Skip to main content

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