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