").unwrap();
3637
3638 let cfg = Config::load_layers(&[machine, repo]).expect("appendable arrays must merge");
3639 assert_eq!(
3640 cfg.verify.e2e,
3641 ["shared-smoke-test".to_owned(), "cargo test".to_owned()]
3642 );
3643 }
3644
3645 #[test]
3646 fn two_layers_declaring_repos_roots_are_both_scanned() {
3647 let dir = tempfile::tempdir().unwrap();
3648 let machine = dir.path().join("machine.toml");
3649 let repo = dir.path().join("magi.toml");
3650 std::fs::write(&machine, "[repos]\nroots = [\"/machine/root\"]\n").unwrap();
3651 std::fs::write(&repo, "[repos]\nroots = [\"/repo/root\"]\n").unwrap();
3652
3653 let cfg = Config::load_layers(&[machine, repo]).expect("appendable arrays must merge");
3654 assert_eq!(
3655 cfg.repos.roots,
3656 [PathBuf::from("/machine/root"), PathBuf::from("/repo/root")]
3657 );
3658 }
3659
3660 #[test]
3661 fn duplicate_gate_commands_across_layers_both_run() {
3662 // Dropping the duplicate would be a silent surprise; the operator
3663 // sees a slower gate, never a missing one.
3664 let dir = tempfile::tempdir().unwrap();
3665 let machine = dir.path().join("machine.toml");
3666 let repo = dir.path().join("magi.toml");
3667 std::fs::write(&machine, "[verify]\ngate = [\"same-command\"]\n").unwrap();
3668 std::fs::write(&repo, "[verify]\ngate = [\"same-command\"]\n").unwrap();
3669
3670 let cfg = Config::load_layers(&[machine, repo]).expect("appendable arrays must merge");
3671 assert_eq!(
3672 cfg.verify.gate,
3673 ["same-command".to_owned(), "same-command".to_owned()]
3674 );
3675 }
3676
3677 #[test]
3678 fn notify_command_is_still_refused_across_two_layers() {
3679 // An argv, not a set: concatenating two of them is not a program.
3680 let dir = tempfile::tempdir().unwrap();
3681 let machine = dir.path().join("machine.toml");
3682 let repo = dir.path().join("magi.toml");
3683 std::fs::write(&machine, "[notify]\ncommand = [\"ntfy\", \"publish\"]\n").unwrap();
3684 std::fs::write(&repo, "[notify]\ncommand = [\"curl\", \"-X\"]\n").unwrap();
3685
3686 let err = Config::load_layers(&[machine.clone(), repo.clone()])
3687 .expect_err("an argv split across layers must not concatenate")
3688 .to_string();
3689 assert!(err.contains("notify.command"), "{err}");
3690 assert!(err.contains("machine.toml"), "{err}");
3691 assert!(err.contains("magi.toml"), "{err}");
3692 }
3693
3694 #[test]
3695 fn one_layer_declaring_verify_gate_runs_unchanged() {
3696 // The classification must not change behaviour for the configuration
3697 // this very repository has today: exactly one layer names the gate.
3698 let dir = tempfile::tempdir().unwrap();
3699 let path = dir.path().join("magi.toml");
3700 std::fs::write(&path, "[verify]\ngate = [\"cargo make check\"]\n").unwrap();
3701
3702 let cfg = Config::load(&path).expect("single layer must still load");
3703 assert_eq!(cfg.verify.gate, ["cargo make check".to_owned()]);
3704 }
3705
3706 #[test]
3707 fn describe_composed_names_the_contributing_layers_only_when_there_are_two() {
3708 let dir = tempfile::tempdir().unwrap();
3709 let machine = dir.path().join("machine.toml");
3710 let repo = dir.path().join("magi.toml");
3711 std::fs::write(&machine, "[verify]\ngate = [\"editorconfig-checker\"]\n").unwrap();
3712 std::fs::write(&repo, "[verify]\ngate = [\"cargo make check\"]\n").unwrap();
3713 let paths = vec![machine.clone(), repo.clone()];
3714
3715 let cfg = Config::load_layers(&paths).expect("appendable arrays must merge");
3716 let described =
3717 Config::describe_composed(&paths, &cfg.verify.gate, "verify.gate", "(none)");
3718 assert!(described.contains("editorconfig-checker && cargo make check"));
3719 assert!(
3720 described.contains(&machine.display().to_string()),
3721 "{described}"
3722 );
3723 assert!(
3724 described.contains(&repo.display().to_string()),
3725 "{described}"
3726 );
3727
3728 // A single contributing layer stays the plain one-line summary.
3729 let single = vec![repo.clone()];
3730 let solo_cfg = Config::load_layers(&single).expect("single layer loads");
3731 let solo_described =
3732 Config::describe_composed(&single, &solo_cfg.verify.gate, "verify.gate", "(none)");
3733 assert_eq!(solo_described, "cargo make check");
3734 }
3735
3736 mod remote_layers {
3737 use super::*;
3738
3739 fn git(dir: &Path, args: &[&str]) {
3740 use crate::proc::Quiet as _;
3741 let out = std::process::Command::new("git")
3742 .args(args)
3743 .quiet()
3744 .current_dir(dir)
3745 .output()
3746 .unwrap();
3747 assert!(out.status.success(), "git {args:?}: {out:?}");
3748 }
3749
3750 /// A checkout whose origin/main carries `remote` as magi.toml and whose
3751 /// working tree carries `disk`.
3752 fn fixture(remote: &str, disk: &str) -> (tempfile::TempDir, PathBuf) {
3753 let tmp = tempfile::tempdir().unwrap();
3754 let bare = tmp.path().join("o.git");
3755 git(tmp.path(), &["init", "--bare", "-b", "trunk", "o.git"]);
3756 let repo = tmp.path().join("repo");
3757 std::fs::create_dir_all(&repo).unwrap();
3758 git(&repo, &["init", "-b", "trunk"]);
3759 git(&repo, &["config", "user.name", "t"]);
3760 git(&repo, &["config", "user.email", "t@example.com"]);
3761 std::fs::write(repo.join("magi.toml"), remote).unwrap();
3762 git(&repo, &["add", "-A"]);
3763 git(&repo, &["commit", "-m", "x"]);
3764 git(&repo, &["remote", "add", "origin", &bare.to_string_lossy()]);
3765 git(&repo, &["push", "origin", "trunk"]);
3766 git(&repo, &["fetch", "origin"]);
3767 git(&repo, &["remote", "set-head", "origin", "trunk"]);
3768 std::fs::write(repo.join("magi.toml"), disk).unwrap();
3769 (tmp, repo)
3770 }
3771
3772 #[test]
3773 fn reads_the_remote_blob_not_the_working_tree() {
3774 let (_t, repo) = fixture("[graph]\ncandidates = 2\n", "[graph]\ncandidates = 5\n");
3775 let (cfg, _) = Config::discover(&repo, None).unwrap();
3776 assert_eq!(cfg.graph.implementers, 2);
3777 }
3778
3779 #[test]
3780 fn detached_head_still_reads_the_remote() {
3781 let (_t, repo) = fixture("[graph]\ncandidates = 2\n", "[graph]\ncandidates = 5\n");
3782 git(&repo, &["checkout", "--detach"]);
3783 let (cfg, _) = Config::discover(&repo, None).unwrap();
3784 assert_eq!(cfg.graph.implementers, 2);
3785 }
3786
3787 #[test]
3788 fn a_missing_ref_is_an_error_not_a_fallback() {
3789 let (_t, repo) = fixture("[graph]\ncandidates = 2\n", "[graph]\ncandidates = 5\n");
3790 git(&repo, &["update-ref", "-d", "refs/remotes/origin/trunk"]);
3791 let err = format!("{:#}", Config::discover(&repo, None).unwrap_err());
3792 assert!(err.contains("origin"), "{err}");
3793 }
3794
3795 #[test]
3796 fn merge_base_in_the_checkout_locates_the_ref() {
3797 let (_t, repo) = fixture("[graph]\ncandidates = 2\n", "[merge]\nbase = \"nope\"\n");
3798 assert!(Config::discover(&repo, None).is_err());
3799 }
3800
3801 #[test]
3802 fn the_magi_dir_layer_is_read_too() {
3803 let (_t, repo) = fixture("[graph]\ncandidates = 2\n", "");
3804 std::fs::write(repo.join("magi.toml"), "[graph]\ncandidates = 2\n").unwrap();
3805 std::fs::create_dir_all(repo.join(".magi")).unwrap();
3806 std::fs::write(repo.join(".magi/config.toml"), "[graph]\njudges = 3\n").unwrap();
3807 git(&repo, &["add", "-A"]);
3808 git(&repo, &["commit", "-m", "y"]);
3809 git(&repo, &["push", "origin", "trunk"]);
3810 git(&repo, &["fetch", "origin"]);
3811 let (cfg, layers) = Config::discover(&repo, None).unwrap();
3812 assert_eq!((cfg.graph.implementers, cfg.graph.judges), (2, 3));
3813 assert_eq!(layers.len(), 2);
3814 }
3815
3816 #[test]
3817 fn an_include_in_a_repository_layer_resolves_against_the_same_commit() {
3818 let (_t, repo) = fixture("include = [\"gate.toml\"]\n", "");
3819 std::fs::write(repo.join("magi.toml"), "include = [\"gate.toml\"]\n").unwrap();
3820 std::fs::write(repo.join("gate.toml"), "[graph]\ncandidates = 2\n").unwrap();
3821 git(&repo, &["add", "-A"]);
3822 git(&repo, &["commit", "-m", "inc"]);
3823 git(&repo, &["push", "origin", "trunk"]);
3824 git(&repo, &["fetch", "origin"]);
3825 // The working tree's copy differs and must not be used.
3826 std::fs::write(repo.join("gate.toml"), "[graph]\ncandidates = 9\n").unwrap();
3827 let (cfg, _) = Config::discover(&repo, None).unwrap();
3828 assert_eq!(cfg.graph.implementers, 2);
3829 }
3830
3831 #[test]
3832 fn a_machine_layer_base_is_rendered_before_locating_the_ref() {
3833 let (t, repo) = fixture("[graph]\ncandidates = 2\n", "[merge]\nbase = \"nope\"\n");
3834 let m = t.path().join("machine.toml");
3835 std::fs::write(
3836 &m,
3837 "[merge]\nbase = \"{{ env.MAGI_TEST_NO_SUCH | default(value='trunk') }}\"\n",
3838 )
3839 .unwrap();
3840 let (remote, base) = {
3841 let probe = probe_merge(&m, &repo).unwrap();
3842 (probe.remote, probe.base)
3843 };
3844 assert_eq!((remote, base), (None, Some("trunk".to_owned())));
3845 }
3846
3847 #[test]
3848 fn a_machine_layer_base_sees_the_checkouts_repo_name() {
3849 let (t, repo) = fixture("[graph]\ncandidates = 2\n", "");
3850 let m = t.path().join("machine.toml");
3851 std::fs::write(&m, "[merge]\nbase = \"{{ repo_name }}\"\n").unwrap();
3852 let probe = probe_merge(&m, &repo).unwrap();
3853 assert_eq!(probe.base.as_deref(), Some("repo"));
3854 }
3855
3856 #[test]
3857 fn every_include_form_is_recognised() {
3858 assert!(declares_include("include = [\"a.toml\"]\n"));
3859 assert!(declares_include("\"include\" = [\"a.toml\"]\n"));
3860 assert!(declares_include("teravars = { include = [\"a.toml\"] }\n"));
3861 assert!(declares_include("[teravars]\ninclude = [\"a.toml\"]\n"));
3862 assert!(!declares_include("[graph]\ncandidates = 2\n"));
3863 }
3864
3865 #[test]
3866 fn an_explicit_config_is_read_as_named() {
3867 let (t, repo) = fixture("[graph]\ncandidates = 2\n", "");
3868 let own = t.path().join("own.toml");
3869 std::fs::write(&own, "[graph]\ncandidates = 4\n").unwrap();
3870 let (cfg, _) = Config::discover(&repo, Some(&own)).unwrap();
3871 assert_eq!(cfg.graph.implementers, 4);
3872 }
3873
3874 #[test]
3875 fn discover_fetched_fails_when_the_fetch_does() {
3876 let (t, repo) = fixture("[graph]\ncandidates = 2\n", "");
3877 std::fs::remove_dir_all(t.path().join("o.git")).unwrap();
3878 let rt = tokio::runtime::Builder::new_current_thread()
3879 .enable_all()
3880 .build()
3881 .unwrap();
3882 let err = rt
3883 .block_on(Config::discover_fetched(&repo, None))
3884 .unwrap_err();
3885 assert!(format!("{err:#}").contains("fetch"), "{err:#}");
3886 }
3887
3888 #[test]
3889 fn discover_fetched_survives_a_fetch_that_leaves_no_origin_head() {
3890 // Never-fetched remote + followRemoteHEAD=never: fetch creates
3891 // origin/trunk but not origin/HEAD, which discovery then re-reads.
3892 let (_t, repo) = fixture("[graph]\ncandidates = 2\n", "");
3893 git(&repo, &["remote", "set-head", "origin", "-d"]);
3894 git(&repo, &["update-ref", "-d", "refs/remotes/origin/trunk"]);
3895 git(
3896 &repo,
3897 &["config", "remote.origin.followRemoteHEAD", "never"],
3898 );
3899 let rt = tokio::runtime::Builder::new_current_thread()
3900 .enable_all()
3901 .build()
3902 .unwrap();
3903 let (cfg, _) = rt.block_on(Config::discover_fetched(&repo, None)).unwrap();
3904 assert_eq!(cfg.graph.implementers, 2);
3905 }
3906
3907 #[test]
3908 fn a_non_git_directory_reads_its_files() {
3909 let dir = tempfile::tempdir().unwrap();
3910 std::fs::write(dir.path().join("magi.toml"), "[graph]\ncandidates = 4\n").unwrap();
3911 let (cfg, _) = Config::discover(dir.path(), None).unwrap();
3912 assert_eq!(cfg.graph.implementers, 4);
3913 }
3914 }
3915}