Skip to main content

magi/
talk.rs

1//! The standing conversation: a place to think out loud with an agent between
2//! tasks, reachable from a phone.
3//!
4//! This is a conversation that stays open. Ask a question, have the agent
5//! read a file or run a command to check something, talk through an idea,
6//! and when it is time to act, tell it to file the work rather than do it
7//! here. The conversation does not end; it is what the operator opens the
8//! next time something comes up.
9//!
10//! # Talking is not implementing
11//!
12//! Every turn here runs with `allow_write: false` by default, for a reason
13//! that is not security, but attribution. An agent that edits a checkout
14//! mid-conversation leaves a diff that belongs to no run and passed no
15//! review, and on a repository entered into magi's blind competition that
16//! makes every candidate's diff unjudgeable. That is why the default holds
17//! regardless of what a repository's own `magi.toml` says about anything
18//! else. When the operator wants a change made, the agent is told to run
19//! `magi task add --solo` ([`briefing`]) rather than reach for an editor: the
20//! change goes through magi's own queue, on the repository's own terms, and
21//! the operator can watch it happen instead of trusting that it did.
22//!
23//! `[talk] allow_write` ([`crate::config::Talk::allow_write`]) lets a
24//! specific repository opt out of that default - a dotfiles or personal
25//! config checkout that is never entered into a competition and never
26//! reviewed has nothing for the restriction to protect, and filing a task for
27//! a one-line edit there is pure overhead. Turning it on does not turn this
28//! conversation into an implementer: [`briefing`] still sends everything
29//! bigger than a small, operator-named edit to the queue, and still tells the
30//! agent to say what it changed.
31//!
32//! With `[talk] allow_write` on, a codex turn also runs unsandboxed
33//! ([`turn_access`]), as a claude turn already does under `bypassPermissions`:
34//! otherwise `magi task add` (writes to magi's queue under the home directory)
35//! or `git fetch` is refused by the sandbox. Unattended seats never get this.
36//!
37//! `--solo` rather than a plain `magi task add` is the point of pairing this
38//! module with [`crate::queue::Task::solo`]. A task that came out of a
39//! conversation the operator just had is a decision already made, not a
40//! design question worth three independent takes - so it runs through one
41//! implementer and straight into review, the way [`crate::graph::Runner`]
42//! already degrades a single-candidate run.
43//!
44//! # Shape
45//!
46//! The same split [`crate::queue`] uses: [`Talk`] is data plus pure helpers,
47//! [`Talks`] owns the I/O and is constructed with its root, so every test
48//! here drives a real store in a temp directory rather than the operator's
49//! own home.
50
51use std::path::{Path, PathBuf};
52use std::sync::{Arc, Mutex, MutexGuard};
53use std::time::Duration;
54
55use anyhow::{Context, Result, bail};
56use jiff::Timestamp;
57use serde::{Deserialize, Serialize};
58
59use crate::agent::{self, Invocation, SeatState};
60use crate::config::{AgentSpec, Config};
61use crate::queue::{Queue, Source, Task};
62
63/// On-disk format for a conversation. Bumped when a field's meaning changes.
64pub const SCHEMA: u32 = 1;
65
66/// Wall-clock limit for one agent turn. See [`crate::config::Graph::timeout_talk`].
67///
68/// An hour by default. This turn is expected to run several shell commands
69/// and read their output before answering one - "what does this function
70/// do", "is this still true", "run the tests and tell me" - which argues for
71/// an hour rather than the five minutes a short budget once assumed, because
72/// the thing that made a short budget matter - an operator watching a
73/// spinner - is not how this conversation gets used: the operator moves on
74/// to something else while a turn runs and checks back later, so a long turn
75/// spends a held seat, not anyone's attention.
76fn turn_timeout(cfg: &Config) -> Duration {
77    Duration::from_secs(cfg.graph.timeout_talk)
78}
79
80/// Seat name for the conversation's agent, scoping its CLI-side session away
81/// from every other seat magi ever opens.
82const SEAT: &str = "talk";
83
84/// Prefix on a turn magi wrote rather than an agent.
85const MAGI_NOTE: &str = "magi: ";
86
87/// Who said something.
88#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
89#[serde(rename_all = "lowercase")]
90pub enum Who {
91    /// The operator.
92    Operator,
93    /// The conversation's agent - or magi itself, reporting that a turn
94    /// failed. See [`MAGI_NOTE`].
95    Agent,
96}
97
98/// Display provenance, independent of the CLI conversation role.
99#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
100#[serde(rename_all = "lowercase")]
101pub enum Sender {
102    /// A real operator send.
103    Operator,
104    /// A reply from the conversation persona.
105    Persona,
106    /// A post injected by magi.
107    Magi,
108}
109
110/// Resolve old records conservatively; operator prose is never classified by text.
111pub fn sender(turn: &Turn, message: Option<&Message>) -> Sender {
112    message
113        .and_then(|m| m.sender)
114        .or(turn.sender)
115        .unwrap_or_else(|| {
116            if turn.posted_by.is_some()
117                || (turn.who == Who::Agent && turn.body.starts_with(MAGI_NOTE))
118            {
119                Sender::Magi
120            } else if turn.who == Who::Operator {
121                Sender::Operator
122            } else {
123                Sender::Persona
124            }
125        })
126}
127
128/// One image the operator attached to a turn.
129///
130/// Never carries the bytes themselves: the picture lives on disk under
131/// [`Talks::attachments_dir`], named by `id` alone. `name` is the filename
132/// the operator's browser reported, kept only for display - it never
133/// contributes to a path, which is what keeps an upload from being able to
134/// traverse outside its own directory.
135#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
136#[serde(deny_unknown_fields)]
137pub struct Attachment {
138    /// Server-minted id; also the file's stem under `attachments_dir`.
139    pub id: String,
140    /// The operator's own filename, for display only.
141    pub name: String,
142    /// Validated by `web` at upload time against a closed whitelist:
143    /// `image/png`, `image/jpeg`, `image/gif`, `image/webp`.
144    pub mime: String,
145    /// Size in bytes, so the phone can show it without a second request.
146    pub bytes: u64,
147}
148
149/// Stable transcript location: a turn and a send within that turn.
150#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
151#[serde(deny_unknown_fields)]
152pub struct ReplyTarget {
153    /// Zero-based index in the append-only transcript.
154    pub turn: usize,
155    /// Zero-based send index; legacy turns have only index zero.
156    pub message: usize,
157}
158
159/// Server-generated, bounded snapshot; never recursively quotes replies.
160#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
161pub struct Reply {
162    /// Display provenance; absent in legacy records.
163    #[serde(default, skip_serializing_if = "Option::is_none")]
164    pub sender: Option<Sender>,
165    /// Original message location.
166    pub target: ReplyTarget,
167    /// Original speaker.
168    pub who: Who,
169    /// At most 240 Unicode characters of the original text.
170    pub excerpt: String,
171}
172
173/// One accepted send, retained even when the queue combines multiple sends.
174#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
175pub struct Message {
176    /// Notification-time facts for the UI; body remains the agent's authoritative input.
177    #[serde(default, skip_serializing_if = "Option::is_none")]
178    pub display: Option<crate::prompt::NoticeDisplay>,
179    /// Display provenance; absent in legacy records.
180    #[serde(default, skip_serializing_if = "Option::is_none")]
181    pub sender: Option<Sender>,
182    /// Unmodified message text.
183    pub body: String,
184    /// Images belonging to this send.
185    pub attachments: Vec<Attachment>,
186    /// Optional reply snapshot.
187    pub reply: Option<Reply>,
188}
189
190/// UI facts may be recovered only from an explicit sender or an exact saved notice.
191pub(crate) fn notice_display(
192    talk: &Talk,
193    turn: &Turn,
194    message: Option<&Message>,
195) -> Option<crate::prompt::NoticeDisplay> {
196    if let Some(display) = message.and_then(|m| m.display.as_ref()) {
197        return Some(display.clone());
198    }
199    let body = message.map_or(turn.body.as_str(), |m| m.body.as_str());
200    if sender(turn, message) == Sender::Magi
201        || talk.completion_notices.iter().any(|r| r.notice == body)
202    {
203        crate::prompt::legacy_notice_display(body)
204    } else {
205        None
206    }
207}
208
209fn resolve_reply(talk: &Talk, target: Option<ReplyTarget>) -> Result<Option<Reply>> {
210    let Some(target) = target else {
211        return Ok(None);
212    };
213    let turn = talk
214        .turns
215        .get(target.turn)
216        .context("invalid reply target")?;
217    let (body, attachments) = if turn.messages.is_empty() {
218        if target.message != 0 {
219            bail!("invalid reply target");
220        }
221        (&turn.body, &turn.attachments)
222    } else {
223        let message = turn
224            .messages
225            .get(target.message)
226            .context("invalid reply target")?;
227        (&message.body, &message.attachments)
228    };
229    let display = notice_display(talk, turn, turn.messages.get(target.message));
230    let body = display.as_ref().map_or(body.as_str(), |d| d.facts.as_str());
231    let excerpt = if body.trim().is_empty() {
232        format!("{} attachment(s)", attachments.len())
233    } else {
234        body.chars().take(240).collect()
235    };
236    Ok(Some(Reply {
237        sender: Some(if display.is_some() {
238            Sender::Magi
239        } else {
240            sender(turn, turn.messages.get(target.message))
241        }),
242        target,
243        who: turn.who,
244        excerpt,
245    }))
246}
247
248/// Quotes are prompt-only: stored text remains suitable for titles and decisions.
249fn message_prompt(turn: &Turn, talk: &Talk, store: &Talks) -> String {
250    if turn.messages.is_empty() {
251        return turn.body.clone();
252    }
253    turn.messages
254        .iter()
255        .map(|message| match &message.reply {
256            Some(reply) => {
257                let mut quote = format!(
258                    "[Reply to {}: {}]",
259                    if reply.sender == Some(Sender::Magi) {
260                        "MAGI".to_owned()
261                    } else {
262                        format!("{:?}", reply.who)
263                    },
264                    reply.excerpt.lines().collect::<Vec<_>>().join(" "),
265                );
266                if let Some(target) = talk.turns.get(reply.target.turn) {
267                    let attachments: &[Attachment] = if target.messages.is_empty() {
268                        &target.attachments
269                    } else if let Some(message) = target.messages.get(reply.target.message) {
270                        &message.attachments
271                    } else {
272                        &[]
273                    };
274                    for attachment in attachments {
275                        if let Some(path) = store.attachment_path(&talk.id, attachment) {
276                            quote.push_str(&format!(
277                                "\nReplied-to image: {} ({})",
278                                path.display(),
279                                attachment.mime
280                            ));
281                        }
282                    }
283                }
284                format!("{quote}\n{}", message.body)
285            }
286            None => message.body.clone(),
287        })
288        .collect::<Vec<_>>()
289        .join("\n\n")
290}
291
292/// Only legacy plain sends can omit metadata; explicit provenance is durable.
293fn messages_need_no_metadata(messages: &[Message]) -> bool {
294    messages.iter().all(|message| {
295        message.sender.is_none()
296            && message.display.is_none()
297            && message.reply.is_none()
298            && message.attachments.is_empty()
299    })
300}
301
302/// One message in the conversation.
303#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
304#[serde(deny_unknown_fields)]
305pub struct Turn {
306    /// Display provenance; absent in legacy records.
307    #[serde(default, skip_serializing_if = "Option::is_none")]
308    pub sender: Option<Sender>,
309    /// Who wrote it.
310    pub who: Who,
311    /// What they said.
312    pub body: String,
313    /// When it was said.
314    pub at: Timestamp,
315    /// Images attached to this turn. `#[serde(default)]` so a conversation
316    /// recorded before attachments existed still reads.
317    #[serde(default)]
318    pub attachments: Vec<Attachment>,
319    /// What the agent's CLI reported reading for this reply; only ever set on
320    /// an agent reply that carried usage. `#[serde(default)]` so a
321    /// conversation recorded before this field existed still reads, which is
322    /// why [`SCHEMA`] stays put: no existing field changed meaning.
323    #[serde(default, skip_serializing_if = "Option::is_none")]
324    pub usage: Option<TurnUsage>,
325    /// Byte offsets in [`Self::body`] where the second and later messages of
326    /// an operator turn begin. Replies queued while a turn runs are joined
327    /// with a blank line, which an owner's own blank lines cannot be told
328    /// apart from, so the boundary is recorded when it is made. `Some(empty)`
329    /// is one message; `None` is unknown (a turn stored before this field, or
330    /// not an operator turn) and readers fall back to paragraphs.
331    #[serde(default, skip_serializing_if = "Option::is_none")]
332    pub breaks: Option<Vec<usize>>,
333    #[serde(default, skip_serializing_if = "Vec::is_empty")]
334    /// Individual sends; empty for legacy turns.
335    pub messages: Vec<Message>,
336    /// The run id of the task seat that appended this agent turn with
337    /// `magi talk post`. Only ever set on such a turn; the chat's own CLI
338    /// session has not seen it, so [`posts_since_last_reply`] replays it in
339    /// the next prompt. `#[serde(default)]`: [`SCHEMA`] stays put.
340    #[serde(default, skip_serializing_if = "Option::is_none")]
341    pub posted_by: Option<String>,
342}
343
344/// Raw usage of one agent reply, stored as the CLI reported it.
345///
346/// Counts only, never a percentage: the window is configuration
347/// ([`Config::context_window`]) and the conversation's model can change, so a
348/// stored percentage would go stale the moment either did. `agent` and
349/// `model` record who read that many tokens, which is how [`context_usage`]
350/// notices the figure predates a switch.
351#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
352pub struct TurnUsage {
353    /// Input-side tokens the CLI reported - see `agent::context_tokens`.
354    pub context_tokens: u64,
355    /// Roster id that answered.
356    pub agent: String,
357    /// That agent's model at the time (`None`: the CLI's own default).
358    #[serde(default)]
359    pub model: Option<String>,
360}
361
362/// How full the conversation's context window is, as the phone shows it.
363///
364/// Derived at read time and never persisted. `None` means unknown, which the
365/// UI says in words - it is never a made-up 0.
366#[derive(Debug, Clone, Default, PartialEq, Eq, Serialize)]
367pub struct ContextUsage {
368    /// Tokens the last agent reply's turn read, if its CLI reported any.
369    pub tokens: Option<u64>,
370    /// Window of the conversation's *current* model, if one is configured.
371    pub window: Option<u64>,
372    /// `tokens / window`, rounded down, not capped (over 100 is real news).
373    pub percent: Option<u64>,
374    /// 80% or more, judged before rounding: the conversation is getting long.
375    pub warn: bool,
376    /// The conversation's agent or model is not the one that produced
377    /// `tokens`: the figure describes the old session and the next turn
378    /// re-measures it.
379    pub since_switch: bool,
380    /// The current model, if the roster names one.
381    pub model: Option<String>,
382    /// `tokens` is a transcript-length estimate, not a CLI measurement (see
383    /// [`estimate_context_tokens`]). A measured figure is never overridden.
384    #[serde(default)]
385    pub estimated: bool,
386}
387
388/// Fraction (in percent) of the window at which [`ContextUsage::warn`] fires.
389const CONTEXT_WARN_PERCENT: u64 = 80;
390
391/// Allowance for the standing prompt when no config is readable to render
392/// the real [`briefing`].
393const STANDING_PROMPT_FALLBACK_CHARS: u64 = 7000;
394
395/// Rough estimate of the context a conversation occupies, for CLIs whose usage
396/// is cumulative and so is never measured (claude / codex / agy).
397///
398/// Counts the characters (not bytes) of every operator and agent turn, skips
399/// magi's own notes (never sent to the agent), adds `standing_chars` for the
400/// standing prompt, and divides by ~3.5 chars per token, rounding up. `None`
401/// when no turn counts: an empty conversation stays unknown.
402///
403/// Known biases: CJK runs near one token per character, and tool output,
404/// images and the CLI's own system prompt are not counted, so this tends to
405/// under-estimate (the 80% warning comes late); a compacted session is still
406/// counted in full, which over-estimates.
407pub fn estimate_context_tokens(talk: &Talk, standing_chars: u64) -> Option<u64> {
408    let mut counted = false;
409    let mut chars = standing_chars;
410    for t in talk.turns.iter().filter(|t| !t.body.starts_with(MAGI_NOTE)) {
411        counted = true;
412        chars += t.body.chars().count() as u64;
413    }
414    // chars / 3.5, rounded up, in integers.
415    counted.then(|| (chars * 2).div_ceil(7))
416}
417
418/// Context usage of `talk`, measured against its current model.
419///
420/// Only the latest agent reply counts (magi's own notes are skipped) and a
421/// reply without usage makes the answer unknown - older replies are never
422/// consulted, since a stale count passed off as current is worse than "unknown".
423/// The window comes from the *current* agent's model, so switching model moves
424/// the denominator at once.
425///
426/// Switching agent (or model, which is an agent change) mints a fresh CLI
427/// session, so the next turn re-sends the whole transcript and the count
428/// resets or jumps. Until that turn lands, the old figure is reported with
429/// `since_switch` set. Deterministic: same talk and config, same answer.
430pub fn context_usage(talk: &Talk, cfg: Option<&Config>) -> ContextUsage {
431    let current = cfg.and_then(|c| c.agents.iter().find(|a| a.id == talk.agent));
432    let model = current.and_then(|a| a.model.clone());
433    let window = cfg
434        .zip(model.as_deref())
435        .and_then(|(c, m)| c.context_window(m))
436        .filter(|w| *w > 0);
437    let usage = talk
438        .turns
439        .iter()
440        .rev()
441        .find(|t| t.who == Who::Agent && !t.body.starts_with(MAGI_NOTE))
442        .and_then(|t| t.usage.as_ref());
443    let measured = usage.map(|u| u.context_tokens);
444    let tokens = measured.or_else(|| {
445        let standing = cfg.map_or(STANDING_PROMPT_FALLBACK_CHARS, |c| {
446            briefing_for(
447                &talk.repo,
448                &c.graph.language,
449                c.talk.allow_write,
450                crate::persona::active(&c.talk.personas, &talk.persona).as_ref(),
451                c.talk.operator_name(),
452                talk.implementers,
453            )
454            .chars()
455            .count() as u64
456        });
457        estimate_context_tokens(talk, standing)
458    });
459    let estimated = measured.is_none() && tokens.is_some();
460    let since_switch =
461        usage.is_some_and(|u| u.agent != talk.agent || (current.is_some() && u.model != model));
462    let (percent, warn) = match (tokens, window) {
463        (Some(t), Some(w)) => (
464            Some(t.saturating_mul(100) / w),
465            t.saturating_mul(100) >= w.saturating_mul(CONTEXT_WARN_PERCENT),
466        ),
467        _ => (None, false),
468    };
469    ContextUsage {
470        tokens,
471        window,
472        percent,
473        warn,
474        since_switch,
475        model,
476        estimated,
477    }
478}
479
480/// Where a conversation is in its life: this conversation can file any
481/// number of tasks without ending, so it only ever moves once, from open to
482/// closed.
483#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
484#[serde(rename_all = "lowercase")]
485pub enum TalkStatus {
486    /// Still open; the operator may say more, and may have already filed work
487    /// out of it.
488    Open,
489    /// Closed by hand. Kept on disk as a record.
490    Closed,
491}
492
493impl TalkStatus {
494    /// Is this conversation still live?
495    pub fn open(self) -> bool {
496        matches!(self, Self::Open)
497    }
498
499    /// Wire form, for the phone and for logs.
500    pub fn as_str(self) -> &'static str {
501        match self {
502            Self::Open => "open",
503            Self::Closed => "closed",
504        }
505    }
506}
507
508/// One standing conversation.
509#[derive(Debug, Clone, Serialize, Deserialize)]
510#[serde(deny_unknown_fields)]
511pub struct Talk {
512    /// On-disk format version.
513    pub schema: u32,
514    /// Conversation id, e.g. `20260904-014455-ab12`.
515    pub id: String,
516    /// Repository this conversation is about.
517    pub repo: PathBuf,
518    /// Roster agent id holding the conversation.
519    pub agent: String,
520    /// Current state.
521    pub status: TalkStatus,
522    /// Everything said, oldest first.
523    pub turns: Vec<Turn>,
524    /// Recommendation-time question snapshots keyed by persona turn position.
525    #[serde(default, skip_serializing_if = "std::collections::BTreeMap::is_empty")]
526    pub recommendations: std::collections::BTreeMap<usize, crate::consult::RecommendationManifest>,
527    /// Exclusive transcript boundary included in the last successful CLI prompt.
528    /// Posts at or beyond it still need replay, even if a reply follows them.
529    /// Missing on older records, which infer it from the last non-post reply.
530    #[serde(default, skip_serializing_if = "Option::is_none")]
531    pub posts_read_through: Option<usize>,
532    /// Text and attachments accepted while the single CLI turn is busy.
533    /// They are durable, but become a real turn only when [`drain`] records them.
534    #[serde(default)]
535    pub pending: String,
536    /// Attachments paired with [`Self::pending`].
537    #[serde(default)]
538    pub pending_attachments: Vec<Attachment>,
539    /// Message boundaries inside [`Self::pending`], as [`Turn::breaks`].
540    #[serde(default, skip_serializing_if = "Option::is_none")]
541    pub pending_breaks: Option<Vec<usize>>,
542    #[serde(default, skip_serializing_if = "Vec::is_empty")]
543    /// Individual sends waiting to be drained.
544    pub pending_messages: Vec<Message>,
545    /// May a failed turn fall back through the rest of `[roles] chatter`?
546    /// True only while the agent was chosen by that chain; an explicit
547    /// `--agent` or an operator's switch pins the conversation to its agent
548    /// even when that agent also appears in the chain.
549    #[serde(default)]
550    pub fallback: bool,
551    /// Persona id (see [`crate::persona`]); empty is the plain default voice.
552    /// Independent of [`Self::agent`]: it survives an agent switch.
553    #[serde(default)]
554    pub persona: String,
555    /// The persona changed after the CLI session was last told about it. The
556    /// next resumed turn carries a persona update; cleared only once a turn
557    /// has succeeded (a fresh seat's briefing already holds the new persona).
558    #[serde(default)]
559    pub persona_dirty: bool,
560    /// How many implementers the tasks this chat files use (1..=3; 1 is
561    /// `--solo`). Only the briefing's wording changes; nothing enforces it.
562    #[serde(default = "solo_implementers")]
563    pub implementers: u8,
564    /// [`Self::implementers`] changed after the CLI session was last told;
565    /// cleared only once a turn has succeeded, like [`Self::persona_dirty`].
566    #[serde(default)]
567    pub implementers_dirty: bool,
568    /// Keys of task-completion notices already accepted into this talk (see
569    /// [`queue_once`]). Kept beside [`Self::pending`] and saved in the same
570    /// write, so a crash between this talk and the task's own record cannot
571    /// post twice; never pruned by a drain or a close.
572    #[serde(default)]
573    pub notified: Vec<String>,
574    /// Automatic notice references, saved atomically with their draft and key.
575    #[serde(default, skip_serializing_if = "Vec::is_empty")]
576    pub completion_notices: Vec<crate::chat_notify::CompletionReference>,
577    /// When the conversation was opened.
578    pub created_at: Timestamp,
579    /// Last change to this file.
580    pub updated_at: Timestamp,
581    /// The CLI-side conversation, so a turn after the first costs one
582    /// sentence instead of the whole transcript. Not `pub`: it is magi's
583    /// bookkeeping, and a caller that edited it would detach the record from
584    /// the conversation the model actually holds.
585    seat: SeatState,
586}
587
588impl Talk {
589    /// Short form used in lists and notifications, matching a run's short id.
590    pub fn short(&self) -> &str {
591        short(&self.id)
592    }
593}
594
595/// A conversation store on disk.
596#[derive(Debug, Clone)]
597pub struct Talks {
598    root: PathBuf,
599    /// Serializes the read-modify-write cycle that reads a talk, decides
600    /// something from its `status`, and writes the whole record back.
601    /// [`close`], [`record`] and the tail of [`turn`] all take this before
602    /// that cycle rather than after just the read: a re-read narrows the
603    /// window another writer can land in, but does not close it, since
604    /// nothing stopped that other writer's own put from landing between this
605    /// call's re-read and its own put. Shared across every clone, since every
606    /// clone is a handle onto the same files.
607    lock: Arc<Mutex<()>>,
608}
609
610impl Talks {
611    /// The operator's conversations, `<home>/talks`.
612    pub fn open() -> Self {
613        Self::at(crate::run::home().join("talks"))
614    }
615
616    /// A store at an explicit root. Tests use this, which is why none of them
617    /// need the operator's real home.
618    pub fn at(root: PathBuf) -> Self {
619        Self {
620            root,
621            lock: Arc::new(Mutex::new(())),
622        }
623    }
624
625    /// Claim the right to read-modify-write a talk's `status`. A plain
626    /// `std::sync::Mutex`, not an async one: every caller holds it across a
627    /// handful of small file operations and never across an `.await`, so
628    /// blocking the thread briefly is the right tool, not a reason to reach
629    /// for `tokio::sync::Mutex`. Poisoning recovers rather than propagates -
630    /// one panicking caller must not wedge every talk in the store the way it
631    /// would wedge the loop's own lock; see [`crate::web`]'s `lock_or_recover`,
632    /// which this mirrors.
633    ///
634    /// The mutex only serializes this process. The cycle is also taken under a
635    /// short file lock beside the records, so the CLI (`magi answer
636    /// --ask-chat`) and the web server cannot overwrite each other's draft.
637    /// The file lock is never skipped: a cycle that cannot get it within
638    /// longer than the lock's own expiry fails instead of writing unguarded.
639    fn guard(&self) -> Result<StoreGuard<'_>> {
640        let answer = self.answer_guard()?;
641        std::fs::create_dir_all(&self.root)
642            .with_context(|| format!("create {}", self.root.display()))?;
643        let path = self.root.join(".store.turn");
644        let deadline = std::time::Instant::now() + TAKEOVER_LOCK_TTL * 2;
645        loop {
646            if let Some(file) = TurnLock::take(&path)? {
647                return Ok(StoreGuard {
648                    _file: file,
649                    _answer: answer,
650                });
651            }
652            if std::time::Instant::now() >= deadline {
653                bail!("the talk store is locked by another process");
654            }
655            std::thread::sleep(Duration::from_millis(10));
656        }
657    }
658
659    /// Try both conversation locks without waiting under a question lock.
660    fn try_guard(&self) -> Result<Option<StoreGuard<'_>>> {
661        let Some(answer) = self.try_answer_guard()? else {
662            return Ok(None);
663        };
664        std::fs::create_dir_all(&self.root)
665            .with_context(|| format!("create {}", self.root.display()))?;
666        Ok(
667            TurnLock::take(&self.root.join(".store.turn"))?.map(|file| StoreGuard {
668                _file: file,
669                _answer: answer,
670            }),
671        )
672    }
673
674    /// Freeze owner evidence until an answer has been persisted. Callers under
675    /// a question lock must use [`Self::try_answer_guard`] instead: waiting here
676    /// can outlive the question lock's expiry. Retain the guard through
677    /// `Questions::update`'s write to prevent a retraction race.
678    pub fn answer_guard(&self) -> Result<AnswerGuard<'_>> {
679        let deadline = std::time::Instant::now() + TAKEOVER_LOCK_TTL * 2;
680        loop {
681            if let Some(guard) = self.try_answer_guard()? {
682                return Ok(guard);
683            }
684            if std::time::Instant::now() >= deadline {
685                bail!("conversation evidence is locked by another process");
686            }
687            std::thread::sleep(Duration::from_millis(10));
688        }
689    }
690
691    /// Acquire owner evidence without waiting while a question lock is held.
692    /// On contention, release the question lock before retrying, then reload
693    /// the question so another answer cannot be overwritten by an old snapshot.
694    pub fn try_answer_guard(&self) -> Result<Option<AnswerGuard<'_>>> {
695        let mutex = match self.lock.try_lock() {
696            Ok(guard) => guard,
697            Err(std::sync::TryLockError::Poisoned(err)) => err.into_inner(),
698            Err(std::sync::TryLockError::WouldBlock) => return Ok(None),
699        };
700        // Keep temporary lock files separate from transcripts so answering
701        // seats can synchronize without permission to alter owner evidence.
702        let root = self.answer_lock_root();
703        std::fs::create_dir_all(&root).context("create conversation evidence lock directory")?;
704        let path = root.join("store.turn");
705        Ok(TurnLock::take(&path)?.map(|file| AnswerGuard {
706            _file: file,
707            _mutex: mutex,
708        }))
709    }
710
711    /// Writable synchronization files, separate from conversation evidence.
712    pub fn answer_lock_root(&self) -> PathBuf {
713        self.root.join(".answer-lock")
714    }
715
716    /// Directory holding the conversation files.
717    pub fn root(&self) -> &Path {
718        &self.root
719    }
720
721    /// Path for one conversation id.
722    pub fn path_of(&self, id: &str) -> PathBuf {
723        self.root.join(format!("{id}.json"))
724    }
725
726    /// Where one conversation's prompts and CLI output are kept, beside the
727    /// record rather than inside it.
728    pub fn artifacts_of(&self, id: &str) -> PathBuf {
729        self.root.join(format!("{id}.artifacts"))
730    }
731
732    /// Where this conversation's attached images live: a subdirectory of
733    /// `artifacts_of`, so deleting the conversation deletes its attachments
734    /// too and nothing here needs its own cleanup path.
735    pub fn attachments_dir(&self, id: &str) -> PathBuf {
736        self.artifacts_of(id).join("attachments")
737    }
738
739    /// Persist one already-validated attachment and return its metadata.
740    ///
741    /// `web::talk_attachment_post` is the only caller: it has already
742    /// checked `mime` against the whitelist and sniffed the bytes, so an
743    /// unrecognised mime reaching here is a bug in that caller, not
744    /// something an operator did. The id is minted here and never taken
745    /// from the client; `name` is stored for display only and never used to
746    /// build a path.
747    pub fn put_attachment(
748        &self,
749        id: &str,
750        mime: &str,
751        name: &str,
752        data: &[u8],
753    ) -> Result<Attachment> {
754        let dir = self.attachments_dir(id);
755        std::fs::create_dir_all(&dir).with_context(|| format!("create {}", dir.display()))?;
756        let ext = attachment_ext(mime).with_context(|| format!("unsupported mime `{mime}`"))?;
757        let att = Attachment {
758            id: new_attachment_id(),
759            name: name.to_owned(),
760            mime: mime.to_owned(),
761            bytes: data.len() as u64,
762        };
763        std::fs::write(dir.join(format!("{}.{ext}", att.id)), data)
764            .with_context(|| format!("write attachment {}", att.id))?;
765        std::fs::write(
766            dir.join(format!("{}.json", att.id)),
767            serde_json::to_string(&att).context("serialize attachment")?,
768        )
769        .with_context(|| format!("write attachment metadata {}", att.id))?;
770        Ok(att)
771    }
772
773    /// Just the metadata, without reading the image bytes back off disk -
774    /// what `web::talk_say` uses to turn an id the operator referenced into
775    /// an [`Attachment`] before appending a [`Turn`], where the bytes
776    /// themselves are of no interest. `None` for an id this conversation
777    /// never stored - including one that merely looks plausible:
778    /// [`valid_attachment_id`] is checked here too, not only by the caller,
779    /// the same defence-in-depth `Questions::panel_asset` uses for its own
780    /// asset ids.
781    pub fn attachment_meta(&self, id: &str, att_id: &str) -> Result<Option<Attachment>> {
782        if !valid_attachment_id(att_id) {
783            return Ok(None);
784        }
785        let meta_path = self.attachments_dir(id).join(format!("{att_id}.json"));
786        if !meta_path.is_file() {
787            return Ok(None);
788        }
789        let att = serde_json::from_str(
790            &std::fs::read_to_string(&meta_path)
791                .with_context(|| format!("read {}", meta_path.display()))?,
792        )
793        .with_context(|| format!("parse {}", meta_path.display()))?;
794        Ok(Some(att))
795    }
796
797    /// A stored attachment's metadata and its bytes together, for serving it
798    /// back on `GET`. `None` under the same conditions as
799    /// [`Talks::attachment_meta`], which this is built on.
800    pub fn read_attachment(&self, id: &str, att_id: &str) -> Result<Option<(Attachment, Vec<u8>)>> {
801        let Some(att) = self.attachment_meta(id, att_id)? else {
802            return Ok(None);
803        };
804        let ext = attachment_ext(&att.mime).with_context(|| {
805            format!("attachment {att_id} has an unsupported mime `{}`", att.mime)
806        })?;
807        let data_path = self.attachments_dir(id).join(format!("{att_id}.{ext}"));
808        let data =
809            std::fs::read(&data_path).with_context(|| format!("read {}", data_path.display()))?;
810        Ok(Some((att, data)))
811    }
812
813    /// Absolute path of one attachment's bytes, for the prompt note [`turn`]
814    /// appends and for [`Invocation::attachments`]. `None` only for a mime
815    /// [`put_attachment`] could never have written, which means the
816    /// attachment did not come from this store.
817    ///
818    /// `self.root` (and so `attachments_dir`) is not guaranteed absolute on
819    /// its own - `run::home()` returns a bare relative `PathBuf` verbatim
820    /// when the operator sets `MAGI_HOME` to a relative path, and nothing
821    /// canonicalizes it on the way in. That is harmless for every other use
822    /// of this store, since its own I/O runs in this process against this
823    /// process's cwd - but this path is handed to a CLI invoked with `cwd:
824    /// &talk.repo`, a different directory, so a relative path here would
825    /// resolve against the wrong place once it reached the prompt.
826    /// `std::path::absolute` fixes it against *this* process's cwd before
827    /// that happens; see `disk::free_bytes_by_os` for the same function used
828    /// the same way elsewhere in this codebase.
829    fn attachment_path(&self, id: &str, att: &Attachment) -> Option<PathBuf> {
830        let ext = attachment_ext(&att.mime)?;
831        let path = self.attachments_dir(id).join(format!("{}.{ext}", att.id));
832        std::path::absolute(&path).ok()
833    }
834
835    /// Write a conversation, atomically, so a process killed mid-write leaves
836    /// the previous state readable rather than a truncated file.
837    ///
838    /// The write-then-rename itself is retried a handful of times - see
839    /// [`write_atomic`] - because a reader with the destination file briefly
840    /// open is exactly the kind of failure that must not cost an agent's
841    /// whole reply; see `turn`'s own tail for what happens when even that is
842    /// not enough.
843    pub fn put(&self, t: &mut Talk) -> Result<()> {
844        std::fs::create_dir_all(&self.root)
845            .with_context(|| format!("create {}", self.root.display()))?;
846        let previous = self.get(&t.id).ok();
847        let start = previous.as_ref().map_or(0, |old| old.turns.len());
848        // Existing manifests are immutable even when a stale handle saves.
849        if let Some(old) = previous {
850            t.recommendations = old.recommendations;
851        }
852        let questions = crate::ask::Questions::at(
853            self.root
854                .parent()
855                .context("talk store has no home")?
856                .join("questions"),
857        );
858        for (index, turn) in t.turns.iter().enumerate().skip(start) {
859            if turn.who == Who::Agent
860                && sender(turn, None) == Sender::Persona
861                && turn.posted_by.is_none()
862            {
863                t.recommendations.insert(
864                    index,
865                    crate::consult::recommendation_manifest(&turn.body, &questions),
866                );
867            }
868        }
869        t.updated_at = Timestamp::now();
870        // Legacy plain messages can still be recovered losslessly from breaks.
871        // Explicit provenance must remain on disk, even without attachments.
872        let mut persisted = t.clone();
873        for turn in &mut persisted.turns {
874            if turn.attachments.is_empty() && messages_need_no_metadata(&turn.messages) {
875                turn.messages.clear();
876            }
877        }
878        if persisted.pending_attachments.is_empty()
879            && messages_need_no_metadata(&persisted.pending_messages)
880        {
881            persisted.pending_messages.clear();
882        }
883        let body = serde_json::to_string_pretty(&persisted).context("serialize talk")?;
884        let path = self.path_of(&t.id);
885        let tmp = path.with_extension("json.tmp");
886        write_atomic(&tmp, &path, &body)
887    }
888
889    /// Load a conversation by id or unambiguous id prefix.
890    pub fn get(&self, id: &str) -> Result<Talk> {
891        let resolved = self.resolve_id(id)?;
892        read_path(&self.path_of(&resolved))
893    }
894
895    /// Every conversation on disk: open first, then newest first, so what the
896    /// operator is still using belongs above what they are done with.
897    pub fn list(&self) -> Vec<Talk> {
898        self.list_counting_unreadable().0
899    }
900
901    /// [`Talks::list`], plus how many `*.json` files could not be read, so a
902    /// caller that reports to the operator can say what it skipped.
903    pub fn list_counting_unreadable(&self) -> (Vec<Talk>, usize) {
904        let mut unreadable = 0;
905        let mut all: Vec<Talk> = std::fs::read_dir(&self.root)
906            .into_iter()
907            .flatten()
908            .flatten()
909            .map(|e| e.path())
910            .filter(|p| p.extension().is_some_and(|x| x == "json"))
911            .filter_map(|p| {
912                let talk = read_path(&p).ok();
913                if talk.is_none() {
914                    unreadable += 1;
915                }
916                talk
917            })
918            .collect();
919        all.sort_unstable_by(|a, b| {
920            let rank = |t: &Talk| u8::from(!t.status.open());
921            rank(a).cmp(&rank(b)).then_with(|| b.id.cmp(&a.id))
922        });
923        (all, unreadable)
924    }
925
926    /// Expand an id prefix to exactly one conversation id.
927    pub fn resolve_id(&self, prefix: &str) -> Result<String> {
928        if self.path_of(prefix).is_file() {
929            return Ok(prefix.to_owned());
930        }
931        let hits: Vec<String> = self
932            .list()
933            .into_iter()
934            .map(|t| t.id)
935            .filter(|id| id.starts_with(prefix) || id.ends_with(prefix))
936            .collect();
937        match hits.len() {
938            1 => Ok(hits.into_iter().next().expect("exactly one hit")),
939            0 => bail!("no talk matches `{prefix}`"),
940            _ => bail!(
941                "`{prefix}` matches {} talks: {}",
942                hits.len(),
943                hits.join(", ")
944            ),
945        }
946    }
947
948    /// Change detection token: the newest modification time in the store, in
949    /// milliseconds.
950    pub fn revision(&self) -> u64 {
951        std::fs::read_dir(&self.root)
952            .into_iter()
953            .flatten()
954            .flatten()
955            .filter(|e| e.path().extension().is_none_or(|x| x != "turn"))
956            .filter(|e| !e.file_name().to_string_lossy().starts_with('.'))
957            .filter_map(|e| e.metadata().ok())
958            .filter_map(|m| m.modified().ok())
959            .filter_map(|t| t.duration_since(std::time::UNIX_EPOCH).ok())
960            .map(|d| d.as_millis() as u64)
961            .max()
962            .unwrap_or(0)
963    }
964
965    /// How many conversations are still open.
966    pub fn count_open(&self) -> usize {
967        self.list().iter().filter(|t| t.status.open()).count()
968    }
969
970    /// Where one conversation's turn lease lives. Not a `*.json`, so
971    /// [`Talks::list`] never sees it.
972    pub fn turn_path(&self, id: &str) -> PathBuf {
973        self.root.join(format!("{id}.turn"))
974    }
975
976    /// Claim the right to run one agent turn in `id`, across processes.
977    ///
978    /// `None` means somebody else holds a fresh lease - the caller reports
979    /// "a turn is already running" and must not start one. A lease is held
980    /// while its last beat is within [`crate::ask::LEASE_TTL`]; pids are never
981    /// consulted. A stale (or unreadable) lease is taken over: renamed to a
982    /// unique name first, so of several takers only the one whose rename wins
983    /// deletes it, and the claim itself is an exclusive create.
984    pub fn claim_turn(&self, id: &str) -> Result<Option<TurnLease>> {
985        self.claim_turn_at(id, Timestamp::now())
986    }
987
988    fn claim_turn_at(&self, id: &str, now: Timestamp) -> Result<Option<TurnLease>> {
989        std::fs::create_dir_all(&self.root)
990            .with_context(|| format!("create {}", self.root.display()))?;
991        let path = self.turn_path(id);
992        let token = fresh_token();
993        if create_turn(&path, &token, now)? {
994            return Ok(Some(TurnLease {
995                talk: id.to_owned(),
996                path,
997                token,
998            }));
999        }
1000        if lease_blocks(&path, now) {
1001            return Ok(None);
1002        }
1003        // Stale or unreadable. Every replacement happens under a short-lived
1004        // exclusive lock file, so two takers cannot each delete the other's
1005        // fresh lease. A taker that finds the lock held simply loses; the lock
1006        // itself ages out, so a taker that died inside it cannot wedge the talk.
1007        let Some(_lock) = TurnLock::take(&path)? else {
1008            return Ok(None);
1009        };
1010        if lease_blocks(&path, now) {
1011            return Ok(None);
1012        }
1013        let _ = std::fs::remove_file(&path);
1014        Ok(create_turn(&path, &token, now)?.then_some(TurnLease {
1015            talk: id.to_owned(),
1016            path,
1017            token,
1018        }))
1019    }
1020
1021    /// Is a turn running in `id` anywhere, by a fresh lease?
1022    pub fn turn_held(&self, id: &str) -> bool {
1023        read_turn(&self.turn_path(id)).is_some_and(|r| r.fresh(Timestamp::now()))
1024    }
1025
1026    /// Remove a conversation from disk, record and artifacts both. The
1027    /// operator's way of saying "not just done, gone" - [`close`] alone
1028    /// leaves the record as history.
1029    ///
1030    /// Takes [`Talks::guard`] for the same reason [`close`] does: a delete
1031    /// racing a [`record`] or the tail of [`turn`] must not land between
1032    /// their own read and write, or the file removed here would look, to
1033    /// them, like a record that simply has not been written yet. The other
1034    /// half of that story is on their side - both check under this same
1035    /// guard that the record they are about to write is still there, and
1036    /// give up without writing if it is not, which is what stops their `put`
1037    /// from resurrecting a conversation this call already removed.
1038    pub fn remove(&self, id: &str) -> Result<()> {
1039        let _guard = self.guard()?;
1040        let resolved = self.resolve_id(id)?;
1041        let path = self.path_of(&resolved);
1042        std::fs::remove_file(&path).with_context(|| format!("remove {}", path.display()))?;
1043        let artifacts = self.artifacts_of(&resolved);
1044        if artifacts.is_dir() {
1045            std::fs::remove_dir_all(&artifacts)
1046                .with_context(|| format!("remove {}", artifacts.display()))?;
1047        }
1048        let _ = std::fs::remove_file(self.turn_path(&resolved));
1049        Ok(())
1050    }
1051}
1052
1053/// What [`Talks::guard`] hands out: the in-process mutex plus the
1054/// cross-process file lock. The file lock is released first.
1055struct StoreGuard<'a> {
1056    _file: TurnLock,
1057    _answer: AnswerGuard<'a>,
1058}
1059
1060/// Holds conversation updates until the corresponding question write completes.
1061pub struct AnswerGuard<'a> {
1062    _file: TurnLock,
1063    _mutex: MutexGuard<'a, ()>,
1064}
1065
1066/// The body of a `<id>.turn` file. `pid` is for a human reading it; nothing
1067/// decides on it.
1068#[derive(Debug, Serialize, Deserialize)]
1069struct TurnRecord {
1070    token: String,
1071    pid: u32,
1072    beat_at: Timestamp,
1073}
1074
1075impl TurnRecord {
1076    fn fresh(&self, now: Timestamp) -> bool {
1077        now.as_second() - self.beat_at.as_second() <= crate::ask::LEASE_TTL.as_secs() as i64
1078    }
1079}
1080
1081/// A token no other caller in this process shares: `rng::entropy` is the
1082/// clock and the pid, so two threads in one clock tick would otherwise get the
1083/// same token, and with it the same temp file name and the same identity.
1084fn fresh_token() -> String {
1085    static SEQ: std::sync::atomic::AtomicU64 = std::sync::atomic::AtomicU64::new(0);
1086    let n = SEQ.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
1087    let seed = crate::rng::entropy() ^ n.wrapping_mul(0x9E37_79B9_7F4A_7C15);
1088    crate::rng::SplitMix64::new(seed).uuid_v4()
1089}
1090
1091fn read_turn(path: &Path) -> Option<TurnRecord> {
1092    serde_json::from_str(&std::fs::read_to_string(path).ok()?).ok()
1093}
1094
1095/// A takeover lock older than this belonged to a taker that died inside it.
1096const TAKEOVER_LOCK_TTL: Duration = Duration::from_secs(10);
1097
1098/// A break ticket older than this belonged to a taker that died holding it;
1099/// the next generation's ticket may then be tried for the same dead token.
1100const TICKET_TTL: Duration = Duration::from_secs(10);
1101
1102/// Most ticket generations tried for one dead token.
1103const TICKET_GENERATIONS: u32 = 16;
1104
1105/// Tickets are forgotten only this long after their last write.
1106const TICKET_SWEEP_AGE: Duration = Duration::from_secs(3600);
1107
1108/// Create `path` exclusively with `body`; `false` when it already exists.
1109fn create_exclusive(path: &Path, body: &str) -> Result<bool> {
1110    use std::io::Write as _;
1111    match std::fs::OpenOptions::new()
1112        .write(true)
1113        .create_new(true)
1114        .open(path)
1115    {
1116        Ok(mut f) => {
1117            if let Err(e) = f.write_all(body.as_bytes()) {
1118                drop(f);
1119                let _ = std::fs::remove_file(path);
1120                return Err(e).with_context(|| format!("write {}", path.display()));
1121            }
1122            Ok(true)
1123        }
1124        Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => Ok(false),
1125        Err(e) => Err(e).with_context(|| format!("create {}", path.display())),
1126    }
1127}
1128
1129fn create_turn(path: &Path, token: &str, now: Timestamp) -> Result<bool> {
1130    let record = TurnRecord {
1131        token: token.to_owned(),
1132        pid: std::process::id(),
1133        beat_at: now,
1134    };
1135    let body = serde_json::to_string(&record).context("serialize turn lease")?;
1136    // Written in full under a private name, then published exclusively: a
1137    // reader never sees a half-written lease.
1138    let tmp = path.with_extension(format!("turn.{token}.new"));
1139    publish_exclusive(path, &tmp, &body)
1140}
1141
1142/// Is there a lease at `path` that must not be taken over? A fresh beat, or
1143/// an empty file young enough to be a [`publish_exclusive`] placeholder whose
1144/// writer has not yet renamed the real record into place.
1145fn lease_blocks(path: &Path, now: Timestamp) -> bool {
1146    if read_turn(path).is_some_and(|r| r.fresh(now)) {
1147        return true;
1148    }
1149    std::fs::metadata(path).is_ok_and(|m| {
1150        m.len() == 0
1151            && m.modified()
1152                .ok()
1153                .and_then(|t| t.elapsed().ok())
1154                .is_some_and(|age| age <= TAKEOVER_LOCK_TTL)
1155    })
1156}
1157
1158/// Does this error mean the filesystem has no hard links (as opposed to an
1159/// ordinary I/O failure, which must not be hidden by a fallback)?
1160fn link_unsupported(e: &std::io::Error) -> bool {
1161    e.kind() == std::io::ErrorKind::Unsupported || (cfg!(windows) && e.raw_os_error() == Some(1))
1162}
1163
1164/// Publish the finished file `tmp` (holding `body`) at `path` unless `path`
1165/// exists; `false` when it does. `tmp` is always removed.
1166///
1167/// The first choice is `hard_link`, which is exclusive and atomic. Where hard
1168/// links are unsupported the fallback is an exclusive `create_new` of `path`
1169/// itself, written through that handle (never a `rename` over `path`, which
1170/// would let a stalled writer clobber a successor). Readers can briefly see
1171/// the file empty: [`lease_blocks`] treats a young empty one as held, and a
1172/// lock side treats it as an "invalid" token that cannot be broken for
1173/// [`TAKEOVER_LOCK_TTL`]. Residual windows, not closable by path alone: a
1174/// reader may see a partly written file for the length of one small write, and
1175/// a writer stalled between create and write for longer than the TTL can have
1176/// its file replaced; it then finds the path no longer carries its body and
1177/// reports `false`.
1178/// The no-hard-link path of [`publish_exclusive`]. A failed write never
1179/// removes `path`: by then it may belong to a successor, so the half-made file
1180/// is left to expire (a young empty one blocks, an unreadable one is stale).
1181fn create_in_place(path: &Path, body: &str) -> Result<bool> {
1182    use std::io::Write as _;
1183    let mut f = match std::fs::OpenOptions::new()
1184        .write(true)
1185        .create_new(true)
1186        .open(path)
1187    {
1188        Ok(f) => f,
1189        Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => return Ok(false),
1190        Err(e) => return Err(e).with_context(|| format!("create {}", path.display())),
1191    };
1192    f.write_all(body.as_bytes())
1193        .with_context(|| format!("write {}", path.display()))?;
1194    // Written through our own handle, so a stalled writer can only write into
1195    // its own file; confirm the path still carries it before claiming it.
1196    Ok(std::fs::read_to_string(path).is_ok_and(|t| t == body))
1197}
1198
1199fn publish_exclusive(path: &Path, tmp: &Path, body: &str) -> Result<bool> {
1200    std::fs::write(tmp, body).with_context(|| format!("write {}", tmp.display()))?;
1201    #[cfg(test)]
1202    let linked = if failpoint::no_link_forced() {
1203        Err(std::io::Error::from(std::io::ErrorKind::Unsupported))
1204    } else {
1205        std::fs::hard_link(tmp, path)
1206    };
1207    #[cfg(not(test))]
1208    let linked = std::fs::hard_link(tmp, path);
1209    let out = match linked {
1210        Ok(()) => Ok(true),
1211        Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => Ok(false),
1212        Err(e) if link_unsupported(&e) => create_in_place(path, body),
1213        Err(e) => Err(e).with_context(|| format!("create {}", path.display())),
1214    };
1215    let _ = std::fs::remove_file(tmp);
1216    out
1217}
1218
1219/// Remove `path` only if it carries exactly `expected`; `true` when it did.
1220///
1221/// There is no atomic compare-and-unlink on a path, so the file is first
1222/// renamed to a private name, which makes the read and the removal act on the
1223/// same inode. If it carried something else (a newer holder published between
1224/// the caller's read and now) it is put back exclusively, so the wrong lock is
1225/// off the path for microseconds instead of deleted. Residual window: in those
1226/// microseconds a fourth party can publish at `path`; the restore then loses,
1227/// the displaced file is dropped (with a warning) and two holders can overlap
1228/// until the survivor's TTL. Path operations alone cannot close that.
1229fn remove_if_carries(path: &Path, expected: &str) -> bool {
1230    let gone = path.with_extension(format!("gone.{}", fresh_token()));
1231    if std::fs::rename(path, &gone).is_err() {
1232        return false;
1233    }
1234    let found = std::fs::read_to_string(&gone);
1235    if found.as_ref().is_ok_and(|t| t == expected) {
1236        let _ = std::fs::remove_file(&gone);
1237        return true;
1238    }
1239    if let Ok(body) = found {
1240        let back = path.with_extension(format!("back.{}", fresh_token()));
1241        match publish_exclusive(path, &back, &body) {
1242            Ok(true) => {}
1243            Ok(false) | Err(_) => tracing::warn!(
1244                "{} was replaced while a stale removal had it aside; \
1245                 the displaced file is dropped",
1246                path.display()
1247            ),
1248        }
1249    }
1250    let _ = std::fs::remove_file(&gone);
1251    false
1252}
1253
1254/// The short exclusive lock every change to an existing lease (takeover, beat,
1255/// release) happens under, so none of them can act on a stale reading. It
1256/// ages out after [`TAKEOVER_LOCK_TTL`] in case its holder died inside it.
1257struct TurnLock {
1258    path: PathBuf,
1259    token: String,
1260}
1261
1262impl TurnLock {
1263    fn token() -> String {
1264        fresh_token()
1265    }
1266
1267    /// Publish `token` at `path` complete (written privately, then published
1268    /// exclusively), so nobody reads a half-written token; `false` if `path`
1269    /// exists.
1270    fn publish(path: &Path, token: &str) -> Result<bool> {
1271        let tmp = path.with_extension(format!("lock.{token}.new"));
1272        publish_exclusive(path, &tmp, token)
1273    }
1274
1275    fn take(lease: &Path) -> Result<Option<Self>> {
1276        let path = lease.with_extension("turn.lock");
1277        let token = Self::token();
1278        if Self::publish(&path, &token)? {
1279            return Ok(Some(Self { path, token }));
1280        }
1281        let Ok(seen) = std::fs::read_to_string(&path) else {
1282            return Ok(None);
1283        };
1284        // A lock that is not a whole token (an older build wrote it in two
1285        // steps) is breakable too, under one fixed ticket name.
1286        let key = if !seen.is_empty() && seen.chars().all(|c| c.is_ascii_alphanumeric() || c == '-')
1287        {
1288            seen.as_str()
1289        } else {
1290            "invalid"
1291        };
1292        let aged = std::fs::metadata(&path)
1293            .and_then(|m| m.modified())
1294            .ok()
1295            .and_then(|t| t.elapsed().ok())
1296            .is_some_and(|age| age > TAKEOVER_LOCK_TTL);
1297        if !aged {
1298            return Ok(None);
1299        }
1300        // Breaking a dead lock is decided by a ticket named after the token
1301        // that was judged dead: exactly one taker per generation can create
1302        // it. The lock itself is never moved aside, so a fresh lock made by
1303        // the winner is never off the path, not even for an instant.
1304        //
1305        // The ticket name carries a generation, not the wall clock, so no
1306        // time boundary can let a second taker in while a ticket is alive.
1307        // Generation n+1 is tried only when ticket n is older than
1308        // `TICKET_TTL` (its owner died). Residual risk, same kind as the lock's
1309        // own TTL: an owner stalled past `TICKET_TTL` between creating its
1310        // ticket and publishing can overlap with the next generation's taker.
1311        let mut won = false;
1312        for n in 0..TICKET_GENERATIONS {
1313            let ticket = path.with_extension(format!("lock.{key}.break.{n}"));
1314            if create_exclusive(&ticket, "")? {
1315                won = true;
1316                break;
1317            }
1318            let stale = std::fs::metadata(&ticket)
1319                .and_then(|m| m.modified())
1320                .ok()
1321                .and_then(|t| t.elapsed().ok())
1322                .is_some_and(|age| age > TICKET_TTL);
1323            if !stale {
1324                return Ok(None);
1325            }
1326        }
1327        if !won {
1328            // Every generation was abandoned. A ticket is never unlinked by
1329            // name while it may still be someone's exclusion: only the sweep
1330            // removes tickets, and only ones far older than any taker lives,
1331            // so recovery resumes once they age out.
1332            Self::sweep_tickets(&path);
1333            return Ok(None);
1334        }
1335        Self::sweep_tickets(&path);
1336        // Only the ticket's owner reaches this point for `seen`, but the
1337        // lock's own holder may still release it (its `Drop`) and a third
1338        // taker publish a new one meanwhile, so the removal is verified
1339        // against the content (see [`remove_if_carries`] for the window that
1340        // remains); if the lock changed, leave it.
1341        if !remove_if_carries(&path, &seen) {
1342            return Ok(None);
1343        }
1344        if Self::publish(&path, &token)? {
1345            return Ok(Some(Self { path, token }));
1346        }
1347        Ok(None)
1348    }
1349
1350    /// Forget tickets far older than the lock's TTL. They stay that long so a
1351    /// slow taker that read the same dead token cannot break it a second time.
1352    fn sweep_tickets(path: &Path) {
1353        let (Some(dir), Some(name)) = (path.parent(), path.file_name().and_then(|n| n.to_str()))
1354        else {
1355            return;
1356        };
1357        let prefix = format!("{name}.");
1358        let Ok(entries) = std::fs::read_dir(dir) else {
1359            return;
1360        };
1361        for entry in entries.flatten() {
1362            let file = entry.file_name();
1363            let Some(file) = file.to_str() else { continue };
1364            if !(file.starts_with(&prefix) && file.contains(".break.")) {
1365                continue;
1366            }
1367            let old = entry
1368                .metadata()
1369                .and_then(|m| m.modified())
1370                .ok()
1371                .and_then(|t| t.elapsed().ok())
1372                .is_some_and(|age| age > TICKET_SWEEP_AGE);
1373            if old {
1374                let _ = std::fs::remove_file(entry.path());
1375            }
1376        }
1377    }
1378
1379    /// Wait briefly for the lock; the holders only do a read and a write.
1380    fn take_patiently(lease: &Path) -> Option<Self> {
1381        for _ in 0..50 {
1382            match Self::take(lease) {
1383                Ok(Some(lock)) => return Some(lock),
1384                Ok(None) => std::thread::sleep(Duration::from_millis(10)),
1385                Err(_) => return None,
1386            }
1387        }
1388        None
1389    }
1390}
1391
1392impl Drop for TurnLock {
1393    fn drop(&mut self) {
1394        // Only our own lock: one that aged out and was taken over is not ours.
1395        remove_if_carries(&self.path, &self.token);
1396    }
1397}
1398
1399/// How often a running turn renews its lease: well inside
1400/// [`crate::ask::LEASE_TTL`].
1401pub const TURN_BEAT: Duration = Duration::from_secs(20);
1402
1403/// One talk's cross-process turn slot, released on drop (success, error,
1404/// panic or a dropped handler future alike). Release only removes the file
1405/// while it still carries this lease's token, so a guard that outlived its
1406/// own expiry cannot delete the lease of whoever took over.
1407#[derive(Debug)]
1408pub struct TurnLease {
1409    talk: String,
1410    path: PathBuf,
1411    token: String,
1412}
1413
1414impl TurnLease {
1415    /// Does the lease file still carry this lease's token? A read only.
1416    pub fn holds(&self) -> bool {
1417        read_turn(&self.path).is_some_and(|r| r.token == self.token)
1418    }
1419
1420    /// Renew the lease. `Ok(false)` means it was taken over or removed, so
1421    /// this turn no longer owns the slot; `Err` is a transient failure (the
1422    /// lock stayed busy, a write failed) and the next beat tries again.
1423    pub fn beat(&self) -> Result<bool> {
1424        let _lock = TurnLock::take_patiently(&self.path)
1425            .with_context(|| format!("lock {} to renew it", self.path.display()))?;
1426        let Some(mut record) = read_turn(&self.path).filter(|r| r.token == self.token) else {
1427            return Ok(false);
1428        };
1429        record.beat_at = Timestamp::now();
1430        let body = serde_json::to_string(&record).context("serialize turn lease")?;
1431        let tmp = self.path.with_extension(format!("turn.{}.tmp", self.token));
1432        write_atomic(&tmp, &self.path, &body)?;
1433        Ok(true)
1434    }
1435
1436    /// Run `fut` while renewing this lease every [`TURN_BEAT`]. A transient
1437    /// beat failure is logged and the turn goes on; a lost lease drops `fut`
1438    /// (the turn stops) and is an error, since somebody else may now be
1439    /// running the same conversation.
1440    pub async fn beating<T>(&self, fut: impl std::future::Future<Output = T>) -> Result<T> {
1441        self.beating_every(TURN_BEAT, fut).await
1442    }
1443
1444    async fn beating_every<T>(
1445        &self,
1446        period: Duration,
1447        fut: impl std::future::Future<Output = T>,
1448    ) -> Result<T> {
1449        tokio::pin!(fut);
1450        loop {
1451            match tokio::time::timeout(period, &mut fut).await {
1452                Ok(out) => return Ok(out),
1453                Err(_) => match self.beat() {
1454                    Ok(true) => {}
1455                    Ok(false) => bail!(
1456                        "the turn lease {} was taken over; this turn is stopped",
1457                        self.path.display()
1458                    ),
1459                    Err(e) => tracing::warn!("{e:#}"),
1460                },
1461            }
1462        }
1463    }
1464}
1465
1466impl Drop for TurnLease {
1467    fn drop(&mut self) {
1468        // Under the lock, so a takeover cannot slip in between the check and
1469        // the removal. If the lock stays busy the lease just ages out.
1470        if let Some(_lock) = TurnLock::take_patiently(&self.path) {
1471            if read_turn(&self.path).is_some_and(|r| r.token == self.token) {
1472                let _ = std::fs::remove_file(&self.path);
1473            }
1474        }
1475    }
1476}
1477
1478/// Open a conversation. Takes no agent turn: there is no idea to answer yet,
1479/// and a conversation the operator has not said anything into yet is a
1480/// normal, valid thing to have sitting on the phone.
1481///
1482/// `agent` beats `[roles] chatter`, which beats [`agent::pick`]'s own default
1483/// order (a claude seat, else the first runnable agent in roster order) when
1484/// nothing names a seat at all - see `[roles] chatter`'s own doc in
1485/// [`crate::config`] for why a dedicated field exists rather than reusing a
1486/// judge seat.
1487pub fn begin(store: &Talks, cfg: &Config, repo: PathBuf, agent: Option<&str>) -> Result<Talk> {
1488    begin_with(store, cfg, repo, agent, &Preferred::default())
1489}
1490
1491/// Remembered choices for a new conversation (`POST /api/talks`). Every field
1492/// is soft: one that does not fit `cfg` is dropped on its own and the default
1493/// stands, so a stale remembered value never stops a conversation opening.
1494#[derive(Debug, Default, Clone)]
1495pub struct Preferred {
1496    /// Remembered agent id.
1497    pub agent: Option<String>,
1498    /// Remembered persona id.
1499    pub persona: Option<String>,
1500    /// Remembered implementer count.
1501    pub implementers: Option<i64>,
1502}
1503
1504/// [`begin`] with remembered choices. An explicit `agent` wins over
1505/// `preferred.agent` and keeps its strict error. A preferred agent that is on
1506/// the roster and runnable is an explicit choice (`fallback = false`) unless
1507/// it is the chatter chain's own head. Persona and implementers are applied
1508/// before the single save, so neither is marked dirty: the new seat's briefing
1509/// already carries them.
1510pub fn begin_with(
1511    store: &Talks,
1512    cfg: &Config,
1513    repo: PathBuf,
1514    agent: Option<&str>,
1515    preferred: &Preferred,
1516) -> Result<Talk> {
1517    // Absolute: a relative path means the wrong repository once anything
1518    // other than this process reads it back.
1519    let repo = repo.canonicalize().unwrap_or(repo);
1520    // An explicit agent is a chain of one; otherwise the first id of
1521    // `[roles] chatter` that can run here (later ones are `turn`'s fallbacks).
1522    let head = || -> Result<AgentSpec> {
1523        Ok(agent::pick_chain(
1524            &cfg.agents,
1525            cfg.roles.chatter.as_ref(),
1526            &agent::installed,
1527            "chatter",
1528        )?
1529        .remove(0))
1530    };
1531    let mut fallback = agent.is_none();
1532    let spec = match agent {
1533        Some(id) => agent::pick(&cfg.agents, Some(id), &agent::installed)?,
1534        None => {
1535            match preferred
1536                .agent
1537                .as_deref()
1538                .map(str::trim)
1539                .filter(|id| !id.is_empty())
1540            {
1541                Some(id) => match agent::pick(&cfg.agents, Some(id), &agent::installed) {
1542                    Ok(spec) => {
1543                        fallback = head().is_ok_and(|head| spec.id == head.id);
1544                        spec
1545                    }
1546                    Err(_) => head()?,
1547                },
1548                None => head()?,
1549            }
1550        }
1551    };
1552    let persona = preferred
1553        .persona
1554        .as_deref()
1555        .and_then(|id| crate::persona::find(&cfg.talk.personas, id))
1556        .filter(|p| !p.is_default())
1557        .map(|p| p.id)
1558        .unwrap_or_default();
1559    let implementers = preferred
1560        .implementers
1561        .and_then(|n| u8::try_from(n).ok())
1562        .and_then(|n| check_implementers(n, cfg).ok())
1563        .unwrap_or(1);
1564
1565    let now = Timestamp::now();
1566    let mut talk = Talk {
1567        recommendations: Default::default(),
1568        posts_read_through: Some(0),
1569        pending_breaks: None,
1570        pending_messages: Vec::new(),
1571        schema: SCHEMA,
1572        id: new_id(),
1573        repo,
1574        agent: spec.id.clone(),
1575        status: TalkStatus::Open,
1576        turns: Vec::new(),
1577        pending: String::new(),
1578        pending_attachments: Vec::new(),
1579        fallback,
1580        persona,
1581        persona_dirty: false,
1582        implementers,
1583        implementers_dirty: false,
1584        notified: Vec::new(),
1585        completion_notices: Vec::new(),
1586        created_at: now,
1587        updated_at: now,
1588        seat: SeatState::new(SEAT, &spec.id, crate::rng::entropy()),
1589    };
1590    store.put(&mut talk)?;
1591    Ok(talk)
1592}
1593
1594/// Append the operator's turn and flush it, without invoking anything.
1595///
1596/// Split out of [`say`] so `POST /api/talks/{id}/say` can answer once the
1597/// message is safely on disk, and run the agent's half in the background -
1598/// holding the connection for a turn that can run fifteen minutes is the
1599/// wrong shape for a phone.
1600pub fn record(
1601    talk: &mut Talk,
1602    store: &Talks,
1603    text: &str,
1604    attachments: Vec<Attachment>,
1605) -> Result<String> {
1606    record_reply(talk, store, text, attachments, None)
1607}
1608
1609/// Record a send with a reply validated under the persistence lock.
1610pub fn record_reply(
1611    talk: &mut Talk,
1612    store: &Talks,
1613    text: &str,
1614    attachments: Vec<Attachment>,
1615    target: Option<ReplyTarget>,
1616) -> Result<String> {
1617    // `web::talk_say` reads the talk, then awaits config discovery before
1618    // calling this - a gap a concurrent `POST /api/talks/{id}/close` can land
1619    // in. The guard held for the rest of this function is what actually closes
1620    // that gap: re-reading status without it only shrinks the window a
1621    // concurrent `close` could land in between this call's own read and its
1622    // `put`, it does not remove it. See [`Talks::guard`] and the matching
1623    // guard in `turn`, which this mirrors.
1624    let _guard = store.guard()?;
1625    // A concurrent `Talks::remove` can have landed in that same gap. `put`
1626    // writes unconditionally, so trusting the stale `talk` here would recreate
1627    // the file a delete just removed - the record must still be there for a
1628    // turn to have anywhere to append to.
1629    let Ok(fresh) = store.get(&talk.id) else {
1630        bail!("talk {} was deleted", talk.short());
1631    };
1632    let reply = resolve_reply(&fresh, target)?;
1633    talk.turns = fresh.turns;
1634    talk.status = fresh.status;
1635    // Do not let this older handle overwrite a draft accepted while it was
1636    // waiting for configuration discovery.
1637    talk.pending = fresh.pending;
1638    talk.pending_breaks = fresh.pending_breaks;
1639    talk.pending_messages = fresh.pending_messages;
1640    talk.pending_attachments = fresh.pending_attachments;
1641    talk.notified = fresh.notified;
1642    if !talk.status.open() {
1643        bail!(
1644            "talk {} is {} and takes no more turns",
1645            talk.short(),
1646            talk.status.as_str()
1647        );
1648    }
1649    // Keep Markdown indentation in owner evidence; trim only to check emptiness.
1650    if text.trim().is_empty() && attachments.is_empty() {
1651        bail!("nothing to say");
1652    }
1653    talk.turns.push(Turn {
1654        sender: Some(Sender::Operator),
1655        who: Who::Operator,
1656        body: text.to_owned(),
1657        at: Timestamp::now(),
1658        messages: vec![Message {
1659            display: None,
1660            sender: Some(Sender::Operator),
1661            body: text.to_owned(),
1662            attachments: attachments.clone(),
1663            reply,
1664        }],
1665        attachments,
1666        usage: None,
1667        breaks: Some(Vec::new()),
1668        posted_by: None,
1669    });
1670    store.put(talk)?;
1671    Ok(text.to_owned())
1672}
1673
1674/// Add an unrecorded message to the durable draft while another turn runs.
1675pub fn queue(
1676    talk: &mut Talk,
1677    store: &Talks,
1678    text: &str,
1679    attachments: Vec<Attachment>,
1680) -> Result<()> {
1681    queue_reply(talk, store, text, attachments, None)
1682}
1683
1684/// Queue a send with its own reply relationship and attachments.
1685pub fn queue_reply(
1686    talk: &mut Talk,
1687    store: &Talks,
1688    text: &str,
1689    attachments: Vec<Attachment>,
1690    target: Option<ReplyTarget>,
1691) -> Result<()> {
1692    queue_keyed(
1693        talk,
1694        store,
1695        text,
1696        attachments,
1697        None,
1698        &|_| false,
1699        (target, Sender::Operator, None),
1700    )
1701    .map(|_| ())
1702}
1703
1704/// [`queue`] for a message that must be accepted at most once: `key` is
1705/// recorded in [`Talk::notified`] in the same write as the draft, so a retry
1706/// after a crash (or a second sweep) finds the key and queues nothing.
1707/// Returns whether this call queued it.
1708pub fn queue_once(talk: &mut Talk, store: &Talks, text: &str, key: &str) -> Result<bool> {
1709    queue_once_matching(talk, store, text, key, &|_| false, None)
1710}
1711
1712/// Accept a key once, also recognizing older equivalent keys under the same
1713/// store lock as the draft write.
1714pub(crate) fn queue_once_matching(
1715    talk: &mut Talk,
1716    store: &Talks,
1717    text: &str,
1718    key: &str,
1719    previous: &dyn Fn(&str) -> bool,
1720    reference: Option<&crate::chat_notify::CompletionReference>,
1721) -> Result<bool> {
1722    queue_keyed(
1723        talk,
1724        store,
1725        text,
1726        Vec::new(),
1727        Some((key, reference)),
1728        previous,
1729        (None, Sender::Magi, None),
1730    )
1731}
1732
1733/// Persist display facts with the notice, reference and deduplication key under one lock.
1734pub(crate) fn queue_notice(
1735    talk: &mut Talk,
1736    store: &Talks,
1737    text: &str,
1738    key: &str,
1739    previous: &dyn Fn(&str) -> bool,
1740    reference: Option<&crate::chat_notify::CompletionReference>,
1741    display: crate::prompt::NoticeDisplay,
1742) -> Result<bool> {
1743    queue_keyed(
1744        talk,
1745        store,
1746        text,
1747        Vec::new(),
1748        Some((key, reference)),
1749        previous,
1750        (None, Sender::Magi, Some(display)),
1751    )
1752}
1753
1754/// Queue an injected message without changing its input role or lease handling.
1755pub fn queue_magi(talk: &mut Talk, store: &Talks, text: &str) -> Result<()> {
1756    queue_magi_display(talk, store, text, None)
1757}
1758
1759/// Save a user-facing snapshot alongside an injected message's agent input.
1760pub(crate) fn queue_magi_display(
1761    talk: &mut Talk,
1762    store: &Talks,
1763    text: &str,
1764    display: Option<crate::prompt::NoticeDisplay>,
1765) -> Result<()> {
1766    queue_keyed(
1767        talk,
1768        store,
1769        text,
1770        Vec::new(),
1771        None,
1772        &|_| false,
1773        (None, Sender::Magi, display),
1774    )
1775    .map(|_| ())
1776}
1777
1778fn queue_keyed(
1779    talk: &mut Talk,
1780    store: &Talks,
1781    text: &str,
1782    attachments: Vec<Attachment>,
1783    key: Option<(&str, Option<&crate::chat_notify::CompletionReference>)>,
1784    previous: &dyn Fn(&str) -> bool,
1785    source: (
1786        Option<ReplyTarget>,
1787        Sender,
1788        Option<crate::prompt::NoticeDisplay>,
1789    ),
1790) -> Result<bool> {
1791    // Keep Markdown indentation in owner evidence; trim only to check emptiness.
1792    if text.trim().is_empty() && attachments.is_empty() {
1793        bail!("nothing to say");
1794    }
1795    let _guard = store.guard()?;
1796    let mut fresh = store
1797        .get(&talk.id)
1798        .with_context(|| format!("talk {} was deleted", talk.short()))?;
1799    if !fresh.status.open() {
1800        bail!(
1801            "talk {} is {} and takes no more turns",
1802            fresh.short(),
1803            fresh.status.as_str()
1804        );
1805    }
1806    if !append_pending(&mut fresh, text, attachments, key, previous, source)? {
1807        *talk = fresh;
1808        return Ok(false);
1809    }
1810    store.put(&mut fresh)?;
1811    *talk = fresh;
1812    Ok(true)
1813}
1814
1815/// Dispatch consultation under both store locks, including reopening, without
1816/// ever waiting while the caller holds the question lock. None means retry
1817/// after releasing that lock and rechecking the question's current state.
1818pub(crate) fn try_queue_consult(
1819    store: &Talks,
1820    id: &str,
1821    text: &str,
1822    display: Option<crate::prompt::NoticeDisplay>,
1823) -> Result<Option<()>> {
1824    let Some(_guard) = store.try_guard()? else {
1825        return Ok(None);
1826    };
1827    let mut fresh = store.get(id).context("read consultation destination")?;
1828    fresh.status = TalkStatus::Open;
1829    append_pending(
1830        &mut fresh,
1831        text,
1832        Vec::new(),
1833        None,
1834        &|_| false,
1835        (None, Sender::Magi, display),
1836    )?;
1837    store.put(&mut fresh)?;
1838    Ok(Some(()))
1839}
1840
1841fn append_pending(
1842    fresh: &mut Talk,
1843    text: &str,
1844    attachments: Vec<Attachment>,
1845    key: Option<(&str, Option<&crate::chat_notify::CompletionReference>)>,
1846    previous: &dyn Fn(&str) -> bool,
1847    source: (
1848        Option<ReplyTarget>,
1849        Sender,
1850        Option<crate::prompt::NoticeDisplay>,
1851    ),
1852) -> Result<bool> {
1853    if let Some((key, reference)) = key {
1854        if fresh.notified.iter().any(|k| k == key || previous(k)) {
1855            return Ok(false);
1856        }
1857        fresh.notified.push(key.to_owned());
1858        if let Some(reference) = reference {
1859            fresh.completion_notices.push(reference.clone());
1860        }
1861    }
1862    let (target, provenance, display) = source;
1863    let reply = resolve_reply(fresh, target)?;
1864    if fresh.pending_messages.is_empty()
1865        && (!fresh.pending.is_empty() || !fresh.pending_attachments.is_empty())
1866    {
1867        fresh.pending_messages.push(Message {
1868            display: None,
1869            sender: None,
1870            body: fresh.pending.clone(),
1871            attachments: fresh.pending_attachments.clone(),
1872            reply: None,
1873        });
1874    }
1875    fresh.pending_messages.push(Message {
1876        display,
1877        sender: Some(provenance),
1878        body: text.to_owned(),
1879        attachments: attachments.clone(),
1880        reply,
1881    });
1882    if !text.is_empty() {
1883        if fresh.pending.is_empty() {
1884            fresh.pending = text.to_owned();
1885            fresh.pending_breaks = Some(Vec::new());
1886        } else {
1887            fresh.pending.push_str("\n\n");
1888            if let Some(b) = fresh.pending_breaks.as_mut() {
1889                b.push(fresh.pending.len());
1890            }
1891            fresh.pending.push_str(text);
1892        }
1893    }
1894    fresh.pending_attachments.extend(attachments);
1895    Ok(true)
1896}
1897
1898/// Append an agent-authored message to the transcript without starting a
1899/// turn (`magi talk post`). `by` is the posting seat's run id. No draft, no
1900/// lease, no agent call: it only writes under the store guard, onto the
1901/// freshest record, so it cannot lose a concurrent reply.
1902pub fn post(store: &Talks, talk_id: &str, body: &str, by: &str) -> Result<()> {
1903    let body = body.trim();
1904    if body.is_empty() {
1905        bail!("nothing to post");
1906    }
1907    let _guard = store.guard()?;
1908    let mut fresh = store
1909        .get(talk_id)
1910        .with_context(|| format!("talk {talk_id} does not exist"))?;
1911    if !fresh.status.open() {
1912        bail!(
1913            "talk {} is closed and takes no more messages",
1914            fresh.short()
1915        );
1916    }
1917    fresh.turns.push(Turn {
1918        sender: Some(Sender::Magi),
1919        who: Who::Agent,
1920        body: body.to_owned(),
1921        at: Timestamp::now(),
1922        attachments: Vec::new(),
1923        usage: None,
1924        breaks: None,
1925        messages: Vec::new(),
1926        posted_by: Some(by.to_owned()),
1927    });
1928    store.put(&mut fresh)
1929}
1930
1931/// Infer the old format's boundary only until an explicit one is saved.
1932fn posts_read_through(talk: &Talk) -> usize {
1933    talk.posts_read_through.unwrap_or_else(|| {
1934        talk.turns
1935            .iter()
1936            .rposition(|t| t.who != Who::Operator && t.posted_by.is_none())
1937            .map_or(0, |i| i + 1)
1938    })
1939}
1940
1941/// Posts not yet included in a successful chat prompt, in transcript order.
1942pub fn posts_since_last_reply(talk: &Talk) -> Option<String> {
1943    let posts: Vec<&Turn> = talk
1944        .turns
1945        .iter()
1946        .skip(posts_read_through(talk))
1947        .filter(|t| t.posted_by.is_some())
1948        .collect();
1949    if posts.is_empty() {
1950        return None;
1951    }
1952    let mut out = String::from("# Posted by a task\n\n");
1953    for t in posts {
1954        out.push_str(&format!(
1955            "Task seat {} posted to this conversation: {}\n\n",
1956            short(t.posted_by.as_deref().unwrap_or_default()),
1957            t.body
1958        ));
1959    }
1960    Some(out)
1961}
1962
1963/// Promote the current durable draft to one operator turn.
1964pub fn drain(talk: &mut Talk, store: &Talks) -> Result<Option<String>> {
1965    let _guard = store.guard()?;
1966    let mut fresh = store
1967        .get(&talk.id)
1968        .with_context(|| format!("talk {} was deleted", talk.short()))?;
1969    if !fresh.status.open() || (fresh.pending.is_empty() && fresh.pending_attachments.is_empty()) {
1970        *talk = fresh;
1971        return Ok(None);
1972    }
1973    let text = std::mem::take(&mut fresh.pending);
1974    let attachments = std::mem::take(&mut fresh.pending_attachments);
1975    let breaks = fresh.pending_breaks.take();
1976    let messages = std::mem::take(&mut fresh.pending_messages);
1977    fresh.turns.push(Turn {
1978        sender: Some(Sender::Operator),
1979        who: Who::Operator,
1980        body: text.clone(),
1981        at: Timestamp::now(),
1982        attachments,
1983        usage: None,
1984        breaks,
1985        messages,
1986        posted_by: None,
1987    });
1988    store.put(&mut fresh)?;
1989    *talk = fresh;
1990    Ok(Some(text))
1991}
1992
1993/// One operator turn and one agent turn, appended - the synchronous form, used
1994/// by tests and by anything that is fine waiting out the turn itself.
1995pub async fn say(
1996    lease: &TurnLease,
1997    talk: &mut Talk,
1998    store: &Talks,
1999    cfg: &Config,
2000    text: &str,
2001    attachments: Vec<Attachment>,
2002) -> Result<()> {
2003    check_lease(lease, talk)?;
2004    let text = record(talk, store, text, attachments)?;
2005    respond(lease, talk, store, cfg, &text).await
2006}
2007
2008/// The agent's half of a turn: invoke, append, flush. Pairs with [`record`].
2009///
2010/// Requires the talk's [`TurnLease`] (from [`Talks::claim_turn`]), so no
2011/// starter can run a turn without holding the cross-process slot. The lease is
2012/// renewed while the turn runs; a lease lost mid-turn stops it with an error.
2013pub async fn respond(
2014    lease: &TurnLease,
2015    talk: &mut Talk,
2016    store: &Talks,
2017    cfg: &Config,
2018    text: &str,
2019) -> Result<()> {
2020    check_lease(lease, talk)?;
2021    lease
2022        .beating(turn(talk, store, cfg, text))
2023        .await
2024        .and_then(|done| done)
2025}
2026
2027fn check_lease(lease: &TurnLease, talk: &Talk) -> Result<()> {
2028    if lease.talk != talk.id {
2029        bail!("the turn lease is for talk {}, not {}", lease.talk, talk.id);
2030    }
2031    // A lease that aged out and was taken over must not start a turn (or
2032    // write the operator's text) at all.
2033    if !lease.holds() {
2034        bail!("the turn lease for talk {} is no longer held", talk.short());
2035    }
2036    Ok(())
2037}
2038
2039/// Close a conversation. Idempotent: closing an already-closed conversation is
2040/// not an error, since the operator's intent - "I am done with this" - is
2041/// already satisfied.
2042///
2043/// Re-reads the record under [`Talks::guard`] rather than trusting the
2044/// caller's copy of `talk`, and writes that fresh copy back rather than the
2045/// one passed in. `web::talk_close` loads `talk` and calls this right after
2046/// with no gap of its own, but without the guard that load can still land
2047/// between a `record` or `turn` elsewhere reading the file and writing it
2048/// back - and a close built on the older snapshot would put it right back,
2049/// silently dropping whatever turn the other call had just appended.
2050///
2051/// If the re-read fails, this errors rather than falling back to the
2052/// caller's stale copy: `talk::begin` always `put`s the record before handing
2053/// out a `Talk`, so the only way a re-read can fail is a concurrent
2054/// [`Talks::remove`] having deleted it, and writing the stale copy back would
2055/// resurrect exactly what that delete removed.
2056pub fn close(talk: &mut Talk, store: &Talks) -> Result<()> {
2057    let _guard = store.guard()?;
2058    let mut fresh = store
2059        .get(&talk.id)
2060        .with_context(|| format!("talk {} was deleted", talk.short()))?;
2061    fresh.status = TalkStatus::Closed;
2062    // A closed conversation must not replay a draft if it is reopened later.
2063    fresh.pending.clear();
2064    fresh.pending_breaks = None;
2065    fresh.pending_messages.clear();
2066    fresh.pending_attachments.clear();
2067    store.put(&mut fresh)?;
2068    *talk = fresh;
2069    Ok(())
2070}
2071
2072/// Reopen a closed conversation. Idempotent for the same reason [`close`] is:
2073/// reopening an already-open conversation is not an error, since the
2074/// operator's intent - "I want to keep talking about this" - is already
2075/// satisfied.
2076///
2077/// Written symmetrically with [`close`]: re-reads the record under
2078/// [`Talks::guard`] rather than trusting the caller's copy of `talk`, writes
2079/// that fresh copy back rather than the one passed in, and errors rather than
2080/// falling back to the stale copy if the re-read fails, for the same reasons
2081/// `close`'s doc gives.
2082pub fn reopen(talk: &mut Talk, store: &Talks) -> Result<()> {
2083    let _guard = store.guard()?;
2084    let mut fresh = store
2085        .get(&talk.id)
2086        .with_context(|| format!("talk {} was deleted", talk.short()))?;
2087    fresh.status = TalkStatus::Open;
2088    store.put(&mut fresh)?;
2089    *talk = fresh;
2090    Ok(())
2091}
2092
2093/// Hand the conversation to another roster agent.
2094///
2095/// A CLI session belongs to one CLI and cannot be carried to another, so the
2096/// seat is minted afresh rather than edited: the next turn finds
2097/// `seat.turns == 0` and re-sends the transcript, since the new agent has
2098/// heard none of it. A magi-written note records the change in the
2099/// transcript. Returns `false` (and writes nothing, not even a note) when the
2100/// stored talk already uses `spec`.
2101///
2102/// Re-reads under [`Talks::guard`], like [`close`], and errors rather than
2103/// resurrecting a record a concurrent delete removed. Refusing a closed talk
2104/// or a turn in flight is the caller's job: only it can see the latter.
2105pub fn switch_agent(talk: &mut Talk, store: &Talks, spec: &AgentSpec) -> Result<bool> {
2106    let _guard = store.guard()?;
2107    let mut fresh = store
2108        .get(&talk.id)
2109        .with_context(|| format!("talk {} was deleted", talk.short()))?;
2110    if fresh.agent == spec.id {
2111        *talk = fresh;
2112        return Ok(false);
2113    }
2114    let from = std::mem::replace(&mut fresh.agent, spec.id.clone());
2115    fresh.seat = SeatState::new(SEAT, &spec.id, crate::rng::entropy());
2116    // A deliberate switch pins the conversation to the agent chosen.
2117    fresh.fallback = false;
2118    fresh.turns.push(Turn {
2119        sender: Some(Sender::Magi),
2120        breaks: None,
2121        messages: Vec::new(),
2122        posted_by: None,
2123        who: Who::Agent,
2124        body: format!("{MAGI_NOTE}agent changed from {from} to {}", spec.id),
2125        at: Timestamp::now(),
2126        attachments: Vec::new(),
2127        usage: None,
2128    });
2129    store.put(&mut fresh)?;
2130    *talk = fresh;
2131    Ok(true)
2132}
2133
2134/// Pick the conversation's persona (`""` or `default` is the plain voice).
2135///
2136/// The CLI session keeps its seat: the change is recorded with
2137/// [`Talk::persona_dirty`] and a magi note, and the next turn tells the session
2138/// about it (or its fresh briefing already does). Returns `false` and writes
2139/// nothing when the stored persona is already `id`. Refusing a closed talk, an
2140/// unknown id or a turn in flight is the caller's job.
2141pub fn switch_persona(talk: &mut Talk, store: &Talks, id: &str) -> Result<bool> {
2142    let id = if id.trim() == crate::persona::DEFAULT_ID {
2143        ""
2144    } else {
2145        id.trim()
2146    };
2147    let _guard = store.guard()?;
2148    let mut fresh = store
2149        .get(&talk.id)
2150        .with_context(|| format!("talk {} was deleted", talk.short()))?;
2151    if fresh.persona == id {
2152        *talk = fresh;
2153        return Ok(false);
2154    }
2155    let from = std::mem::replace(&mut fresh.persona, id.to_owned());
2156    fresh.persona_dirty = true;
2157    let label = |p: &str| {
2158        if p.is_empty() {
2159            crate::persona::DEFAULT_ID.to_owned()
2160        } else {
2161            p.to_owned()
2162        }
2163    };
2164    fresh.turns.push(Turn {
2165        sender: Some(Sender::Magi),
2166        breaks: None,
2167        messages: Vec::new(),
2168        posted_by: None,
2169        who: Who::Agent,
2170        body: format!(
2171            "{MAGI_NOTE}persona changed from {} to {}",
2172            label(&from),
2173            label(id)
2174        ),
2175        at: Timestamp::now(),
2176        attachments: Vec::new(),
2177        usage: None,
2178    });
2179    store.put(&mut fresh)?;
2180    *talk = fresh;
2181    Ok(true)
2182}
2183
2184/// Discard the durable draft without adding a transcript turn.
2185pub fn clear_pending(talk: &mut Talk, store: &Talks) -> Result<()> {
2186    let _guard = store.guard()?;
2187    let mut fresh = store
2188        .get(&talk.id)
2189        .with_context(|| format!("talk {} was deleted", talk.short()))?;
2190    fresh.pending.clear();
2191    fresh.pending_breaks = None;
2192    fresh.pending_messages.clear();
2193    fresh.pending_attachments.clear();
2194    store.put(&mut fresh)?;
2195    *talk = fresh;
2196    Ok(())
2197}
2198
2199/// Clear a draft only when the caller still sees its complete snapshot.
2200pub fn clear_pending_if_matches(
2201    talk: &mut Talk,
2202    store: &Talks,
2203    expected_text: &str,
2204    expected_attachments: &[String],
2205) -> Result<bool> {
2206    let _guard = store.guard()?;
2207    let mut fresh = store
2208        .get(&talk.id)
2209        .with_context(|| format!("talk {} was deleted", talk.short()))?;
2210    if !pending_matches(&fresh, expected_text, expected_attachments) {
2211        *talk = fresh;
2212        return Ok(false);
2213    }
2214    fresh.pending.clear();
2215    fresh.pending_breaks = None;
2216    fresh.pending_messages.clear();
2217    fresh.pending_attachments.clear();
2218    store.put(&mut fresh)?;
2219    *talk = fresh;
2220    Ok(true)
2221}
2222
2223/// Replace just the text of the durable draft, but only if the caller's
2224/// snapshot still identifies the entire draft. This refuses to overwrite a
2225/// message another client queued or a draft the drain already promoted.
2226pub fn edit_pending_text(
2227    talk: &mut Talk,
2228    store: &Talks,
2229    text: &str,
2230    expected_text: &str,
2231    expected_attachments: &[String],
2232) -> Result<bool> {
2233    edit_pending_message(talk, store, text, expected_text, expected_attachments, None)
2234}
2235
2236/// Edit one queued send while retaining its reply and attachments.
2237pub fn edit_pending_message(
2238    talk: &mut Talk,
2239    store: &Talks,
2240    text: &str,
2241    expected_text: &str,
2242    expected_attachments: &[String],
2243    message: Option<usize>,
2244) -> Result<bool> {
2245    let _guard = store.guard()?;
2246    let mut fresh = store
2247        .get(&talk.id)
2248        .with_context(|| format!("talk {} was deleted", talk.short()))?;
2249    if !pending_matches(&fresh, expected_text, expected_attachments) {
2250        *talk = fresh;
2251        return Ok(false);
2252    }
2253    if fresh
2254        .pending_messages
2255        .iter()
2256        .enumerate()
2257        .any(|(i, m)| m.sender == Some(Sender::Magi) && message.is_none_or(|index| index == i))
2258    {
2259        bail!("MAGI messages cannot be edited");
2260    }
2261    if let Some(index) = message {
2262        let edited = fresh
2263            .pending_messages
2264            .get_mut(index)
2265            .context("invalid queued message")?;
2266        if text.trim().is_empty() && edited.attachments.is_empty() {
2267            fresh.pending_messages.remove(index);
2268        } else {
2269            edited.body = text.to_owned();
2270        }
2271        fresh.pending.clear();
2272        fresh.pending_breaks = Some(Vec::new());
2273        fresh.pending_attachments.clear();
2274        for message in &fresh.pending_messages {
2275            if !message.body.is_empty() {
2276                if !fresh.pending.is_empty() {
2277                    fresh.pending.push_str("\n\n");
2278                    fresh
2279                        .pending_breaks
2280                        .as_mut()
2281                        .expect("initialized")
2282                        .push(fresh.pending.len());
2283                }
2284                fresh.pending.push_str(&message.body);
2285            }
2286            fresh
2287                .pending_attachments
2288                .extend(message.attachments.clone());
2289        }
2290    } else {
2291        if fresh.pending_messages.len() > 1
2292            && fresh.pending_messages.iter().any(|m| m.reply.is_some())
2293        {
2294            bail!("edit queued replies individually");
2295        }
2296        let reply = fresh.pending_messages.first().and_then(|m| m.reply.clone());
2297        fresh.pending_messages = vec![Message {
2298            display: None,
2299            sender: Some(Sender::Operator),
2300            body: text.to_owned(),
2301            attachments: fresh.pending_attachments.clone(),
2302            reply,
2303        }];
2304        fresh.pending = text.to_owned();
2305        fresh.pending_breaks = Some(Vec::new());
2306    }
2307    store.put(&mut fresh)?;
2308    *talk = fresh;
2309    Ok(true)
2310}
2311
2312fn pending_matches(talk: &Talk, expected_text: &str, expected_attachments: &[String]) -> bool {
2313    talk.pending == expected_text
2314        && talk
2315            .pending_attachments
2316            .iter()
2317            .map(|attachment| &attachment.id)
2318            .eq(expected_attachments.iter())
2319}
2320
2321/// Invoke the conversation's agent once and append what it said.
2322///
2323/// The first turn ever taken carries the full [`briefing`], because nothing
2324/// else has told the agent what this conversation is or what it may do.
2325/// Every turn after that resends nothing when the CLI can resume its own
2326/// session, and falls back to [`transcript`] only when it cannot.
2327async fn turn(talk: &mut Talk, store: &Talks, cfg: &Config, text: &str) -> Result<()> {
2328    // Freeze both boundaries before invoking the CLI. Failure notes must not
2329    // hide unread posts in old records, and posts arriving during any attempt
2330    // (including fallback) were absent from this snapshot's prompt.
2331    let text = talk
2332        .turns
2333        .last()
2334        .filter(|t| t.who == Who::Operator && t.body == text)
2335        .map(|turn| message_prompt(turn, talk, store))
2336        .unwrap_or_else(|| text.to_owned());
2337    let evidence = crate::chat_notify::response_evidence(talk).await;
2338    let text = format!("{text}{evidence}");
2339    let prompt_turns = talk.turns.len();
2340    talk.posts_read_through = Some(posts_read_through(talk));
2341    let spec = cfg
2342        .agents
2343        .iter()
2344        .find(|a| a.id == talk.agent)
2345        .with_context(|| {
2346            format!(
2347                "talk {} was opened with agent `{}`, which is no longer in \
2348                 the roster; restore it in magi.toml or start a new \
2349                 conversation",
2350                talk.short(),
2351                talk.agent
2352            )
2353        })?;
2354
2355    // The newest turn is always the operator message this call is answering
2356    // - `record` appended it before `turn` was ever called - so its own
2357    // attachments are what belong at the end of *this* prompt.
2358    let last_note = attachment_note(
2359        store,
2360        &talk.id,
2361        talk.turns
2362            .last()
2363            .map_or(&[][..], |t| t.attachments.as_slice()),
2364    );
2365
2366    // Every attachment this conversation has ever held, not only this
2367    // turn's: a resumed session gets a fresh process every turn, so a CLI
2368    // whose sandbox needs `--add-dir` (see `agent::build_command`) needs the
2369    // grant again to open an image from an earlier turn, even when nothing
2370    // new was attached just now.
2371    let attachment_paths: Vec<PathBuf> = talk
2372        .turns
2373        .iter()
2374        .flat_map(|t| t.attachments.iter())
2375        .filter_map(|a| store.attachment_path(&talk.id, a))
2376        .collect();
2377
2378    // A question handed to this conversation is answered with `magi answer`,
2379    // which writes the question store; a read-only sandbox refuses that. So
2380    // while such a question is still open - the hand-over turn and the turns
2381    // in which the owner decides - the turn may write, with the question store
2382    // as a writable root, plus execution roots for an authorized dismissal.
2383    // It goes back to read-only once the question is
2384    // answered or abandoned. As for the deputy, what the seat may touch beyond
2385    // that rests on the prompt, not the sandbox.
2386    // `try_home`, not `Questions::open`: a unit test that never pinned a home
2387    // has no store to consult, and must not abort the turn on `run::home()`.
2388    let questions =
2389        crate::run::try_home().map(|home| crate::ask::Questions::at(home.join("questions")));
2390    let consult_roots = questions.as_ref().map_or_else(Vec::new, |questions| {
2391        crate::consult::answer_write_roots(questions, store, talk)
2392    });
2393    let consulted = !consult_roots.is_empty();
2394
2395    let (allow_write, unsandboxed) = turn_access(cfg.talk.allow_write, consulted);
2396
2397    let artifacts = store.artifacts_of(&talk.id);
2398    // From the transcript, not the seat: a switched agent's seat restarts at
2399    // zero and must not overwrite an earlier turn's artifacts.
2400    let operator_turns = talk.turns.iter().filter(|t| t.who == Who::Operator).count();
2401    let stem = format!("turn-{}", operator_turns.max(1));
2402    // The chat's build cache is the same shared one the graph's seats get, so
2403    // a conversation that compiles does not mint another multi-GB target dir.
2404    let cache_dir = cfg.cache_dir();
2405
2406    // The agent holding the conversation, then - only when it came from
2407    // `[roles] chatter` - the rest of that chain, each at most once.
2408    let mut chain = vec![spec.clone()];
2409    if let Some(choice) = cfg.roles.chatter.as_ref()
2410        && talk.fallback
2411    {
2412        for id in choice.ids() {
2413            if id == talk.agent || chain.iter().any(|s| s.id == id) {
2414                continue;
2415            }
2416            match agent::pick(&cfg.agents, Some(id), &agent::installed) {
2417                Ok(s) => chain.push(s),
2418                Err(e) => tracing::warn!("[roles] chatter: skipping `{id}`: {e:#}"),
2419            }
2420        }
2421    }
2422
2423    let persona = crate::persona::active(&cfg.talk.personas, &talk.persona);
2424    let operator_name = cfg.talk.operator_name();
2425    let persona_update = if talk.persona_dirty {
2426        format!(
2427            "{}\n\n",
2428            crate::persona::update_block_for(persona.as_ref(), operator_name)
2429        )
2430    } else {
2431        String::new()
2432    };
2433
2434    // The filing command is not in the transcript either, so a non-Solo count
2435    // rides on every turn that has no session to remember it.
2436    let filing_update = if talk.implementers_dirty {
2437        format!("{}\n\n", filing_update_block(talk.implementers))
2438    } else {
2439        String::new()
2440    };
2441
2442    let mut outcome = None;
2443    let mut fell_back_from: Option<String> = None;
2444    // What the conversation looked like after the first agent's failed try,
2445    // so an exhausted chain leaves exactly what a single failed seat would.
2446    let mut first_try: Option<(String, SeatState)> = None;
2447    for (n, spec) in chain.iter().enumerate() {
2448        if n > 0 {
2449            if first_try.is_none() {
2450                first_try = Some((talk.agent.clone(), talk.seat.clone()));
2451            }
2452            tracing::warn!("chat: falling back from `{}` to `{}`", talk.agent, spec.id);
2453            // A new CLI has none of the old one's conversation: a fresh seat
2454            // puts `has_session` at false and the full transcript is re-sent.
2455            fell_back_from.get_or_insert_with(|| talk.agent.clone());
2456            talk.agent = spec.id.clone();
2457            talk.seat = SeatState::new(SEAT, &spec.id, crate::rng::entropy());
2458        }
2459        let resuming = agent::has_session(spec.kind, &talk.seat, cfg.graph.sessions);
2460        let first_ever = talk.turns.len() <= 1;
2461        let body = if talk.seat.turns == 0 && first_ever {
2462            format!(
2463                "{}\n\n# Operator\n\n{text}{last_note}",
2464                briefing_for(
2465                    &talk.repo,
2466                    &cfg.graph.language,
2467                    cfg.talk.allow_write,
2468                    persona.as_ref(),
2469                    operator_name,
2470                    talk.implementers
2471                )
2472            )
2473        } else if talk.seat.turns == 0 {
2474            // A fresh seat on a conversation that already has history (the
2475            // agent was switched): the briefing, then everything said so far.
2476            format!(
2477                "{}\n\n{}\n\n# Operator\n\n{text}{last_note}",
2478                briefing_for(
2479                    &talk.repo,
2480                    &cfg.graph.language,
2481                    cfg.talk.allow_write,
2482                    persona.as_ref(),
2483                    operator_name,
2484                    talk.implementers
2485                ),
2486                transcript(talk, store)
2487            )
2488        } else if resuming {
2489            let posts = posts_since_last_reply(talk).unwrap_or_default();
2490            format!("{persona_update}{filing_update}{posts}{text}{last_note}")
2491        } else {
2492            // No session to hold the persona: the transcript never stores it,
2493            // so a non-default persona is re-sent on every such turn (the
2494            // update block already carries it when the choice just changed).
2495            let mut standing = match (&persona, talk.persona_dirty) {
2496                (Some(p), false) => format!("{}\n", crate::persona::section_for(p, operator_name)),
2497                _ => persona_update.clone(),
2498            };
2499            // The default voice has no persona section to carry the name, and
2500            // the transcript never stores the briefing's addressing line.
2501            if let (None, Some(n)) = (&persona, operator_name) {
2502                standing.push_str(&format!(
2503                    "# Addressing the operator\n{}\n",
2504                    crate::persona::addressing(n)
2505                ));
2506            }
2507            if talk.implementers_dirty || talk.implementers != 1 {
2508                standing.push_str(&format!("{}\n\n", filing_update_block(talk.implementers)));
2509            }
2510            format!("{}\n\n{standing}{text}{last_note}", transcript(talk, store))
2511        };
2512        // A resumed conversation may remember a legacy merge/hold-only card.
2513        // Re-send current persisted choices, without altering the owner's words.
2514        let current_approvals = questions
2515            .as_ref()
2516            .map(|store| {
2517                store
2518                    .list()
2519                    .iter()
2520                    .filter(|q| {
2521                        q.status.open()
2522                            && q.node == crate::land::APPROVAL_NODE
2523                            && q.consult.as_ref().is_some_and(|c| c.talk == talk.id)
2524                    })
2525                    .map(crate::prompt::chat_consult)
2526                    .collect::<Vec<_>>()
2527                    .join("\n\n")
2528            })
2529            .unwrap_or_default();
2530        let body = if current_approvals.is_empty() {
2531            body
2532        } else {
2533            format!("{current_approvals}\n\n# Current operator message\n\n{body}")
2534        };
2535        let attempt_stem = if n == 0 {
2536            stem.clone()
2537        } else {
2538            format!("{stem}-{}", spec.id)
2539        };
2540        let inv = Invocation {
2541            cwd: &talk.repo,
2542            prompt: &body,
2543            timeout: turn_timeout(cfg),
2544            // Off unless this repository's own config opts in - see
2545            // `crate::config::Talk::allow_write` and this module's doc for why
2546            // the default keeps a conversational edit from landing in a checkout
2547            // no run or review can claim.
2548            allow_write,
2549            unsandboxed,
2550            sessions: cfg.graph.sessions,
2551            artifacts: &artifacts,
2552            stem: &attempt_stem,
2553            // The conversation's own id, so `magi task add` run from inside it is
2554            // attributed to this conversation - see `Source::Agent`.
2555            run: &talk.id,
2556            node: crate::queue::CHAT_NODE,
2557            cache_dir: cache_dir.as_deref(),
2558            attachments: &attachment_paths,
2559            writable: &consult_roots,
2560        };
2561        let wait = crate::repo_lock::WaitPolicy::new(
2562            Duration::from_secs(cfg.graph.timeout_resource_wait),
2563            || false,
2564        );
2565        let result = wait.scope(agent::invoke(spec, &mut talk.seat, &inv)).await;
2566        let advance = agent::chain_advances(&result);
2567        if n == 0 || !advance {
2568            outcome = Some(result);
2569        } else {
2570            // A later failure is only logged; the note describes the first.
2571            tracing::warn!("chat: fallback agent `{}` also failed", spec.id);
2572        }
2573        if !advance {
2574            break;
2575        }
2576    }
2577    if outcome.as_ref().is_some_and(agent::chain_advances) {
2578        // Exhausted: back to the agent the conversation had, so the note
2579        // below names it and the next turn starts from it again.
2580        if let Some((id, seat)) = first_try {
2581            talk.agent = id;
2582            talk.seat = seat;
2583            fell_back_from = None;
2584        }
2585    }
2586    let outcome = outcome.expect("a chain holds at least one agent");
2587    let note = |why: String| Turn {
2588        sender: Some(Sender::Magi),
2589        breaks: None,
2590        messages: Vec::new(),
2591        posted_by: None,
2592        who: Who::Agent,
2593        body: format!("{MAGI_NOTE}{why}"),
2594        at: Timestamp::now(),
2595        attachments: Vec::new(),
2596        usage: None,
2597    };
2598    let (reply, failure) = match outcome {
2599        Err(e) => (
2600            note(format!("could not run agent `{}`: {e}", talk.agent)),
2601            Some(format!("could not run agent `{}`: {e}", talk.agent)),
2602        ),
2603        Ok(out) if out.quota_exhausted() => {
2604            let reset = out
2605                .quota
2606                .as_ref()
2607                .and_then(|q| q.reset.clone())
2608                .map_or_else(String::new, |r| format!(" (resets {r})"));
2609            let why = format!(
2610                "agent `{}` is out of quota{reset}; your message is saved, so \
2611                 say it again when the window reopens",
2612                talk.agent
2613            );
2614            (note(why.clone()), Some(why))
2615        }
2616        Ok(out) if out.timed_out => {
2617            let why = format!(
2618                "agent `{}` did not answer within {}s; your message is saved",
2619                talk.agent,
2620                turn_timeout(cfg).as_secs()
2621            );
2622            (note(why.clone()), Some(why))
2623        }
2624        Ok(out) if !out.usable() => {
2625            let why = format!(
2626                "agent `{}` produced no answer (exit {}); your message is saved",
2627                talk.agent,
2628                out.exit_code
2629                    .map_or_else(|| "unknown".to_owned(), |c| c.to_string())
2630            );
2631            (note(why.clone()), Some(why))
2632        }
2633        Ok(out) => (
2634            Turn {
2635                sender: Some(Sender::Persona),
2636                breaks: None,
2637                messages: Vec::new(),
2638                posted_by: None,
2639                who: Who::Agent,
2640                body: out.text.trim().to_owned(),
2641                at: Timestamp::now(),
2642                attachments: Vec::new(),
2643                // `talk.agent` is whoever actually answered: a fallback has
2644                // already moved it, and an exhausted chain never reaches here.
2645                usage: out.context_tokens.map(|context_tokens| TurnUsage {
2646                    context_tokens,
2647                    agent: talk.agent.clone(),
2648                    model: cfg
2649                        .agents
2650                        .iter()
2651                        .find(|a| a.id == talk.agent)
2652                        .and_then(|a| a.model.clone()),
2653                }),
2654            },
2655            None,
2656        ),
2657    };
2658
2659    // A close landed on disk while this turn was in flight is read back here
2660    // rather than trusted from the snapshot this call started with. `store`
2661    // holds nothing else this function does not itself own - the turn guard
2662    // in `web::Ui::begin_talk_turn` keeps `turns` and `seat` this call's
2663    // alone to mutate - but `status` is not behind that guard, and an
2664    // operator's close must stick: the whole point of ending a conversation
2665    // is that an agent's answer to the last message before the close cannot
2666    // silently reopen it. The guard is what makes that read-then-write
2667    // section atomic with `close`'s own - taken only for this tail and not
2668    // for the whole invocation above, so one talk's fifteen-minute turn does
2669    // not block another talk's close from proceeding.
2670    let _guard = store.guard()?;
2671    // A delete is the more final version of that same race: `put` writes
2672    // unconditionally, so a talk removed while this turn was in flight must
2673    // stay removed rather than being written back with this turn's reply
2674    // appended to it. The reply is simply given up on - there is no
2675    // conversation left for it to belong to.
2676    let Ok(fresh) = store.get(&talk.id) else {
2677        return Ok(());
2678    };
2679    talk.status = fresh.status;
2680    // `queue` may have accepted another operator message while the CLI was
2681    // running. This handle predates that write, so preserving only `status`
2682    // would overwrite the durable draft when the reply is appended below.
2683    talk.pending = fresh.pending;
2684    talk.pending_breaks = fresh.pending_breaks;
2685    talk.pending_messages = fresh.pending_messages;
2686    talk.pending_attachments = fresh.pending_attachments;
2687    talk.notified = fresh.notified;
2688    // A `magi talk post` landed while the CLI ran: this handle predates it.
2689    if fresh.turns.len() > talk.turns.len() {
2690        let posted: Vec<Turn> = fresh.turns[talk.turns.len()..]
2691            .iter()
2692            .filter(|t| t.posted_by.is_some())
2693            .cloned()
2694            .collect();
2695        talk.turns.extend(posted);
2696    }
2697    if let Some(from) = fell_back_from.filter(|_| failure.is_none()) {
2698        // The switch persists: quota coming back does not move the chat
2699        // home, an operator's switch does.
2700        talk.turns.push(note(format!(
2701            "agent changed from {from} to {} (fallback)",
2702            talk.agent
2703        )));
2704    }
2705    // The persona is in the session's context only once a turn has answered
2706    // (a failed one may never have reached the agent), so only then is it clean.
2707    if failure.is_none() {
2708        talk.persona_dirty = false;
2709        talk.implementers_dirty = false;
2710        talk.posts_read_through = Some(prompt_turns);
2711    }
2712    talk.turns.push(reply);
2713    if let Err(put_err) = store.put(talk) {
2714        // `Talks::put` already retried the write itself - reaching here
2715        // means a passing race is not what this is. An agent's answer,
2716        // possibly the result of an hour-long call, must not vanish with
2717        // nothing to show for it just because the very last step failed:
2718        // pop it back off, stash its text beside the conversation, and
2719        // replace it with a note the operator can actually see, the same
2720        // mechanism the failure branches above already use for a quota or a
2721        // timeout.
2722        let lost = talk.turns.pop().expect("just pushed above");
2723        let stash = stash_lost_turn(store, &talk.id, &stem, &lost);
2724        let why = match &stash {
2725            Ok(path) => format!(
2726                "agent `{}` answered, but the reply could not be saved to \
2727                 this conversation ({put_err:#}); the raw text was kept at \
2728                 {} - your message is saved, ask again",
2729                talk.agent,
2730                path.display()
2731            ),
2732            Err(stash_err) => format!(
2733                "agent `{}` answered, but the reply could not be saved to \
2734                 this conversation ({put_err:#}), and it could not be kept \
2735                 anywhere else either ({stash_err:#}); your message is \
2736                 saved, ask again",
2737                talk.agent
2738            ),
2739        };
2740        talk.turns.push(note(why.clone()));
2741        // Writing the note also carries the seat this call already advanced -
2742        // `agent::invoke` incremented `turns` and, for a vendor that reports
2743        // its own session id, recorded that too. That is what keeps the next
2744        // turn resuming the session the CLI is already holding instead of
2745        // re-opening it, so losing the reply costs the transcript a turn but
2746        // not the conversation.
2747        return match store.put(talk) {
2748            Ok(()) => bail!("{why}"),
2749            Err(note_err) => {
2750                // Even the short note failed to save, which means this
2751                // conversation's file cannot be written at all right now -
2752                // nothing is left for this call to retry or record. Pop the
2753                // note so `talk.turns` matches the transcript on disk, and
2754                // surface both failures for whoever reads the log.
2755                //
2756                // `talk.seat` is deliberately not wound back to match. The
2757                // CLI really did take the turn and really did consume this
2758                // seat's session id; pretending otherwise would be a second
2759                // untruth on top of the unwritable file, and the handle is
2760                // reloaded from disk by the next `drain` or `get` anyway -
2761                // see `web::drain_loop`. What the seat cannot do is reach
2762                // disk, so the record stays a turn behind the CLI until some
2763                // later write lands, and a turn taken before then re-opens a
2764                // session id the CLI already holds. That is the desync
2765                // `20260907-011805-fb57` is about, and tolerating it belongs
2766                // there rather than here: no write this branch could make
2767                // would help, since a failed write is exactly what put it in
2768                // this position twice over.
2769                talk.turns.pop();
2770                Err(note_err).context(why)
2771            }
2772        };
2773    }
2774
2775    match failure {
2776        Some(why) => bail!("{why}"),
2777        None => Ok(()),
2778    }
2779}
2780
2781/// Everything said so far, as prose, for a CLI that cannot resume its own
2782/// conversation.
2783fn transcript(talk: &Talk, store: &Talks) -> String {
2784    let mut out = String::from(
2785        "This conversation cannot resume on the CLI's side, so here is \
2786         everything said so far; answer only the last message.\n",
2787    );
2788    for t in &talk.turns {
2789        let who = match t.who {
2790            Who::Operator => "operator",
2791            Who::Agent if t.body.starts_with(MAGI_NOTE) => "magi",
2792            Who::Agent => "you",
2793        };
2794        out.push_str(&format!(
2795            "\n## {who}\n\n{}\n",
2796            message_prompt(t, talk, store).trim()
2797        ));
2798        out.push_str(&attachment_note(store, &talk.id, &t.attachments));
2799    }
2800    out
2801}
2802
2803/// The section named at the end of a turn's body, listing every attachment's
2804/// absolute path and mime so the agent knows exactly what to open. Empty
2805/// when `attachments` is, which is every turn but the rare one carrying an
2806/// image, so a turn with none changes nothing about the prompt.
2807fn attachment_note(store: &Talks, talk_id: &str, attachments: &[Attachment]) -> String {
2808    if attachments.is_empty() {
2809        return String::new();
2810    }
2811    let mut out = String::from(
2812        "\n\nThe operator attached the image(s) below to this message. Open \
2813         and look at each one before you answer.\n",
2814    );
2815    for att in attachments {
2816        if let Some(path) = store.attachment_path(talk_id, att) {
2817            out.push_str(&format!("\n- {} ({})", path.display(), att.mime));
2818        }
2819    }
2820    out.push('\n');
2821    out
2822}
2823
2824/// The briefing the agent opens with, sent once as part of its first turn.
2825///
2826/// Pure, so the properties that matter can be asserted without an interview:
2827/// it names `magi task add --solo` (the route this conversation always has to
2828/// changing anything) and it never tells the agent to write a task *file* of
2829/// its own - that would compete with filing through the queue.
2830/// What a turn may do: `(allow_write, unsandboxed)`. Only the repository's own
2831/// `[talk] allow_write` lifts the CLI sandbox (codex's counterpart of claude's
2832/// `bypassPermissions`); a turn writable merely because a handed-over question
2833/// is open keeps the sandbox and gets just the question store as a writable root.
2834pub(crate) fn turn_access(talk_allow_write: bool, consulted: bool) -> (bool, bool) {
2835    (talk_allow_write || consulted, talk_allow_write)
2836}
2837
2838/// `allow_write` only ever adds an extra permission on top of that; it never
2839/// removes the queue as an option, which is why both branches keep the same
2840/// `# When the operator wants something done` section - `write_policy` is
2841/// the only part that changes.
2842///
2843/// It also tells the agent that `--repo` is not stuck naming this
2844/// conversation's own directory: `resolve_repo` (`src/main.rs`) now accepts a
2845/// short `owner/repo` or bare `repo` name and resolves it against
2846/// `[repos] roots`, the same local checkouts `magi repos` lists. Without this
2847/// line an agent asked to change some other repository has no way to know
2848/// that option exists, and the only path it can see - asking the operator to
2849/// dictate a full path - is exactly the friction this change exists to
2850/// remove. A miss or an ambiguous name still fails the command outright, so
2851/// the instruction is to ask rather than guess when that happens - the
2852/// silent-decision line this task must not cross.
2853pub fn briefing(repo: &Path, language: &str, allow_write: bool) -> String {
2854    briefing_with(repo, language, allow_write, None, None)
2855}
2856
2857/// [`briefing`] plus the selected persona's tone-only section and, when the
2858/// operator has a configured name, how to address them. `None` for both is
2859/// byte-for-byte the plain briefing. A name changed mid-session reaches a
2860/// running CLI session only on its next fresh seat (nothing is persisted).
2861pub fn briefing_with(
2862    repo: &Path,
2863    language: &str,
2864    allow_write: bool,
2865    persona: Option<&crate::persona::Persona>,
2866    operator_name: Option<&str>,
2867) -> String {
2868    briefing_for(repo, language, allow_write, persona, operator_name, 1)
2869}
2870
2871/// The most implementers a chat can ask its filed tasks to use.
2872pub const MAX_IMPLEMENTERS: u8 = 3;
2873
2874fn solo_implementers() -> u8 {
2875    1
2876}
2877
2878/// The `magi task add` flag for `n` implementers: `--solo` for one,
2879/// `--implementers N` otherwise (the two are never combined).
2880fn filing_flag(n: u8) -> String {
2881    if n <= 1 {
2882        "--solo".to_owned()
2883    } else {
2884        format!("--implementers {n}")
2885    }
2886}
2887
2888/// The note a resumed (or transcript-replayed) turn opens with after the
2889/// implementer count changed, or while it is not Solo.
2890pub fn filing_update_block(n: u8) -> String {
2891    let flag = filing_flag(n);
2892    if n <= 1 {
2893        format!(
2894            "# Task filing update\n\nThe operator changed how many implementers tasks use. \
2895             From now on file with `magi task add {flag} --repo <repo> <instruction>` \
2896             (one implementer, straight into review).\n"
2897        )
2898    } else {
2899        format!(
2900            "# Task filing update\n\nThe operator changed how many implementers tasks use. \
2901             From now on file with `magi task add {flag} --repo <repo> <instruction>` \
2902             ({n} independent implementations compete). Do not pass `--solo` \
2903             together with `--implementers`; drop any earlier `--solo`.\n"
2904        )
2905    }
2906}
2907
2908/// Validate a requested implementer count: 1..=3, and one the repository's
2909/// config can actually resolve. Never clamps.
2910pub fn check_implementers(n: u8, cfg: &crate::config::Config) -> std::result::Result<u8, String> {
2911    if !(1..=MAX_IMPLEMENTERS).contains(&n) {
2912        return Err(format!(
2913            "implementers must be 1..={MAX_IMPLEMENTERS}, got {n}"
2914        ));
2915    }
2916    if n > 1 {
2917        let mut cfg = cfg.clone();
2918        crate::queue::RunOverrides {
2919            candidates: Some(usize::from(n)),
2920            ..Default::default()
2921        }
2922        .apply(&mut cfg);
2923        cfg.resolve_roles()
2924            .map_err(|e| format!("{n} implementers is not allowed here: {e:#}"))?;
2925    }
2926    Ok(n)
2927}
2928
2929/// Set how many implementers this chat files tasks with. Same shape as
2930/// [`switch_persona`]: `false` and no write when unchanged.
2931pub fn switch_implementers(talk: &mut Talk, store: &Talks, n: u8) -> Result<bool> {
2932    let _guard = store.guard()?;
2933    let mut fresh = store
2934        .get(&talk.id)
2935        .with_context(|| format!("talk {} was deleted", talk.short()))?;
2936    if fresh.implementers == n {
2937        *talk = fresh;
2938        return Ok(false);
2939    }
2940    let from = std::mem::replace(&mut fresh.implementers, n);
2941    fresh.implementers_dirty = true;
2942    fresh.turns.push(Turn {
2943        sender: Some(Sender::Magi),
2944        breaks: None,
2945        messages: Vec::new(),
2946        posted_by: None,
2947        who: Who::Agent,
2948        body: format!("{MAGI_NOTE}implementers changed from {from} to {n}"),
2949        at: Timestamp::now(),
2950        attachments: Vec::new(),
2951        usage: None,
2952    });
2953    store.put(&mut fresh)?;
2954    *talk = fresh;
2955    Ok(true)
2956}
2957
2958/// [`briefing_with`] for a chat that files tasks with `implementers`
2959/// implementers (1 is the unchanged `--solo` text).
2960pub fn briefing_for(
2961    repo: &Path,
2962    language: &str,
2963    allow_write: bool,
2964    persona: Option<&crate::persona::Persona>,
2965    operator_name: Option<&str>,
2966    implementers: u8,
2967) -> String {
2968    let flag = filing_flag(implementers);
2969    let shape = if implementers <= 1 {
2970        "Use --solo: it runs the task through one implementer \
2971         straight into review instead of the usual multi-agent competition, \
2972         which is the right shape for a change this conversation has already \
2973         settled, rather than one still worth several independent takes."
2974            .to_owned()
2975    } else {
2976        format!(
2977            "Use --implementers {implementers}: it has {implementers} \
2978             implementers work on the task independently and compete, which \
2979             is the right shape when several independent takes are worth \
2980             having. Never add --solo to it; the two do not go together."
2981        )
2982    };
2983    let write_policy = if allow_write {
2984        "Write access is enabled for this conversation (`allow_write = \
2985         true`), so you may write files - but only a small, \
2986         already-decided edit the operator names outright in this \
2987         conversation, not an implementation. This is a permission on the \
2988         conversation as a whole, not a property of whichever repository \
2989         it happened to start in: if the operator names a different \
2990         repository for that small edit, the policy allows it there too. \
2991         Your own tool may still confine writes to the repository this \
2992         conversation started in regardless - if a write elsewhere is \
2993         refused, say so plainly rather than working around it. Once you \
2994         have made an edit, say plainly what you edited. Anything bigger, \
2995         or anything still open-ended, still goes through the queue below \
2996         rather than being done here."
2997    } else {
2998        "Do not write files. Implementing a change is not this \
2999         conversation's job; a separate, blind competition of agents does \
3000         that, and a repository this conversation has already edited would \
3001         make their diffs unjudgeable."
3002    };
3003    let mut out = format!(
3004        "You are magi's standing conversation partner for its operator, who \
3005         usually has this open on a phone. Keep replies short: no preamble, \
3006         no restating what they just said.\n\n\
3007         # Repository\n\n{repo}\n\n\
3008         You may look around: read files, run shell commands, search history, \
3009         run tests - whatever answers the question. {write_policy}\n\n\
3010         A short, command-shaped message (\"list\", \"info <id>\", \"show \
3011         3cbf\") is almost always the operator asking you to look something \
3012         up, not an instruction to file - answer it yourself with `magi \
3013         list`, `magi show <id>`, `magi task list`, or the like, the same way \
3014         you would answer any other question in this conversation.\n\n\
3015         # Answering existing questions\n\n\
3016         Consultation delivery does not restrict an explicit owner decision. \
3017         Use `magi answer <full-question-id> --reply <exact offered choice> --quote <entire latest owner message>`. \
3018         Direct instructions bind `<full-question-id>: <exact offered choice>`. \
3019         An explicit batch instruction (including `おっけーそれぞれ回答してー`) authorizes \
3020         the immediately preceding persisted recommendations, one row per question: \
3021         `<full-question-id>: <exact offered choice>` or a table with separate ID and choice cells. \
3022         Hold, obsolete-approval dismissal, repair/re-review and non-destructive recovery choices \
3023         may be delegated this way, including cross-chat and never-consulted questions. \
3024         Merge requires `<full-question-id>: merge <displayed head>`. Withdrawal, PR closure, \
3025         replacement that closes the old PR, discard and task completion require an individual \
3026         exact consequence instruction; a batch instruction is insufficient. Public posting requires \
3027         the exact question/content fingerprint reported by the answer command. `use my text` \
3028         requires the question/content fingerprint followed by a newline and the literal replacement title. Report the exact missing evidence. \
3029         Never create another driver or task to apply an answer: existing execution transitions \
3030         retain their state, ownership and confirmation checks. A busy original consultation must \
3031         finish before a non-hold answer. A hold grants no closure or withdrawal. \
3032         Thanks, silence, recommendations alone, conditional/retracted instructions and generated \
3033         messages are not owner authorization.\n\n\
3034         # When the operator wants something done\n\n\
3035         Run:\n\n\
3036         magi task add {flag} --repo {repo} <instruction>\n\n\
3037         and tell the operator the task id it prints, so they can follow it \
3038         from the Queue. If it refuses with a duplicate warning (the \
3039         instruction names a branch, commit or pull request that an \
3040         unfinished task, run or PR already owns), do not repeat it with \
3041         --force yourself: tell the operator what it matched and let them \
3042         decide. Write <instruction> so that an implementer who has \
3043         never seen this conversation can act on it alone - it is everything \
3044         they get. {shape}\n\n\
3045         If the operator asks for something in a different repository, \
3046         --repo does not have to be a full path: --repo owner/repo (or just \
3047         repo, when that is unambiguous) is resolved against local checkouts \
3048         the same way `magi repos` lists them. If the command fails because \
3049         nothing matches or more than one checkout shares that name, ask the \
3050         operator which repository they mean (or run `magi repos` yourself \
3051         to see the candidates) rather than guessing.\n\n\
3052         You cannot wait, and you cannot send the operator a message on your \
3053         own: you only speak when they write. A task you file reports back \
3054         into this conversation automatically when it finishes, is held or \
3055         fails, as a message headed \"{notice}\".\n\n\
3056         For a background command you run yourself, register it with \
3057         `magi talk watch --timeout 900 -- <executable> <arguments...>`, for \
3058         example `magi talk watch -- cargo test --locked`. It uses your \
3059         current directory; arguments are passed literally. For shell syntax, \
3060         explicitly call `sh -c`. Only after watch registration succeeds may \
3061         you promise a completion report. Watches default to 900 seconds, \
3062         accept at most 3600 seconds, and share four concurrent slots. Their \
3063         automatic reports resume this conversation through its own turn. \
3064         Never promise to report back on work that is neither a filed task \
3065         nor a successfully registered watch.\n\n\
3066         The current state of the code is whatever origin/main holds, not \
3067         whatever a working tree shows: a primary checkout often lags \
3068         upstream, sits on a detached HEAD and carries uncommitted changes. \
3069         Before answering about code, run `git fetch origin` in that \
3070         repository if it is cheap, then read through \
3071         `git show origin/main:<path>` or `git grep <pattern> origin/main`. \
3072         If the working tree differs, say so; if the fetch fails, say that \
3073         too, so the operator knows the answer may be stale.\n\n\
3074         If the operator attached an image (a screenshot, say) that the task \
3075         is about, pass it with `--attach <path>`, using the absolute path \
3076         the turn's attachment note gives; repeat the flag for several. \
3077         `magi task add {flag} --attach <path> <instruction>` copies the \
3078         file into the task, so the implementer receives it. Do not paste the \
3079         path into <instruction> instead: deleting this conversation deletes \
3080         its attachments, and then that path reaches no one.\n",
3081        repo = repo.display(),
3082        notice = crate::prompt::CHAT_TASK_NOTICE_HEADING,
3083    );
3084    out.push_str(&language_note(language));
3085    match (persona, operator_name) {
3086        (Some(p), n) => out.push_str(&crate::persona::section_for(p, n)),
3087        (None, Some(n)) => {
3088            out.push_str("\n# Addressing the operator\n");
3089            out.push_str(&crate::persona::addressing(n));
3090        }
3091        (None, None) => {}
3092    }
3093    out
3094}
3095
3096/// The operator is talking, so their language matters here more than in most
3097/// prompts magi sends.
3098fn language_note(language: &str) -> String {
3099    if language.trim().is_empty() || language.eq_ignore_ascii_case("en") {
3100        String::new()
3101    } else {
3102        format!("\nHold this conversation in {language}.\n")
3103    }
3104}
3105
3106/// Queue tasks this conversation has filed, oldest first.
3107///
3108/// A task is this conversation's when its [`Source::Agent`] names this
3109/// conversation's id as `run` - which is exactly what happens when
3110/// `magi task add` is run from inside a turn, because [`turn`] passes the
3111/// conversation's own id as [`Invocation::run`].
3112pub fn tasks_of(queue: &Queue, talk_id: &str) -> Vec<Task> {
3113    let mut tasks: Vec<Task> = queue
3114        .list()
3115        .into_iter()
3116        .filter(|t| matches!(&t.source, Source::Agent { run, .. } if run == talk_id))
3117        .collect();
3118    tasks.sort_unstable_by(|a, b| a.id.cmp(&b.id));
3119    tasks
3120}
3121
3122fn read_path(path: &Path) -> Result<Talk> {
3123    let body = std::fs::read_to_string(path).with_context(|| format!("read {}", path.display()))?;
3124    let mut talk: Talk =
3125        serde_json::from_str(&body).with_context(|| format!("parse {}", path.display()))?;
3126    for turn in &mut talk.turns {
3127        if turn.who == Who::Operator && turn.messages.is_empty() && turn.attachments.is_empty() {
3128            turn.messages = plain_messages(&turn.body, turn.breaks.as_deref());
3129        }
3130    }
3131    if talk.pending_messages.is_empty() && talk.pending_attachments.is_empty() {
3132        talk.pending_messages = plain_messages(&talk.pending, talk.pending_breaks.as_deref());
3133    }
3134    Ok(talk)
3135}
3136
3137/// Rebuild only explicitly recorded boundaries; legacy paragraphs stay intact.
3138fn plain_messages(body: &str, breaks: Option<&[usize]>) -> Vec<Message> {
3139    let Some(breaks) = breaks else {
3140        return Vec::new();
3141    };
3142    if body.is_empty() {
3143        return Vec::new();
3144    }
3145    let mut messages = Vec::new();
3146    let mut start = 0;
3147    for end in breaks
3148        .iter()
3149        .copied()
3150        .chain(std::iter::once(body.len() + 2))
3151    {
3152        let Some(text) = end.checked_sub(2).and_then(|end| body.get(start..end)) else {
3153            return Vec::new();
3154        };
3155        messages.push(Message {
3156            display: None,
3157            sender: None,
3158            body: text.to_owned(),
3159            attachments: Vec::new(),
3160            reply: None,
3161        });
3162        start = end;
3163    }
3164    messages
3165}
3166
3167/// How many times [`write_atomic`] retries a failed write-then-rename before
3168/// giving up.
3169const PUT_RETRIES: u32 = 5;
3170
3171/// Write `body` to `tmp` and rename it onto `path`, retrying the whole thing
3172/// a handful of times with a short sleep in between.
3173///
3174/// The only failure this is meant to absorb is a passing one - most
3175/// concretely, a reader elsewhere in this process (or another `magi`
3176/// process) with `path` briefly open for `read_to_string` at the exact
3177/// moment this call tries to rename over it. That clears in milliseconds
3178/// once the reader lets go; a caller still failing after several short
3179/// sleeps has something more durable wrong (a full disk, a permissions
3180/// change) that a longer sleep would not fix either, and is left to report
3181/// it.
3182fn write_atomic(tmp: &Path, path: &Path, body: &str) -> Result<()> {
3183    let mut last_err = None;
3184    for attempt in 0..PUT_RETRIES {
3185        if attempt > 0 {
3186            std::thread::sleep(Duration::from_millis(20 * u64::from(attempt)));
3187        }
3188        match try_write_atomic(tmp, path, body) {
3189            Ok(()) => return Ok(()),
3190            Err(e) => last_err = Some(e),
3191        }
3192    }
3193    Err(last_err.expect("the loop above always runs at least once"))
3194}
3195
3196fn try_write_atomic(tmp: &Path, path: &Path, body: &str) -> Result<()> {
3197    #[cfg(test)]
3198    if failpoint::take_forced_put_failure() {
3199        bail!("simulated write failure (test)");
3200    }
3201    std::fs::write(tmp, body).with_context(|| format!("write {}", tmp.display()))?;
3202    std::fs::rename(tmp, path).with_context(|| format!("replace {}", path.display()))?;
3203    Ok(())
3204}
3205
3206/// Last resort when `turn`'s own `store.put` fails even after
3207/// [`write_atomic`]'s retries: keep the generated text somewhere still
3208/// findable rather than let the whole of an agent's answer disappear along
3209/// with the write that was supposed to record it.
3210fn stash_lost_turn(store: &Talks, id: &str, stem: &str, reply: &Turn) -> Result<PathBuf> {
3211    let dir = store.artifacts_of(id);
3212    std::fs::create_dir_all(&dir).with_context(|| format!("create {}", dir.display()))?;
3213    let path = dir.join(format!("{stem}-lost.txt"));
3214    std::fs::write(&path, &reply.body).with_context(|| format!("write {}", path.display()))?;
3215    Ok(path)
3216}
3217
3218/// A test-only seam that lets [`try_write_atomic`] simulate the kind of
3219/// passing I/O race [`write_atomic`] is meant to retry through, without
3220/// depending on real OS-level file-locking behaviour, which differs across
3221/// the three platforms this crate ships on (and, on the one platform where a
3222/// reader really does block a rename, is awkward to trigger deterministically
3223/// in a unit test).
3224#[cfg(test)]
3225mod failpoint {
3226    use std::cell::Cell;
3227
3228    thread_local! {
3229        static FORCE_PUT_FAILURES: Cell<u32> = const { Cell::new(0) };
3230        static FORCE_NO_LINK: Cell<bool> = const { Cell::new(false) };
3231    }
3232
3233    /// While set, `hard_link` on this thread fails as unsupported, so the
3234    /// create-and-rename fallback runs.
3235    pub(super) fn force_no_link(on: bool) {
3236        FORCE_NO_LINK.with(|c| c.set(on));
3237    }
3238
3239    pub(super) fn no_link_forced() -> bool {
3240        FORCE_NO_LINK.with(Cell::get)
3241    }
3242
3243    /// Arrange for the next `count` calls into [`super::try_write_atomic`] to
3244    /// fail before touching the filesystem at all.
3245    pub(super) fn force_put_failures(count: u32) {
3246        FORCE_PUT_FAILURES.with(|c| c.set(count));
3247    }
3248
3249    /// Consumed once per attempt inside [`super::try_write_atomic`]; `true`
3250    /// means simulate this attempt failing.
3251    pub(super) fn take_forced_put_failure() -> bool {
3252        FORCE_PUT_FAILURES.with(|c| {
3253            let n = c.get();
3254            if n == 0 {
3255                false
3256            } else {
3257                c.set(n - 1);
3258                true
3259            }
3260        })
3261    }
3262}
3263
3264fn short(id: &str) -> &str {
3265    id.split('-').next_back().unwrap_or(id)
3266}
3267
3268fn new_id() -> String {
3269    let stamp = jiff::Zoned::now().strftime("%Y%m%d-%H%M%S");
3270    let seed = crate::rng::entropy();
3271    format!("{stamp}-{:04x}", (seed ^ (seed >> 32)) & 0xffff)
3272}
3273
3274/// Extension an attachment's bytes are stored under, from its (already
3275/// validated) mime. The one place this mapping exists on the write side;
3276/// `web`'s own whitelist is what actually decides which mimes are accepted
3277/// in the first place.
3278fn attachment_ext(mime: &str) -> Option<&'static str> {
3279    match mime {
3280        "image/png" => Some("png"),
3281        "image/jpeg" => Some("jpg"),
3282        "image/gif" => Some("gif"),
3283        "image/webp" => Some("webp"),
3284        _ => None,
3285    }
3286}
3287
3288/// Is `id` a shape [`put_attachment`](Talks::put_attachment) could have
3289/// produced? 32 lowercase hex digits and nothing else, checked before an id
3290/// that came from the client is ever allowed to build a path - so `..` and a
3291/// path separator are never even possible.
3292pub fn valid_attachment_id(id: &str) -> bool {
3293    id.len() == 32
3294        && id
3295            .bytes()
3296            .all(|b| b.is_ascii_digit() || (b'a'..=b'f').contains(&b))
3297}
3298
3299/// A fresh attachment id: 128 bits of process entropy as lowercase hex - the
3300/// same "mint it, never take it from the client" rule [`new_id`] follows for
3301/// conversation ids.
3302fn new_attachment_id() -> String {
3303    let mut r = crate::rng::SplitMix64::new(crate::rng::entropy());
3304    format!("{:016x}{:016x}", r.next_u64(), r.next_u64())
3305}
3306
3307#[cfg(test)]
3308mod tests {
3309    #[test]
3310    fn turn_access_lifts_the_sandbox_only_for_the_talk_opt_in() {
3311        use super::turn_access;
3312        assert_eq!(turn_access(false, false), (false, false));
3313        assert_eq!(turn_access(true, false), (true, true));
3314        // A handed-over question opens the question store, never the sandbox.
3315        assert_eq!(turn_access(false, true), (true, false));
3316        assert_eq!(turn_access(true, true), (true, true));
3317    }
3318
3319    #[test]
3320    fn the_briefing_points_at_origin_main_not_the_working_tree() {
3321        let b = briefing(Path::new("/r"), "en", false);
3322        assert!(b.contains("origin/main"));
3323        assert!(b.contains("git show origin/main:"));
3324    }
3325    use std::collections::BTreeMap;
3326
3327    use crate::config::{AgentChoice, AgentKind, AgentSpec, Graph};
3328    use crate::queue::{Queue, Source, Task};
3329
3330    use super::*;
3331
3332    fn ctx_agent(id: &str, model: Option<&str>) -> AgentSpec {
3333        AgentSpec {
3334            id: id.to_owned(),
3335            kind: AgentKind::Command,
3336            model: model.map(str::to_owned),
3337            command: Vec::new(),
3338            extra_args: Vec::new(),
3339            env: BTreeMap::new(),
3340            prompt_delivery: None,
3341        }
3342    }
3343
3344    fn ctx_talk(agent: &str, turns: Vec<Turn>) -> Talk {
3345        Talk {
3346            recommendations: Default::default(),
3347            posts_read_through: None,
3348            pending_breaks: None,
3349            pending_messages: Vec::new(),
3350            schema: SCHEMA,
3351            id: "20260904-014455-ab12".to_owned(),
3352            repo: PathBuf::from("."),
3353            agent: agent.to_owned(),
3354            status: TalkStatus::Open,
3355            turns,
3356            pending: String::new(),
3357            pending_attachments: Vec::new(),
3358            fallback: false,
3359            persona: String::new(),
3360            persona_dirty: false,
3361            implementers: 1,
3362            implementers_dirty: false,
3363            notified: Vec::new(),
3364            completion_notices: Vec::new(),
3365            created_at: Timestamp::now(),
3366            updated_at: Timestamp::now(),
3367            seat: SeatState::new(SEAT, agent, 1),
3368        }
3369    }
3370
3371    fn reply(body: &str, usage: Option<(u64, &str, Option<&str>)>) -> Turn {
3372        Turn {
3373            sender: None,
3374            breaks: None,
3375            messages: Vec::new(),
3376            posted_by: None,
3377            who: Who::Agent,
3378            body: body.to_owned(),
3379            at: Timestamp::now(),
3380            attachments: Vec::new(),
3381            usage: usage.map(|(t, a, m)| TurnUsage {
3382                context_tokens: t,
3383                agent: a.to_owned(),
3384                model: m.map(str::to_owned),
3385            }),
3386        }
3387    }
3388
3389    fn ctx_config(windows: &[(&str, u64)]) -> Config {
3390        Config {
3391            agents: vec![
3392                ctx_agent("small", Some("small-model")),
3393                ctx_agent("big", Some("big-model")),
3394                ctx_agent("plain", None),
3395            ],
3396            context_windows: windows.iter().map(|(k, v)| ((*k).to_owned(), *v)).collect(),
3397            ..Config::default()
3398        }
3399    }
3400
3401    #[test]
3402    fn context_usage_computes_percent_and_warns_at_eighty() {
3403        let cfg = ctx_config(&[("small-model", 1000)]);
3404        let at = |tokens| {
3405            let t = ctx_talk(
3406                "small",
3407                vec![reply("hi", Some((tokens, "small", Some("small-model"))))],
3408            );
3409            context_usage(&t, Some(&cfg))
3410        };
3411        let u = at(799);
3412        assert_eq!((u.percent, u.warn, u.window), (Some(79), false, Some(1000)));
3413        let u = at(800);
3414        assert_eq!((u.percent, u.warn), (Some(80), true));
3415        let u = at(1500);
3416        assert_eq!((u.percent, u.warn), (Some(150), true));
3417        assert!(!u.since_switch);
3418    }
3419
3420    #[test]
3421    fn context_usage_is_unknown_without_usage_and_never_looks_back() {
3422        let cfg = ctx_config(&[("small-model", 1000)]);
3423        let t = ctx_talk(
3424            "small",
3425            vec![
3426                reply("old", Some((900, "small", Some("small-model")))),
3427                reply("new", None),
3428            ],
3429        );
3430        let u = context_usage(&t, Some(&cfg));
3431        // No measurement in the latest reply: an estimate, never the stale 900.
3432        assert!(u.estimated);
3433        assert_ne!(u.tokens, Some(900));
3434        assert!(u.tokens.is_some());
3435        // A magi note after the reply neither hides nor replaces it.
3436        let t = ctx_talk(
3437            "small",
3438            vec![
3439                reply("old", Some((900, "small", Some("small-model")))),
3440                reply("magi: could not run agent", None),
3441            ],
3442        );
3443        assert_eq!(context_usage(&t, Some(&cfg)).tokens, Some(900));
3444        assert_eq!(
3445            context_usage(&ctx_talk("small", Vec::new()), Some(&cfg)).tokens,
3446            None
3447        );
3448    }
3449
3450    #[test]
3451    fn estimate_counts_chars_both_sides_and_standing_prompt() {
3452        let mut t = ctx_talk("small", vec![reply("abcdefg", None)]);
3453        assert_eq!(estimate_context_tokens(&t, 0), Some(2)); // 7 chars -> 2
3454        let op = Turn {
3455            sender: None,
3456            who: Who::Operator,
3457            ..reply("abcdefg", None)
3458        };
3459        t.turns.push(op);
3460        assert_eq!(estimate_context_tokens(&t, 0), Some(4));
3461        assert!(
3462            estimate_context_tokens(&t, 700).unwrap() > estimate_context_tokens(&t, 0).unwrap()
3463        );
3464        // Characters, not bytes: 7 kanji are 7 chars.
3465        let ja = ctx_talk("small", vec![reply("日本語日本語日", None)]);
3466        assert_eq!(estimate_context_tokens(&ja, 0), Some(2));
3467        // magi notes are not sent to the agent; with nothing else, unknown.
3468        let note = ctx_talk("small", vec![reply("magi: could not run agent", None)]);
3469        assert_eq!(estimate_context_tokens(&note, 1000), None);
3470        assert_eq!(
3471            estimate_context_tokens(&ctx_talk("small", Vec::new()), 1000),
3472            None
3473        );
3474    }
3475
3476    #[test]
3477    fn context_usage_measured_wins_and_estimate_gets_percent_and_warn() {
3478        let cfg = ctx_config(&[("small-model", 1000)]);
3479        let t = ctx_talk(
3480            "small",
3481            vec![reply(
3482                &"x".repeat(5000),
3483                Some((10, "small", Some("small-model"))),
3484            )],
3485        );
3486        let u = context_usage(&t, Some(&cfg));
3487        assert_eq!((u.tokens, u.estimated), (Some(10), false));
3488        let t = ctx_talk("small", vec![reply(&"x".repeat(5000), None)]);
3489        let u = context_usage(&t, Some(&cfg));
3490        assert!(u.estimated && !u.since_switch);
3491        assert_eq!(u.window, Some(1000));
3492        assert!(u.warn && u.percent.unwrap() >= 80);
3493        let t = ctx_talk("small", vec![reply("hi", None)]);
3494        let u = context_usage(&t, Some(&cfg));
3495        assert!(u.estimated && u.percent.is_some());
3496    }
3497
3498    #[test]
3499    fn context_usage_without_a_window_shows_tokens_only() {
3500        let cfg = ctx_config(&[]);
3501        // No model at all, and a model nothing matches.
3502        let t = ctx_talk("plain", vec![reply("hi", Some((5000, "plain", None)))]);
3503        let u = context_usage(&t, Some(&cfg));
3504        assert_eq!(
3505            (u.tokens, u.window, u.percent, u.warn),
3506            (Some(5000), None, None, false)
3507        );
3508        let t = ctx_talk(
3509            "small",
3510            vec![reply("hi", Some((5000, "small", Some("small-model"))))],
3511        );
3512        assert_eq!(context_usage(&t, Some(&cfg)).percent, None);
3513        // No readable config: same, and no panic.
3514        assert_eq!(context_usage(&t, None).window, None);
3515    }
3516
3517    #[test]
3518    fn context_usage_switching_model_changes_the_denominator() {
3519        let cfg = ctx_config(&[("small-model", 1000), ("big-model", 10_000)]);
3520        let used = reply("hi", Some((900, "small", Some("small-model"))));
3521        let before = context_usage(&ctx_talk("small", vec![used.clone()]), Some(&cfg));
3522        assert_eq!(
3523            (before.percent, before.warn, before.since_switch),
3524            (Some(90), true, false)
3525        );
3526        // Same turns, conversation now on the big model: new denominator, and
3527        // the figure is flagged as describing the previous session.
3528        let after = context_usage(&ctx_talk("big", vec![used]), Some(&cfg));
3529        assert_eq!(after.window, Some(10_000));
3530        assert_eq!(
3531            (after.percent, after.warn, after.since_switch),
3532            (Some(9), false, true)
3533        );
3534        assert_eq!(after.model.as_deref(), Some("big-model"));
3535    }
3536
3537    #[test]
3538    fn a_turn_recorded_before_usage_existed_still_reads() {
3539        let old = r#"{"who":"agent","body":"hi","at":"2026-09-04T01:44:55Z"}"#;
3540        let turn: Turn = serde_json::from_str(old).expect("old turn reads");
3541        assert!(turn.usage.is_none());
3542        let json = serde_json::to_string(&turn).expect("serialize");
3543        assert!(
3544            !json.contains("usage"),
3545            "absent usage is not written: {json}"
3546        );
3547    }
3548
3549    /// A store of its own, with no process-global state.
3550    fn store() -> (tempfile::TempDir, Talks) {
3551        let tmp = tempfile::tempdir().expect("tempdir");
3552        let talks = Talks::at(tmp.path().join("talks"));
3553        (tmp, talks)
3554    }
3555
3556    /// A `kind = "command"` agent whose whole behaviour is a POSIX shell
3557    /// script - see `chat`'s tests for why no test here may spawn a real
3558    /// agent CLI.
3559    fn mock_agent(dir: &Path, script: &str, env: BTreeMap<String, String>) -> AgentSpec {
3560        let path = dir.join("mock-talk-agent.sh");
3561        std::fs::write(&path, script).expect("write mock");
3562        AgentSpec {
3563            id: "mock".to_owned(),
3564            kind: AgentKind::Command,
3565            model: None,
3566            command: vec!["sh".to_owned(), path.to_string_lossy().into_owned()],
3567            extra_args: Vec::new(),
3568            env,
3569            prompt_delivery: None,
3570        }
3571    }
3572
3573    fn config(spec: AgentSpec) -> Config {
3574        Config {
3575            agents: vec![spec],
3576            graph: Graph {
3577                language: "en".to_owned(),
3578                ..Graph::default()
3579            },
3580            ..Config::default()
3581        }
3582    }
3583
3584    /// Echo a canned reply, ignoring the prompt on stdin.
3585    const REPLY: &str = "#!/bin/sh\ncat >/dev/null\nprintf '%s\\n' \"$MOCK_REPLY\"\n";
3586
3587    /// Say nothing and fail, the way a CLI that cannot start does.
3588    const BROKEN: &str = "#!/bin/sh\ncat >/dev/null\nexit 3\n";
3589
3590    /// Reply with the prompt it was given, so a test can inspect exactly what
3591    /// the agent received on stdin.
3592    const ECHO: &str = "#!/bin/sh\ncat\n";
3593
3594    fn env(reply: &str) -> BTreeMap<String, String> {
3595        BTreeMap::from([("MOCK_REPLY".to_owned(), reply.to_owned())])
3596    }
3597
3598    #[test]
3599    fn the_frozen_json_field_names_round_trip_through_disk() {
3600        let (tmp, talks) = store();
3601        let mut talk = Talk {
3602            recommendations: Default::default(),
3603            posts_read_through: None,
3604            pending_breaks: None,
3605            pending_messages: Vec::new(),
3606            schema: SCHEMA,
3607            id: "20260904-014455-ab12".to_owned(),
3608            repo: tmp.path().to_owned(),
3609            agent: "sonnet".to_owned(),
3610            status: TalkStatus::Open,
3611            turns: Vec::new(),
3612            pending: String::new(),
3613            pending_attachments: Vec::new(),
3614            fallback: false,
3615            persona: String::new(),
3616            persona_dirty: false,
3617            implementers: 1,
3618            implementers_dirty: false,
3619            notified: Vec::new(),
3620            completion_notices: Vec::new(),
3621            created_at: Timestamp::now(),
3622            updated_at: Timestamp::now(),
3623            seat: SeatState::new(SEAT, "sonnet", 7),
3624        };
3625        talks.put(&mut talk).expect("put");
3626
3627        let raw = std::fs::read_to_string(talks.path_of(&talk.id)).expect("read back");
3628        let v: serde_json::Value = serde_json::from_str(&raw).expect("parse");
3629        for field in [
3630            "schema",
3631            "id",
3632            "repo",
3633            "agent",
3634            "status",
3635            "turns",
3636            "created_at",
3637            "updated_at",
3638        ] {
3639            assert!(v.get(field).is_some(), "missing field `{field}`");
3640        }
3641        assert_eq!(v["schema"], 1);
3642        assert_eq!(v["status"], "open");
3643
3644        let back = talks.get(&talk.id).expect("get");
3645        assert_eq!(back.posts_read_through, None);
3646        assert_eq!(back.id, talk.id);
3647        assert_eq!(back.status, TalkStatus::Open);
3648    }
3649
3650    #[test]
3651    fn opening_a_talk_takes_no_agent_turn() {
3652        let (tmp, talks) = store();
3653        // A script that would fail loudly if it were ever run: `begin` must
3654        // not invoke anything, since there is nothing yet for an agent to
3655        // answer.
3656        let spec = mock_agent(tmp.path(), BROKEN, BTreeMap::new());
3657        let cfg = config(spec);
3658
3659        let talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
3660        assert_eq!(talk.status, TalkStatus::Open);
3661        assert!(talk.turns.is_empty(), "nothing has been said yet");
3662
3663        let on_disk = talks.get(&talk.id).expect("get");
3664        assert_eq!(on_disk.turns.len(), 0);
3665    }
3666
3667    /// `[roles] chatter`, when set, decides who holds this conversation; unset,
3668    /// it falls back to [`agent::pick`]'s own default order (a claude seat,
3669    /// else the first runnable agent in roster order) rather than to any
3670    /// other role - see `[roles] chatter`'s own doc in [`crate::config`] for
3671    /// why a dedicated field exists at all: opening this against the same
3672    /// seat as a judge is what produced the `agent ... did not answer within
3673    /// 300s` timeout that led to it.
3674    #[test]
3675    fn chatter_wins_when_set_and_falls_back_to_pick_s_default_order_otherwise() {
3676        let (tmp, talks) = store();
3677        let first_spec = mock_agent(tmp.path(), BROKEN, BTreeMap::new());
3678        let mut chatter_spec = mock_agent(tmp.path(), BROKEN, BTreeMap::new());
3679        chatter_spec.id = "chatter-mock".to_owned();
3680
3681        let mut cfg = Config {
3682            agents: vec![first_spec.clone(), chatter_spec.clone()],
3683            graph: Graph {
3684                language: "en".to_owned(),
3685                ..Graph::default()
3686            },
3687            ..Config::default()
3688        };
3689        cfg.roles.chatter = Some(chatter_spec.id.as_str().into());
3690
3691        let talk =
3692            begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin with chatter set");
3693        assert_eq!(talk.agent, chatter_spec.id, "an explicit chatter must win");
3694
3695        cfg.roles.chatter = None;
3696        let fallback =
3697            begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin with chatter unset");
3698        assert_eq!(
3699            fallback.agent, first_spec.id,
3700            "unset chatter must fall back to agent::pick's own default order"
3701        );
3702    }
3703
3704    /// A conversation recorded before attachments existed - schema 1, no
3705    /// `attachments` key on any turn - must still read.
3706    #[test]
3707    fn a_talk_recorded_without_attachments_still_reads() {
3708        let (tmp, talks) = store();
3709        let path = talks.path_of("20260904-014455-ab12");
3710        std::fs::create_dir_all(talks.root()).expect("talks dir");
3711        std::fs::write(
3712            &path,
3713            serde_json::json!({
3714                "schema": 1,
3715                "id": "20260904-014455-ab12",
3716                "repo": tmp.path(),
3717                "agent": "sonnet",
3718                "status": "open",
3719                "turns": [
3720                    { "who": "operator", "body": "still there?",
3721                      "at": Timestamp::now().to_string() },
3722                ],
3723                "created_at": Timestamp::now().to_string(),
3724                "updated_at": Timestamp::now().to_string(),
3725                "seat": SeatState::new(SEAT, "sonnet", 7),
3726            })
3727            .to_string(),
3728        )
3729        .expect("write pre-attachments talk");
3730
3731        let talk = talks.get("20260904-014455-ab12").expect("must still read");
3732        assert!(talk.turns[0].attachments.is_empty());
3733    }
3734
3735    fn lease_store() -> (tempfile::TempDir, Talks, Talks) {
3736        let tmp = tempfile::TempDir::new().expect("tmp");
3737        let root = tmp.path().join("talks");
3738        (tmp, Talks::at(root.clone()), Talks::at(root))
3739    }
3740
3741    #[test]
3742    fn two_starters_on_one_talk_one_wins_and_the_other_is_refused() {
3743        let (_tmp, a, b) = lease_store();
3744        let won = a.claim_turn("t1").expect("claim").expect("first wins");
3745        assert!(
3746            b.claim_turn("t1").expect("claim").is_none(),
3747            "second is refused"
3748        );
3749        assert!(b.turn_held("t1"));
3750        assert!(
3751            b.claim_turn("t2").expect("claim").is_some(),
3752            "other talks are free"
3753        );
3754        drop(won);
3755    }
3756
3757    #[test]
3758    fn a_stale_lease_is_taken_over_and_the_old_guard_cannot_release_it() {
3759        let (_tmp, a, b) = lease_store();
3760        let old = a.claim_turn("t1").expect("claim").expect("held");
3761        let later = Timestamp::now()
3762            .checked_add(jiff::SignedDuration::from_secs(
3763                crate::ask::LEASE_TTL.as_secs() as i64 + 5,
3764            ))
3765            .expect("later");
3766        let new = b
3767            .claim_turn_at("t1", later)
3768            .expect("claim")
3769            .expect("a stale lease is taken over");
3770        drop(old);
3771        assert!(a.turn_held("t1"), "the old guard left the new lease alone");
3772        assert!(new.beat().expect("beat"), "the new owner still beats");
3773        drop(new);
3774        assert!(!a.turn_held("t1"));
3775    }
3776
3777    #[tokio::test]
3778    async fn a_turn_whose_lease_was_taken_over_is_stopped() {
3779        let (_tmp, a, b) = lease_store();
3780        let old = a.claim_turn("t1").expect("claim").expect("held");
3781        let later = Timestamp::now()
3782            .checked_add(jiff::SignedDuration::from_secs(
3783                crate::ask::LEASE_TTL.as_secs() as i64 + 5,
3784            ))
3785            .expect("later");
3786        let _new = b
3787            .claim_turn_at("t1", later)
3788            .expect("claim")
3789            .expect("taken over");
3790        let out = old
3791            .beating_every(Duration::from_millis(10), std::future::pending::<()>())
3792            .await;
3793        assert!(out.is_err(), "the displaced turn must stop, not run on");
3794    }
3795
3796    #[tokio::test]
3797    async fn a_turn_that_finishes_is_returned_and_keeps_its_lease_beating() {
3798        let (_tmp, a, _b) = lease_store();
3799        let lease = a.claim_turn("t1").expect("claim").expect("held");
3800        let out = lease
3801            .beating_every(Duration::from_millis(5), async {
3802                tokio::time::sleep(Duration::from_millis(40)).await;
3803                7
3804            })
3805            .await
3806            .expect("still ours");
3807        assert_eq!(out, 7);
3808        assert!(a.turn_held("t1"));
3809    }
3810
3811    #[test]
3812    fn an_unreadable_lease_counts_as_stale() {
3813        let (_tmp, a, b) = lease_store();
3814        std::fs::create_dir_all(a.root()).expect("dir");
3815        std::fs::write(a.turn_path("t1"), "not json").expect("write");
3816        assert!(!a.turn_held("t1"));
3817        assert!(b.claim_turn("t1").expect("claim").is_some());
3818    }
3819
3820    #[test]
3821    fn without_hard_links_a_claim_is_still_exclusive_and_a_young_placeholder_blocks() {
3822        let (_tmp, a, b) = lease_store();
3823        failpoint::force_no_link(true);
3824        let held = a.claim_turn("t1").expect("claim").expect("first wins");
3825        assert!(b.claim_turn("t1").expect("claim").is_none());
3826        assert!(a.turn_held("t1"));
3827        drop(held);
3828        assert!(!a.turn_held("t1"));
3829        // A writer between its exclusive create and its write: an empty file.
3830        let path = a.turn_path("t1");
3831        assert!(create_exclusive(&path, "").expect("placeholder"));
3832        assert!(b.claim_turn("t1").expect("claim").is_none(), "young: held");
3833        age_file(&path);
3834        assert!(b.claim_turn("t1").expect("claim").is_some(), "old: stale");
3835        // The lock publishes through the same path.
3836        let lease = a.turn_path("t2");
3837        let lock = TurnLock::take(&lease).expect("take").expect("lock");
3838        assert!(TurnLock::take(&lease).expect("take").is_none());
3839        drop(lock);
3840        assert!(TurnLock::take(&lease).expect("take").is_some());
3841        failpoint::force_no_link(false);
3842    }
3843
3844    #[test]
3845    fn remove_if_carries_removes_only_the_expected_content() {
3846        let (_tmp, a, _b) = lease_store();
3847        std::fs::create_dir_all(a.root()).expect("dir");
3848        let p = a.root().join("x.turn.lock");
3849        std::fs::write(&p, "mine").expect("write");
3850        assert!(!remove_if_carries(&p, "other"));
3851        assert_eq!(std::fs::read_to_string(&p).expect("kept"), "mine");
3852        assert!(remove_if_carries(&p, "mine"));
3853        assert!(!p.exists());
3854        assert!(!remove_if_carries(&p, "mine"), "absent is not a removal");
3855    }
3856
3857    #[test]
3858    fn a_dropped_lock_does_not_remove_a_lock_taken_over_since() {
3859        let (_tmp, a, _b) = lease_store();
3860        std::fs::create_dir_all(a.root()).expect("dir");
3861        let lease = a.turn_path("t1");
3862        let lock = TurnLock::take(&lease).expect("take").expect("lock");
3863        let path = lock.path.clone();
3864        std::fs::write(&path, "someone-else").expect("replace");
3865        drop(lock);
3866        assert_eq!(
3867            std::fs::read_to_string(&path).expect("kept"),
3868            "someone-else"
3869        );
3870    }
3871
3872    #[test]
3873    fn concurrent_takeovers_of_a_stale_lease_have_one_winner() {
3874        let (_tmp, a, _b) = lease_store();
3875        drop(a.claim_turn("t1").expect("claim").expect("held"));
3876        std::fs::write(
3877            a.turn_path("t1"),
3878            serde_json::to_string(&TurnRecord {
3879                token: "gone".into(),
3880                pid: 1,
3881                beat_at: Timestamp::from_second(1).expect("ts"),
3882            })
3883            .expect("json"),
3884        )
3885        .expect("write");
3886        let wins: Vec<_> = std::thread::scope(|sc| {
3887            let hs: Vec<_> = (0..8)
3888                .map(|_| {
3889                    let s = a.clone();
3890                    sc.spawn(move || s.claim_turn("t1").expect("claim"))
3891                })
3892                .collect();
3893            hs.into_iter().map(|h| h.join().expect("join")).collect()
3894        });
3895        assert_eq!(wins.iter().filter(|w| w.is_some()).count(), 1);
3896    }
3897
3898    fn age_file(path: &Path) {
3899        let f = std::fs::OpenOptions::new()
3900            .write(true)
3901            .open(path)
3902            .expect("open");
3903        f.set_modified(std::time::SystemTime::now() - std::time::Duration::from_secs(60))
3904            .expect("age");
3905    }
3906
3907    #[test]
3908    fn a_late_taker_of_a_broken_lock_cannot_disturb_its_replacement() {
3909        let (_tmp, a, _b) = lease_store();
3910        let lease = a.turn_path("t1");
3911        let lock = lease.with_extension("turn.lock");
3912        std::fs::create_dir_all(lock.parent().expect("dir")).expect("dir");
3913        std::fs::write(&lock, "t1-dead").expect("dead lock");
3914        age_file(&lock);
3915        // B breaks the dead lock and holds a fresh one.
3916        let b = TurnLock::take(&lease).expect("take").expect("b wins");
3917        let fresh = std::fs::read_to_string(&lock).expect("read");
3918        assert_eq!(fresh, b.token);
3919        // C read the same dead token earlier: its ticket is already spent, and
3920        // the fresh lock is never moved, so a newcomer cannot slip in.
3921        let ticket = std::fs::read_dir(lock.parent().expect("dir"))
3922            .expect("dir")
3923            .flatten()
3924            .map(|e| e.path())
3925            .find(|p| p.to_string_lossy().ends_with(".break.0"))
3926            .expect("ticket");
3927        assert!(!create_exclusive(&ticket, "").expect("ticket"));
3928        assert!(TurnLock::take(&lease).expect("take").is_none());
3929        assert_eq!(std::fs::read_to_string(&lock).expect("read"), fresh);
3930        // A later generation is breakable once the old ticket is stale too.
3931        age_file(&lock);
3932        age_file(&ticket);
3933        std::mem::forget(b);
3934        let c = TurnLock::take(&lease)
3935            .expect("take")
3936            .expect("next generation");
3937        assert_ne!(c.token, fresh);
3938    }
3939
3940    #[test]
3941    fn a_live_ticket_blocks_and_a_stale_one_hands_over_to_the_next_generation() {
3942        let (_tmp, a, _b) = lease_store();
3943        let lease = a.turn_path("t1");
3944        let lock = lease.with_extension("turn.lock");
3945        std::fs::create_dir_all(lock.parent().expect("dir")).expect("dir");
3946        std::fs::write(&lock, "t1-dead").expect("dead lock");
3947        age_file(&lock);
3948        let t0 = lock.with_extension("lock.t1-dead.break.0");
3949        assert!(create_exclusive(&t0, "").expect("ticket"));
3950        // A fresh ticket 0 blocks the same dead token, whatever the clock says.
3951        assert!(TurnLock::take(&lease).expect("take").is_none());
3952        assert_eq!(std::fs::read_to_string(&lock).expect("read"), "t1-dead");
3953        // Once ticket 0 is stale, exactly one taker gets through via ticket 1.
3954        age_file(&t0);
3955        let c = TurnLock::take(&lease).expect("take").expect("generation 1");
3956        assert_eq!(std::fs::read_to_string(&lock).expect("read"), c.token);
3957        assert!(lock.with_extension("lock.t1-dead.break.1").exists());
3958        assert!(TurnLock::take(&lease).expect("take").is_none());
3959    }
3960
3961    #[test]
3962    fn exhausted_ticket_generations_recover_once_the_sweep_ages_them_out() {
3963        let (_tmp, a, _b) = lease_store();
3964        let lease = a.turn_path("t1");
3965        let lock = lease.with_extension("turn.lock");
3966        std::fs::create_dir_all(lock.parent().expect("dir")).expect("dir");
3967        std::fs::write(&lock, "t1-dead").expect("dead lock");
3968        age_file(&lock);
3969        let tickets: Vec<_> = (0..TICKET_GENERATIONS)
3970            .map(|n| lock.with_extension(format!("lock.t1-dead.break.{n}")))
3971            .collect();
3972        for t in &tickets {
3973            assert!(create_exclusive(t, "").expect("ticket"));
3974            age_file(t);
3975        }
3976        // Stale but not yet swept: no recovery, and no ticket is unlinked.
3977        assert!(TurnLock::take(&lease).expect("take").is_none());
3978        assert!(tickets.iter().all(|t| t.exists()));
3979        // Past the sweep age they go, and the next attempt recovers.
3980        for t in &tickets {
3981            let f = std::fs::OpenOptions::new()
3982                .write(true)
3983                .open(t)
3984                .expect("open");
3985            f.set_modified(std::time::SystemTime::now() - TICKET_SWEEP_AGE * 2)
3986                .expect("age");
3987        }
3988        assert!(TurnLock::take(&lease).expect("take").is_none());
3989        assert!(TurnLock::take(&lease).expect("take").is_some());
3990    }
3991
3992    #[test]
3993    fn an_aged_empty_lock_is_broken() {
3994        let (_tmp, a, _b) = lease_store();
3995        let lease = a.turn_path("t1");
3996        let lock = lease.with_extension("turn.lock");
3997        std::fs::create_dir_all(lock.parent().expect("dir")).expect("dir");
3998        std::fs::write(&lock, "").expect("empty lock");
3999        age_file(&lock);
4000        assert!(TurnLock::take(&lease).expect("take").is_some());
4001    }
4002
4003    #[test]
4004    fn reply_plain_text_sends_keep_provenance_and_boundaries() {
4005        let (tmp, talks) = store();
4006        let cfg = config(mock_agent(tmp.path(), REPLY, env("reply")));
4007        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4008        record(&mut talk, &talks, "original", Vec::new()).expect("record");
4009        queue(&mut talk, &talks, "first", Vec::new()).expect("queue");
4010        queue(&mut talk, &talks, "second", Vec::new()).expect("queue");
4011        let wire: serde_json::Value =
4012            serde_json::from_str(&std::fs::read_to_string(talks.path_of(&talk.id)).unwrap())
4013                .unwrap();
4014        assert_eq!(wire["pending_messages"][0]["sender"], "operator");
4015        assert_eq!(wire["turns"][0]["messages"][0]["sender"], "operator");
4016        drain(&mut talk, &talks).expect("drain");
4017        let saved = talks.get(&talk.id).expect("reload");
4018        assert_eq!(saved.turns[1].body, "first\n\nsecond");
4019        assert_eq!(saved.turns[1].breaks, Some(vec![7]));
4020        let wire: serde_json::Value =
4021            serde_json::from_str(&std::fs::read_to_string(talks.path_of(&saved.id)).unwrap())
4022                .unwrap();
4023        // Explicit provenance and send boundaries survive disk storage.
4024        for turn in wire["turns"].as_array().expect("turns") {
4025            for field in turn.as_object().expect("turn").keys() {
4026                assert!(
4027                    [
4028                        "who",
4029                        "body",
4030                        "at",
4031                        "attachments",
4032                        "usage",
4033                        "breaks",
4034                        "posted_by",
4035                        "sender",
4036                        "messages"
4037                    ]
4038                    .contains(&field.as_str()),
4039                    "unknown turn field: {field}"
4040                );
4041            }
4042        }
4043        record_reply(
4044            &mut talk,
4045            &talks,
4046            "reply",
4047            Vec::new(),
4048            Some(ReplyTarget {
4049                turn: 1,
4050                message: 1,
4051            }),
4052        )
4053        .expect("reply to reloaded plain text");
4054        assert_eq!(
4055            talk.turns[2].messages[0].reply.as_ref().unwrap().excerpt,
4056            "second"
4057        );
4058    }
4059
4060    #[tokio::test]
4061    async fn reply_image_identity_reaches_fresh_and_followup_prompts_after_reload() {
4062        let (tmp, talks) = store();
4063        let cfg = config(mock_agent(tmp.path(), ECHO, BTreeMap::new()));
4064        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4065        let first = talks
4066            .put_attachment(&talk.id, "image/png", "first.png", b"first")
4067            .expect("first");
4068        let second = talks
4069            .put_attachment(&talk.id, "image/png", "second.png", b"second")
4070            .expect("second");
4071        let first_path = talks.attachment_path(&talk.id, &first).unwrap();
4072        let second_path = talks.attachment_path(&talk.id, &second).unwrap();
4073        // Two attachment-only sends share one turn but retain their ownership.
4074        queue(&mut talk, &talks, "", vec![first]).expect("queue first");
4075        queue(&mut talk, &talks, "", vec![second]).expect("queue second");
4076        drain(&mut talk, &talks).expect("drain");
4077        let lease = talks.claim_turn(&talk.id).expect("claim").expect("lease");
4078        for (message, selected, other) in [
4079            (0, &first_path, &second_path),
4080            (1, &second_path, &first_path),
4081        ] {
4082            record_reply(
4083                &mut talk,
4084                &talks,
4085                "describe this",
4086                Vec::new(),
4087                Some(ReplyTarget { turn: 0, message }),
4088            )
4089            .expect("reply");
4090            talk = talks.get(&talk.id).expect("reload");
4091            let quote = message_prompt(talk.turns.last().unwrap(), &talk, &talks);
4092            assert!(quote.contains(selected.to_str().unwrap()));
4093            assert!(!quote.contains(other.to_str().unwrap()));
4094            respond(&lease, &mut talk, &talks, &cfg, "describe this")
4095                .await
4096                .expect("respond");
4097            assert!(
4098                talk.turns
4099                    .last()
4100                    .unwrap()
4101                    .body
4102                    .contains(&format!("Replied-to image: {}", selected.display()))
4103            );
4104        }
4105    }
4106
4107    #[test]
4108    fn reply_sends_survive_queue_edit_drain_reload_and_legacy_records() {
4109        let (tmp, talks) = store();
4110        let cfg = config(mock_agent(tmp.path(), REPLY, env("reply")));
4111        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4112        record(&mut talk, &talks, &"猫".repeat(300), Vec::new()).expect("record");
4113        // Old records omit the new metadata entirely.
4114        talk.turns[0].messages.clear();
4115        talks.put(&mut talk).expect("legacy save");
4116        let target = ReplyTarget {
4117            turn: 0,
4118            message: 0,
4119        };
4120        queue_reply(&mut talk, &talks, "first", Vec::new(), Some(target.clone())).expect("first");
4121        queue_reply(&mut talk, &talks, "second", Vec::new(), None).expect("second");
4122        assert!(
4123            edit_pending_message(&mut talk, &talks, "edited", "first\n\nsecond", &[], Some(0))
4124                .expect("edit")
4125        );
4126        drain(&mut talk, &talks).expect("drain");
4127        let saved = talks.get(&talk.id).expect("reload");
4128        assert_eq!(saved.turns[1].body, "edited\n\nsecond");
4129        assert_eq!(saved.turns[1].messages.len(), 2);
4130        let reply = saved.turns[1].messages[0].reply.as_ref().expect("reply");
4131        assert_eq!(reply.target, target);
4132        assert_eq!(reply.excerpt, "猫".repeat(240));
4133        assert!(message_prompt(&saved.turns[1], &saved, &talks).contains("[Reply to Operator:"));
4134        assert!(transcript(&saved, &talks).contains("[Reply to Operator:"));
4135        queue_reply(
4136            &mut talk,
4137            &talks,
4138            "third",
4139            Vec::new(),
4140            Some(ReplyTarget {
4141                turn: 1,
4142                message: 1,
4143            }),
4144        )
4145        .expect("specific send");
4146        assert_eq!(
4147            talk.pending_messages[0]
4148                .reply
4149                .as_ref()
4150                .expect("reply")
4151                .excerpt,
4152            "second"
4153        );
4154        clear_pending(&mut talk, &talks).expect("clear");
4155        assert!(talk.pending_messages.is_empty());
4156        assert!(
4157            record_reply(
4158                &mut talk,
4159                &talks,
4160                "invalid",
4161                Vec::new(),
4162                Some(ReplyTarget {
4163                    turn: 1,
4164                    message: 2
4165                })
4166            )
4167            .is_err()
4168        );
4169        assert_eq!(talks.get(&talk.id).expect("reload").turns.len(), 2);
4170    }
4171
4172    #[tokio::test]
4173    async fn reply_quotes_reach_fresh_and_followup_prompts_without_changing_body() {
4174        let (tmp, talks) = store();
4175        let cfg = config(mock_agent(tmp.path(), ECHO, BTreeMap::new()));
4176        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4177        record(&mut talk, &talks, "original", Vec::new()).expect("record");
4178        let target = ReplyTarget {
4179            turn: 0,
4180            message: 0,
4181        };
4182        record_reply(
4183            &mut talk,
4184            &talks,
4185            "answer",
4186            Vec::new(),
4187            Some(target.clone()),
4188        )
4189        .expect("reply");
4190        let lease = talks.claim_turn(&talk.id).expect("claim").expect("lease");
4191        respond(&lease, &mut talk, &talks, &cfg, "answer")
4192            .await
4193            .expect("respond");
4194        assert!(
4195            talk.turns
4196                .last()
4197                .expect("response")
4198                .body
4199                .contains("[Reply to Operator: original]\nanswer")
4200        );
4201        assert_eq!(talk.turns[1].body, "answer");
4202        record_reply(&mut talk, &talks, "followup", Vec::new(), Some(target)).expect("reply");
4203        respond(&lease, &mut talk, &talks, &cfg, "followup")
4204            .await
4205            .expect("respond");
4206        assert!(
4207            talk.turns
4208                .last()
4209                .expect("response")
4210                .body
4211                .contains("[Reply to Operator: original]\nfollowup")
4212        );
4213    }
4214
4215    #[test]
4216    fn owner_indentation_survives_record_queue_and_edit() {
4217        let (tmp, talks) = store();
4218        let cfg = config(mock_agent(tmp.path(), REPLY, env("reply")));
4219        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4220        for text in ["    Post original", "\tUse rewrite"] {
4221            record_reply(&mut talk, &talks, text, Vec::new(), None).unwrap();
4222            assert_eq!(
4223                talks.get(&talk.id).unwrap().turns.last().unwrap().body,
4224                text
4225            );
4226            queue(&mut talk, &talks, "draft", Vec::new()).unwrap();
4227            assert!(edit_pending_message(&mut talk, &talks, text, "draft", &[], Some(0)).unwrap());
4228            let saved = talks.get(&talk.id).unwrap();
4229            assert_eq!(saved.pending, text);
4230            assert_eq!(saved.pending_messages[0].body, text);
4231            assert!(edit_pending_message(&mut talk, &talks, text, text, &[], None).unwrap());
4232            let saved = talks.get(&talk.id).unwrap();
4233            assert_eq!(saved.pending, text);
4234            assert_eq!(saved.pending_messages[0].body, text);
4235            assert_eq!(drain(&mut talk, &talks).unwrap().as_deref(), Some(text));
4236        }
4237    }
4238
4239    #[test]
4240    fn queued_text_is_durable_combined_and_drained_as_one_operator_turn() {
4241        let (tmp, talks) = store();
4242        let cfg = config(mock_agent(tmp.path(), REPLY, env("reply")));
4243        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4244
4245        queue(&mut talk, &talks, "first", Vec::new()).expect("queue first");
4246        queue(&mut talk, &talks, "second", Vec::new()).expect("queue second");
4247        let saved = talks.get(&talk.id).expect("reload queued talk");
4248        assert_eq!(saved.pending, "first\n\nsecond");
4249        assert_eq!(saved.pending_breaks, Some(vec!["first\n\n".len()]));
4250        assert!(saved.turns.is_empty(), "a draft is not a transcript turn");
4251
4252        let drained = drain(&mut talk, &talks).expect("drain");
4253        assert_eq!(drained.as_deref(), Some("first\n\nsecond"));
4254        let saved = talks.get(&talk.id).expect("reload drained talk");
4255        assert!(saved.pending.is_empty());
4256        assert_eq!(saved.turns.len(), 1);
4257        assert_eq!(saved.turns[0].body, "first\n\nsecond");
4258        assert_eq!(saved.turns[0].breaks, Some(vec!["first\n\n".len()]));
4259        assert_eq!(saved.pending_breaks, None);
4260    }
4261
4262    #[test]
4263    fn editing_a_queued_draft_preserves_its_attachments_and_rejects_a_stale_snapshot() {
4264        let (tmp, talks) = store();
4265        let cfg = config(mock_agent(tmp.path(), REPLY, env("reply")));
4266        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4267        let attachment = Attachment {
4268            id: "a".repeat(32),
4269            name: "shot.png".to_owned(),
4270            mime: "image/png".to_owned(),
4271            bytes: 3,
4272        };
4273
4274        queue(&mut talk, &talks, "first", vec![attachment.clone()]).expect("queue");
4275        assert!(
4276            edit_pending_text(
4277                &mut talk,
4278                &talks,
4279                "corrected",
4280                "first",
4281                std::slice::from_ref(&attachment.id),
4282            )
4283            .expect("edit")
4284        );
4285        let saved = talks.get(&talk.id).expect("reload edited draft");
4286        assert_eq!(saved.pending, "corrected");
4287        assert_eq!(saved.pending_attachments, vec![attachment]);
4288
4289        queue(&mut talk, &talks, "later", Vec::new()).expect("queue concurrent draft");
4290        assert!(
4291            !edit_pending_text(
4292                &mut talk,
4293                &talks,
4294                "stale edit",
4295                "corrected",
4296                &["a".repeat(32)],
4297            )
4298            .expect("stale edit is a conflict")
4299        );
4300        assert_eq!(
4301            talks.get(&talk.id).expect("reload after conflict").pending,
4302            "corrected\n\nlater"
4303        );
4304        assert!(
4305            !clear_pending_if_matches(&mut talk, &talks, "corrected", &["a".repeat(32)])
4306                .expect("stale clear is a conflict")
4307        );
4308        assert_eq!(
4309            talks
4310                .get(&talk.id)
4311                .expect("reload after stale clear")
4312                .pending,
4313            "corrected\n\nlater"
4314        );
4315    }
4316
4317    #[tokio::test]
4318    async fn a_reply_save_preserves_pending_accepted_while_the_cli_runs() {
4319        let (tmp, talks) = store();
4320        let slow = "#!/bin/sh\ncat >/dev/null\nsleep 0.1\nprintf reply\n";
4321        let cfg = config(mock_agent(tmp.path(), slow, BTreeMap::new()));
4322        let mut running = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4323        let id = running.id.clone();
4324        let first = record(&mut running, &talks, "first", Vec::new()).expect("record");
4325
4326        let response_talks = talks.clone();
4327        let response_cfg = cfg.clone();
4328        let reply = tokio::spawn(async move {
4329            respond(
4330                &response_talks.claim_turn(&running.id).unwrap().unwrap(),
4331                &mut running,
4332                &response_talks,
4333                &response_cfg,
4334                &first,
4335            )
4336            .await
4337        });
4338        tokio::time::sleep(std::time::Duration::from_millis(20)).await;
4339
4340        let mut queued = talks.get(&id).expect("queued handle");
4341        queue(&mut queued, &talks, "next", Vec::new()).expect("queue");
4342        reply.await.expect("join").expect("reply");
4343
4344        let saved = talks.get(&id).expect("reload");
4345        assert_eq!(saved.pending, "next");
4346        assert_eq!(saved.pending_breaks, Some(Vec::new()));
4347        assert_eq!(saved.turns.len(), 2, "operator message and reply remain");
4348    }
4349
4350    #[tokio::test]
4351    async fn posts_during_a_resumed_turn_are_replayed_once_even_after_a_reply_save_failure() {
4352        for lose_reply in [false, true] {
4353            let (tmp, talks) = store();
4354            let cfg = config(mock_agent(tmp.path(), ECHO, BTreeMap::new()));
4355            let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4356            say(
4357                &talks.claim_turn(&talk.id).unwrap().unwrap(),
4358                &mut talk,
4359                &talks,
4360                &cfg,
4361                "establish the session",
4362                Vec::new(),
4363            )
4364            .await
4365            .expect("first turn");
4366            assert!(agent::has_session(
4367                AgentKind::Command,
4368                &talk.seat,
4369                cfg.graph.sessions
4370            ));
4371
4372            post(&talks, &talk.id, "already available", "task-earlier").expect("earlier post");
4373            talk = talks.get(&talk.id).expect("reload earlier post");
4374            let ready = tmp.path().join("ready");
4375            let release = tmp.path().join("release");
4376            let slow = "#!/bin/sh\ncat >/dev/null\ntouch \"$READY\"\nn=0\nwhile [ ! -f \"$RELEASE\" ] && [ \"$n\" -lt 200 ]; do sleep 0.01; n=$((n + 1)); done\nprintf reply\n";
4377            let slow_cfg = config(mock_agent(
4378                tmp.path(),
4379                slow,
4380                BTreeMap::from([
4381                    ("READY".to_owned(), ready.to_string_lossy().into_owned()),
4382                    ("RELEASE".to_owned(), release.to_string_lossy().into_owned()),
4383                ]),
4384            ));
4385            let text = record(&mut talk, &talks, "keep going", Vec::new()).expect("record");
4386            let prompt_turns = talk.turns.len();
4387            let id = talk.id.clone();
4388            let response_talks = talks.clone();
4389            let running = tokio::spawn(async move {
4390                respond(
4391                    &response_talks.claim_turn(&talk.id).unwrap().unwrap(),
4392                    &mut talk,
4393                    &response_talks,
4394                    &slow_cfg,
4395                    &text,
4396                )
4397                .await
4398            });
4399            tokio::time::timeout(std::time::Duration::from_secs(5), async {
4400                while !ready.exists() {
4401                    tokio::time::sleep(std::time::Duration::from_millis(10)).await;
4402                }
4403            })
4404            .await
4405            .expect("CLI read its prompt");
4406            post(&talks, &id, "concurrent task result", "task-one").expect("post");
4407            post(&talks, &id, "second task result", "task-two").expect("second post");
4408            if lose_reply {
4409                failpoint::force_put_failures(PUT_RETRIES);
4410            }
4411            std::fs::write(release, "go").expect("release CLI");
4412            let result = running.await.expect("join");
4413            if lose_reply {
4414                assert!(
4415                    result
4416                        .unwrap_err()
4417                        .to_string()
4418                        .contains("could not be saved")
4419                );
4420            } else {
4421                result.expect("reply");
4422            }
4423
4424            let mut saved = talks.get(&id).expect("reload");
4425            assert_eq!(saved.posts_read_through, Some(prompt_turns));
4426            assert_eq!(saved.turns[prompt_turns].body, "concurrent task result");
4427            assert_eq!(saved.turns[prompt_turns + 1].body, "second task result");
4428            assert!(posts_since_last_reply(&saved).is_some());
4429            let cfg = config(mock_agent(tmp.path(), ECHO, BTreeMap::new()));
4430            say(
4431                &talks.claim_turn(&id).unwrap().unwrap(),
4432                &mut saved,
4433                &talks,
4434                &cfg,
4435                "read the results",
4436                Vec::new(),
4437            )
4438            .await
4439            .expect("replay turn");
4440            let prompt = &saved.turns.last().unwrap().body;
4441            assert!(prompt.contains("# Posted by a task"), "{prompt}");
4442            assert!(!prompt.contains("already available"), "{prompt}");
4443            assert!(
4444                prompt
4445                    .find("concurrent task result")
4446                    .expect("first post in prompt")
4447                    < prompt
4448                        .find("second task result")
4449                        .expect("second post in prompt")
4450            );
4451            let mut saved = talks.get(&id).expect("reload after replay");
4452            assert!(posts_since_last_reply(&saved).is_none());
4453            say(
4454                &talks.claim_turn(&id).unwrap().unwrap(),
4455                &mut saved,
4456                &talks,
4457                &cfg,
4458                "continue",
4459                Vec::new(),
4460            )
4461            .await
4462            .expect("next turn");
4463            assert!(!saved.turns.last().unwrap().body.contains("task result"));
4464        }
4465    }
4466
4467    /// A mock whose script counts its own calls in `<dir>/<id>.calls` before
4468    /// running `body`.
4469    fn counting_agent(dir: &Path, id: &str, body: &str) -> AgentSpec {
4470        let calls = dir.join(format!("{id}.calls"));
4471        let script = format!(
4472            "#!/bin/sh\necho x >> '{}'\n{body}\n",
4473            calls.to_string_lossy()
4474        );
4475        let path = dir.join(format!("mock-{id}.sh"));
4476        std::fs::write(&path, script).expect("write mock");
4477        AgentSpec {
4478            id: id.to_owned(),
4479            kind: AgentKind::Command,
4480            model: None,
4481            command: vec!["sh".to_owned(), path.to_string_lossy().into_owned()],
4482            extra_args: Vec::new(),
4483            env: BTreeMap::new(),
4484            prompt_delivery: None,
4485        }
4486    }
4487
4488    fn calls(dir: &Path, id: &str) -> usize {
4489        std::fs::read_to_string(dir.join(format!("{id}.calls"))).map_or(0, |s| s.lines().count())
4490    }
4491
4492    fn chain_config(specs: Vec<AgentSpec>, ids: &[&str]) -> Config {
4493        let mut cfg = config(specs[0].clone());
4494        cfg.agents = specs;
4495        cfg.roles.chatter = Some(AgentChoice::Chain(
4496            ids.iter().map(|s| (*s).to_owned()).collect(),
4497        ));
4498        cfg
4499    }
4500
4501    #[tokio::test]
4502    async fn completion_evidence_reaches_fresh_resumed_and_fallback_prompts() {
4503        for fallback in [false, true] {
4504            let (tmp, talks) = store();
4505            let cfg = if fallback {
4506                chain_config(
4507                    vec![
4508                        counting_agent(tmp.path(), "a", "cat >/dev/null\nexit 3"),
4509                        counting_agent(tmp.path(), "b", "cat"),
4510                    ],
4511                    &["a", "b"],
4512                )
4513            } else {
4514                config(mock_agent(tmp.path(), ECHO, BTreeMap::new()))
4515            };
4516            let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).unwrap();
4517            let lease = talks.claim_turn(&talk.id).unwrap().unwrap();
4518            for n in 0..2 {
4519                let text = format!(
4520                    "# {}\n\n- Task: `unavailable-{n}`",
4521                    crate::prompt::CHAT_TASK_NOTICE_HEADING
4522                );
4523                let reference = crate::chat_notify::CompletionReference {
4524                    task: format!("unavailable-{n}"),
4525                    run: None,
4526                    notice: text.clone(),
4527                };
4528                let key = format!("unavailable-{n}:done");
4529                assert!(
4530                    queue_once_matching(
4531                        &mut talk,
4532                        &talks,
4533                        &text,
4534                        &key,
4535                        &|_| false,
4536                        Some(&reference)
4537                    )
4538                    .unwrap()
4539                );
4540                assert!(
4541                    !queue_once_matching(
4542                        &mut talk,
4543                        &talks,
4544                        &text,
4545                        &key,
4546                        &|_| false,
4547                        Some(&reference)
4548                    )
4549                    .unwrap()
4550                );
4551                talk = talks.get(&talk.id).unwrap();
4552                assert_eq!(talk.completion_notices.len(), n + 1);
4553                let text = drain(&mut talk, &talks).unwrap().unwrap();
4554                respond(&lease, &mut talk, &talks, &cfg, &text)
4555                    .await
4556                    .unwrap();
4557                let reply = &talk.turns.last().unwrap().body;
4558                assert!(reply.contains("# Verified outcome evidence"), "{reply}");
4559                assert!(reply.contains("Outcome evidence unavailable"));
4560                assert!(reply.contains("Do not file new tasks or edit repositories solely"));
4561                assert!(reply.contains("Explicitly say when no action remains"));
4562            }
4563            record(&mut talk, &talks, "ordinary follow-up", vec![]).unwrap();
4564            assert!(
4565                crate::chat_notify::response_evidence(&talk)
4566                    .await
4567                    .is_empty(),
4568                "history does not trigger lookups"
4569            );
4570            record(
4571                &mut talk,
4572                &talks,
4573                &format!("# {}\n\nlegacy", crate::prompt::CHAT_TASK_NOTICE_HEADING),
4574                vec![],
4575            )
4576            .unwrap();
4577            assert!(
4578                crate::chat_notify::response_evidence(&talk)
4579                    .await
4580                    .contains("Legacy notification has no structured task/run reference")
4581            );
4582        }
4583    }
4584
4585    #[tokio::test]
4586    async fn a_chatter_chain_falls_back_resends_the_transcript_and_sticks() {
4587        let (tmp, talks) = store();
4588        let a = counting_agent(tmp.path(), "a", "cat >/dev/null\nexit 3");
4589        let b = counting_agent(tmp.path(), "b", "cat");
4590        let cfg = chain_config(vec![a, b], &["a", "b"]);
4591        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4592        assert_eq!(talk.agent, "a");
4593        post(&talks, &talk.id, "fallback task result", "task").expect("post");
4594        talk = talks.get(&talk.id).expect("reload");
4595
4596        say(
4597            &talks.claim_turn(&talk.id).unwrap().unwrap(),
4598            &mut talk,
4599            &talks,
4600            &cfg,
4601            "hello there",
4602            Vec::new(),
4603        )
4604        .await
4605        .expect("turn");
4606        assert_eq!(calls(tmp.path(), "a"), 1, "each id is tried once");
4607        assert_eq!(calls(tmp.path(), "b"), 1);
4608        assert_eq!(talk.agent, "b", "the switch persists");
4609        assert!(talks.get(&talk.id).unwrap().agent == "b");
4610        let reply = talk.turns.last().unwrap();
4611        assert!(reply.body.contains("hello there"));
4612        assert!(reply.body.contains("fallback task result"));
4613        let saved = talks.get(&talk.id).expect("reload fallback");
4614        assert_eq!(saved.posts_read_through, Some(2));
4615        assert!(posts_since_last_reply(&saved).is_none());
4616        assert!(
4617            reply.body.contains("magi task add --solo"),
4618            "a fresh seat gets the full briefing"
4619        );
4620        assert!(
4621            talk.turns
4622                .iter()
4623                .any(|t| t.body.contains("agent changed from a to b")),
4624            "the switch is noted"
4625        );
4626    }
4627
4628    #[tokio::test]
4629    async fn an_exhausted_chatter_chain_fails_like_a_single_seat_and_stays_put() {
4630        let (tmp, talks) = store();
4631        let a = counting_agent(tmp.path(), "a", "cat >/dev/null\nexit 3");
4632        let b = counting_agent(tmp.path(), "b", "cat >/dev/null\nexit 4");
4633        let cfg = chain_config(vec![a, b], &["a", "b", "a"]);
4634        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4635
4636        let err = say(
4637            &talks.claim_turn(&talk.id).unwrap().unwrap(),
4638            &mut talk,
4639            &talks,
4640            &cfg,
4641            "hi",
4642            Vec::new(),
4643        )
4644        .await
4645        .expect_err("every agent failed");
4646        assert!(err.to_string().contains("`a`"), "{err:#}");
4647        assert_eq!(calls(tmp.path(), "a"), 1);
4648        assert_eq!(calls(tmp.path(), "b"), 1);
4649        assert_eq!(talk.agent, "a", "an exhausted chain leaves the agent alone");
4650    }
4651
4652    #[test]
4653    fn a_chatter_chain_skips_an_unknown_id_at_begin() {
4654        let (tmp, talks) = store();
4655        let b = counting_agent(tmp.path(), "b", "cat");
4656        let cfg = chain_config(vec![b], &["ghost", "b"]);
4657        let talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4658        assert_eq!(talk.agent, "b");
4659    }
4660
4661    #[tokio::test]
4662    async fn an_explicit_agent_inside_the_chatter_chain_stays_pinned() {
4663        let (tmp, talks) = store();
4664        let a = counting_agent(tmp.path(), "a", "cat >/dev/null\nexit 3");
4665        let b = counting_agent(tmp.path(), "b", "cat");
4666        let cfg = chain_config(vec![a, b], &["a", "b"]);
4667        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), Some("a")).expect("begin");
4668        say(
4669            &talks.claim_turn(&talk.id).unwrap().unwrap(),
4670            &mut talk,
4671            &talks,
4672            &cfg,
4673            "hi",
4674            Vec::new(),
4675        )
4676        .await
4677        .expect_err("a alone, and it fails");
4678        assert_eq!(calls(tmp.path(), "b"), 0);
4679        assert_eq!(talk.agent, "a");
4680    }
4681
4682    #[tokio::test]
4683    async fn an_explicit_agent_does_not_borrow_the_chatter_chain() {
4684        let (tmp, talks) = store();
4685        let a = counting_agent(tmp.path(), "a", "cat >/dev/null\nexit 3");
4686        let b = counting_agent(tmp.path(), "b", "cat");
4687        let c = counting_agent(tmp.path(), "c", "cat >/dev/null\nexit 3");
4688        let cfg = chain_config(vec![a, b, c], &["a", "b"]);
4689        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), Some("c")).expect("begin");
4690        say(
4691            &talks.claim_turn(&talk.id).unwrap().unwrap(),
4692            &mut talk,
4693            &talks,
4694            &cfg,
4695            "hi",
4696            Vec::new(),
4697        )
4698        .await
4699        .expect_err("c alone, and it fails");
4700        assert_eq!(calls(tmp.path(), "b"), 0);
4701    }
4702
4703    #[test]
4704    fn briefing_names_the_operator_with_or_without_a_persona() {
4705        let plain = briefing(Path::new("/repo"), "en", false);
4706        let named = briefing_with(Path::new("/repo"), "en", false, None, Some("Commander"));
4707        assert!(named.starts_with(&plain));
4708        assert!(named.contains("# Addressing the operator"));
4709        assert!(named.contains("\"Commander\""));
4710        let rei = crate::persona::builtin_catalog()
4711            .into_iter()
4712            .find(|p| p.id == "rei")
4713            .expect("rei");
4714        let with = briefing_with(
4715            Path::new("/repo"),
4716            "en",
4717            false,
4718            Some(&rei),
4719            Some("Commander"),
4720        );
4721        assert!(with.contains("\"Commander\""));
4722        assert!(!with.contains("# Addressing the operator"));
4723    }
4724
4725    #[test]
4726    fn briefing_carries_a_persona_section_only_when_one_is_chosen() {
4727        let plain = briefing(Path::new("/repo"), "en", false);
4728        assert_eq!(
4729            plain,
4730            briefing_with(Path::new("/repo"), "en", false, None, None)
4731        );
4732        assert!(!plain.contains("Persona"));
4733        let rei = crate::persona::builtin_catalog()
4734            .into_iter()
4735            .find(|p| p.id == "rei")
4736            .expect("rei");
4737        let with = briefing_with(Path::new("/repo"), "en", false, Some(&rei), None);
4738        assert!(with.starts_with(&plain), "the plain briefing is untouched");
4739        assert!(with.contains("# Persona (tone only)"));
4740        assert!(with.contains("TONE ONLY"));
4741        assert!(with.contains("task ids"));
4742        assert!(with.contains("write policy"));
4743        assert!(with.contains("`magi task add`"));
4744        assert!(with.contains("Rei Ayanami"));
4745    }
4746
4747    #[test]
4748    fn a_talk_written_before_personas_still_loads() {
4749        let (tmp, talks) = store();
4750        let spec = mock_agent(tmp.path(), ECHO, BTreeMap::new());
4751        let cfg = config(spec);
4752        let talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4753        let path = talks.root.join(format!("{}.json", talk.id));
4754        let mut v: serde_json::Value =
4755            serde_json::from_str(&std::fs::read_to_string(&path).unwrap()).unwrap();
4756        v.as_object_mut().unwrap().remove("persona");
4757        v.as_object_mut().unwrap().remove("persona_dirty");
4758        std::fs::write(&path, v.to_string()).unwrap();
4759        let loaded = talks.get(&talk.id).expect("old record loads");
4760        assert_eq!(loaded.persona, "");
4761        assert!(!loaded.persona_dirty);
4762    }
4763
4764    #[test]
4765    fn the_briefing_files_with_solo_or_the_chosen_implementer_count() {
4766        let repo = Path::new("/repo");
4767        let solo = briefing_for(repo, "", false, None, None, 1);
4768        assert_eq!(solo, briefing(repo, "", false));
4769        assert!(solo.contains("magi task add --solo --repo"));
4770        assert!(solo.contains("magi task add --solo --attach"));
4771        for n in [2u8, 3] {
4772            let b = briefing_for(repo, "", false, None, None, n);
4773            assert!(
4774                b.contains(&format!("magi task add --implementers {n} --repo")),
4775                "{b}"
4776            );
4777            assert!(b.contains(&format!("magi task add --implementers {n} --attach")));
4778            assert!(!b.contains("task add --solo"), "{b}");
4779            assert!(!b.contains("Use --solo"), "{b}");
4780        }
4781    }
4782
4783    #[test]
4784    fn begin_with_applies_remembered_choices_and_drops_unfit_ones_alone() {
4785        let (tmp, talks) = store();
4786        let first = mock_agent(tmp.path(), ECHO, BTreeMap::new());
4787        let mut second = first.clone();
4788        second.id = "second".to_owned();
4789        let mut cfg = config(first.clone());
4790        cfg.agents.push(second.clone());
4791        let open = |p: &Preferred| {
4792            begin_with(&talks, &cfg, tmp.path().to_owned(), None, p).expect("begin")
4793        };
4794
4795        let talk = open(&Preferred {
4796            agent: Some("second".into()),
4797            persona: Some("rei".into()),
4798            implementers: Some(i64::from(check_implementers(3, &cfg).unwrap_or(1))),
4799        });
4800        assert_eq!(talk.agent, "second");
4801        assert!(
4802            !talk.fallback,
4803            "a remembered non-head agent is an explicit choice"
4804        );
4805        assert_eq!(talk.persona, "rei");
4806        assert_eq!(talk.implementers, check_implementers(3, &cfg).unwrap_or(1));
4807        assert!(!talk.persona_dirty && !talk.implementers_dirty);
4808
4809        // The chain head chosen again stays a fallback-capable talk.
4810        let head = open(&Preferred {
4811            agent: Some(first.id.clone()),
4812            ..Preferred::default()
4813        });
4814        assert!(head.fallback);
4815
4816        // Each unfit field falls back by itself.
4817        for implementers in [0, 4, -1, 1000] {
4818            let t = open(&Preferred {
4819                agent: Some("gone".into()),
4820                persona: Some("no-such-persona".into()),
4821                implementers: Some(implementers),
4822            });
4823            assert_eq!(t.agent, first.id);
4824            assert!(t.fallback);
4825            assert_eq!(t.persona, "");
4826            assert_eq!(t.implementers, 1);
4827        }
4828
4829        // `default` stays the empty string; a good field survives a bad one.
4830        let t = open(&Preferred {
4831            agent: Some("second".into()),
4832            persona: Some("default".into()),
4833            implementers: Some(9),
4834        });
4835        assert_eq!(
4836            (t.agent.as_str(), t.persona.as_str(), t.implementers),
4837            ("second", "", 1)
4838        );
4839
4840        // An explicit agent still wins and keeps its strict error.
4841        let strict = begin_with(
4842            &talks,
4843            &cfg,
4844            tmp.path().to_owned(),
4845            Some("first-unknown"),
4846            &Preferred {
4847                agent: Some("second".into()),
4848                ..Preferred::default()
4849            },
4850        );
4851        assert!(strict.is_err());
4852    }
4853
4854    #[test]
4855    fn remembered_agent_can_open_when_the_chatter_chain_cannot_resolve() {
4856        let (tmp, talks) = store();
4857        let spec = mock_agent(tmp.path(), ECHO, BTreeMap::new());
4858        // A missing chain member makes head resolution fail independently of
4859        // which subscription CLIs happen to be installed on this machine.
4860        let cfg = chain_config(vec![spec.clone()], &["missing-chatter"]);
4861        assert!(begin(&talks, &cfg, tmp.path().to_owned(), None).is_err());
4862        let talk = begin_with(
4863            &talks,
4864            &cfg,
4865            tmp.path().to_owned(),
4866            None,
4867            &Preferred {
4868                agent: Some(spec.id.clone()),
4869                ..Preferred::default()
4870            },
4871        )
4872        .expect("a valid remembered agent does not require a default chain");
4873        assert_eq!(talk.agent, spec.id);
4874        assert!(!talk.fallback);
4875        assert_eq!(talks.get(&talk.id).unwrap().agent, spec.id);
4876        assert!(
4877            begin_with(
4878                &talks,
4879                &cfg,
4880                tmp.path().to_owned(),
4881                None,
4882                &Preferred {
4883                    agent: Some("missing-preference".into()),
4884                    ..Preferred::default()
4885                },
4886            )
4887            .is_err()
4888        );
4889    }
4890
4891    #[test]
4892    fn implementers_are_validated_and_old_records_load_as_solo() {
4893        let (tmp, talks) = store();
4894        let cfg = config(mock_agent(tmp.path(), ECHO, BTreeMap::new()));
4895        assert_eq!(check_implementers(1, &cfg), Ok(1));
4896        assert_eq!(check_implementers(3, &cfg), Ok(3));
4897        assert!(check_implementers(0, &cfg).is_err());
4898        assert!(check_implementers(4, &cfg).is_err());
4899        let mut empty = cfg.clone();
4900        empty.agents.clear();
4901        assert!(check_implementers(2, &empty).is_err());
4902
4903        let talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4904        let path = talks.path_of(&talk.id);
4905        let mut v: serde_json::Value =
4906            serde_json::from_str(&std::fs::read_to_string(&path).unwrap()).unwrap();
4907        v.as_object_mut().unwrap().remove("implementers");
4908        v.as_object_mut().unwrap().remove("implementers_dirty");
4909        std::fs::write(&path, v.to_string()).unwrap();
4910        let loaded = talks.get(&talk.id).expect("load");
4911        assert_eq!(loaded.implementers, 1);
4912        assert!(!loaded.implementers_dirty);
4913    }
4914
4915    #[tokio::test]
4916    async fn an_implementers_switch_is_noted_once_and_kept_across_a_failed_turn() {
4917        let (tmp, talks) = store();
4918        let spec = mock_agent(tmp.path(), ECHO, BTreeMap::new());
4919        let cfg = config(spec);
4920        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4921        let lease = talks.claim_turn(&talk.id).unwrap().unwrap();
4922        say(&lease, &mut talk, &talks, &cfg, "hello", Vec::new())
4923            .await
4924            .expect("first turn");
4925
4926        assert!(switch_implementers(&mut talk, &talks, 2).expect("switch"));
4927        assert!(!switch_implementers(&mut talk, &talks, 2).expect("same"));
4928        assert!(talk.implementers_dirty);
4929        say(&lease, &mut talk, &talks, &cfg, "next", Vec::new())
4930            .await
4931            .expect("turn");
4932        let prompt = &talk.turns.last().unwrap().body;
4933        assert!(prompt.contains("# Task filing update"), "{prompt}");
4934        assert!(prompt.contains("--implementers 2"));
4935        assert!(!talk.implementers_dirty, "cleared after a successful turn");
4936
4937        say(&lease, &mut talk, &talks, &cfg, "again", Vec::new())
4938            .await
4939            .expect("turn");
4940        assert!(
4941            !talk
4942                .turns
4943                .last()
4944                .unwrap()
4945                .body
4946                .contains("# Task filing update")
4947        );
4948
4949        // Without a resumable session the current policy rides on every turn.
4950        let mut no_sessions = cfg.clone();
4951        no_sessions.graph.sessions = false;
4952        say(&lease, &mut talk, &talks, &no_sessions, "one", Vec::new())
4953            .await
4954            .expect("turn");
4955        say(&lease, &mut talk, &talks, &no_sessions, "two", Vec::new())
4956            .await
4957            .expect("turn");
4958        assert!(talk.turns.last().unwrap().body.contains("--implementers 2"));
4959    }
4960
4961    #[tokio::test]
4962    async fn a_persona_switch_notes_marks_dirty_and_updates_the_next_turn_once() {
4963        let (tmp, talks) = store();
4964        let spec = mock_agent(tmp.path(), ECHO, BTreeMap::new());
4965        let cfg = config(spec);
4966        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
4967        let lease = talks.claim_turn(&talk.id).unwrap().unwrap();
4968        say(&lease, &mut talk, &talks, &cfg, "hello", Vec::new())
4969            .await
4970            .expect("first turn");
4971        assert!(!talk.turns[1].body.contains("Persona"));
4972
4973        assert!(switch_persona(&mut talk, &talks, "misato").expect("switch"));
4974        assert!(!switch_persona(&mut talk, &talks, "misato").expect("same"));
4975        assert!(talk.persona_dirty);
4976        assert!(talk.turns.last().unwrap().body.contains("persona changed"));
4977
4978        say(&lease, &mut talk, &talks, &cfg, "next", Vec::new())
4979            .await
4980            .expect("turn");
4981        let prompt = &talk.turns.last().unwrap().body;
4982        assert!(prompt.contains("# Persona update"), "{prompt}");
4983        assert!(prompt.contains("Misato Katsuragi"));
4984        assert!(!talk.persona_dirty, "cleared after a successful turn");
4985
4986        say(&lease, &mut talk, &talks, &cfg, "again", Vec::new())
4987            .await
4988            .expect("turn");
4989        assert!(!talk.turns.last().unwrap().body.contains("# Persona update"));
4990
4991        assert!(switch_persona(&mut talk, &talks, "default").expect("back"));
4992        assert_eq!(talk.persona, "");
4993        say(&lease, &mut talk, &talks, &cfg, "plain", Vec::new())
4994            .await
4995            .expect("turn");
4996        assert!(
4997            talk.turns
4998                .last()
4999                .unwrap()
5000                .body
5001                .contains("turned the persona off")
5002        );
5003
5004        // Without a resumable session every turn re-sends the persona.
5005        assert!(switch_persona(&mut talk, &talks, "rei").expect("rei"));
5006        let mut no_sessions = cfg.clone();
5007        no_sessions.graph.sessions = false;
5008        say(&lease, &mut talk, &talks, &no_sessions, "one", Vec::new())
5009            .await
5010            .expect("turn");
5011        say(&lease, &mut talk, &talks, &no_sessions, "two", Vec::new())
5012            .await
5013            .expect("turn");
5014        let last = &talk.turns.last().unwrap().body;
5015        assert!(!talk.persona_dirty);
5016        assert!(last.contains("# Persona (tone only)"), "{last}");
5017        assert!(last.contains("Rei Ayanami"));
5018    }
5019
5020    #[tokio::test]
5021    async fn the_first_turn_carries_the_briefing_and_later_turns_do_not() {
5022        let (tmp, talks) = store();
5023        let spec = mock_agent(tmp.path(), ECHO, BTreeMap::new());
5024        let cfg = config(spec);
5025        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5026
5027        say(
5028            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5029            &mut talk,
5030            &talks,
5031            &cfg,
5032            "what does the queue module do?",
5033            Vec::new(),
5034        )
5035        .await
5036        .expect("first turn");
5037        let first_prompt = &talk.turns[1].body;
5038        assert!(first_prompt.contains("magi task add --solo"));
5039        assert!(first_prompt.contains("what does the queue module do?"));
5040
5041        say(
5042            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5043            &mut talk,
5044            &talks,
5045            &cfg,
5046            "and how is it locked?",
5047            Vec::new(),
5048        )
5049        .await
5050        .expect("second turn");
5051        let second_prompt = &talk.turns[3].body;
5052        assert!(
5053            !second_prompt.contains("magi task add --solo"),
5054            "the briefing is sent once, not on every turn: {second_prompt}"
5055        );
5056        assert!(second_prompt.contains("and how is it locked?"));
5057    }
5058
5059    #[tokio::test]
5060    async fn switching_agent_resets_the_seat_notes_it_and_resends_the_transcript() {
5061        let (tmp, talks) = store();
5062        let a = mock_agent(tmp.path(), ECHO, BTreeMap::new());
5063        let mut b = a.clone();
5064        b.id = "other".to_owned();
5065        let mut cfg = config(a.clone());
5066        cfg.agents.push(b.clone());
5067        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), Some(&a.id)).expect("begin");
5068        say(
5069            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5070            &mut talk,
5071            &talks,
5072            &cfg,
5073            "remember the walrus",
5074            Vec::new(),
5075        )
5076        .await
5077        .expect("first turn");
5078        let old_session = talk.seat.claude_session.clone();
5079        assert_eq!(talk.seat.turns, 1);
5080
5081        assert!(switch_agent(&mut talk, &talks, &b).expect("switch"));
5082        assert_eq!(talk.agent, "other");
5083        assert_eq!(talk.seat.turns, 0);
5084        assert_eq!(talk.seat.agent, "other");
5085        assert_ne!(talk.seat.claude_session, old_session);
5086        let note = talk.turns.last().expect("note");
5087        assert_eq!(note.who, Who::Agent);
5088        assert!(note.body.starts_with(MAGI_NOTE), "{}", note.body);
5089        assert!(note.body.contains("changed from"), "{}", note.body);
5090        assert_eq!(talks.get(&talk.id).expect("reload").agent, "other");
5091
5092        let before = talk.turns.len();
5093        assert!(!switch_agent(&mut talk, &talks, &b).expect("same agent"));
5094        assert_eq!(talk.turns.len(), before, "a no-op writes no note");
5095
5096        say(
5097            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5098            &mut talk,
5099            &talks,
5100            &cfg,
5101            "what did I say?",
5102            Vec::new(),
5103        )
5104        .await
5105        .expect("turn after switch");
5106        let prompt = &talk.turns.last().expect("reply").body;
5107        assert!(prompt.contains("remember the walrus"), "{prompt}");
5108        assert!(prompt.contains("## magi"), "{prompt}");
5109        assert!(prompt.contains("what did I say?"), "{prompt}");
5110    }
5111
5112    #[tokio::test]
5113    async fn say_appends_the_operator_turn_then_the_agent_turn() {
5114        let (tmp, talks) = store();
5115        let spec = mock_agent(tmp.path(), REPLY, env("go ahead"));
5116        let cfg = config(spec);
5117        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5118
5119        say(
5120            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5121            &mut talk,
5122            &talks,
5123            &cfg,
5124            "can I rename this function?",
5125            Vec::new(),
5126        )
5127        .await
5128        .expect("say");
5129
5130        assert_eq!(talk.turns.len(), 2);
5131        assert_eq!(talk.turns[0].who, Who::Operator);
5132        assert_eq!(talk.turns[0].body, "can I rename this function?");
5133        assert_eq!(talk.turns[1].who, Who::Agent);
5134        assert_eq!(talk.turns[1].body, "go ahead");
5135        assert_eq!(talks.get(&talk.id).expect("get").turns, talk.turns);
5136    }
5137
5138    #[tokio::test]
5139    async fn a_failed_turn_keeps_the_operator_message_and_says_what_happened() {
5140        let (tmp, talks) = store();
5141        let spec = mock_agent(tmp.path(), BROKEN, BTreeMap::new());
5142        let cfg = config(spec);
5143        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5144
5145        talk.posts_read_through = None;
5146        talks.put(&mut talk).expect("save old-format boundary");
5147        post(&talks, &talk.id, "unread task result", "task").expect("post");
5148        talk = talks.get(&talk.id).expect("reload");
5149
5150        let err = say(
5151            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5152            &mut talk,
5153            &talks,
5154            &cfg,
5155            "check the tests",
5156            Vec::new(),
5157        )
5158        .await
5159        .expect_err("a turn with no answer is an error");
5160        assert!(err.to_string().contains("no answer"), "{err}");
5161
5162        let on_disk = talks.get(&talk.id).expect("get");
5163        assert_eq!(on_disk.turns.len(), 3);
5164        assert_eq!(on_disk.posts_read_through, Some(0));
5165        assert!(
5166            posts_since_last_reply(&on_disk)
5167                .unwrap()
5168                .contains("unread task result")
5169        );
5170        assert_eq!(on_disk.turns[1].body, "check the tests");
5171        let note = &on_disk.turns[2];
5172        assert_eq!(note.who, Who::Agent);
5173        assert!(note.body.starts_with(MAGI_NOTE), "{}", note.body);
5174        assert!(note.body.contains("your message is saved"));
5175    }
5176
5177    /// The failure this stands in for: a reader elsewhere briefly has the
5178    /// talk file open right when `turn` tries to save the reply, and the
5179    /// write-then-rename fails once or twice before the reader lets go.
5180    /// `write_atomic`'s own retries must absorb that with nobody the wiser -
5181    /// no gap in the transcript, no dropped turn.
5182    #[tokio::test]
5183    async fn a_passing_write_failure_while_saving_the_reply_does_not_lose_it() {
5184        let (tmp, talks) = store();
5185        let spec = mock_agent(tmp.path(), REPLY, env("go ahead"));
5186        let cfg = config(spec);
5187        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5188
5189        let text =
5190            record(&mut talk, &talks, "can I rename this function?", Vec::new()).expect("record");
5191        // One fewer failure than `write_atomic` will retry through, so the
5192        // very last attempt must succeed.
5193        failpoint::force_put_failures(PUT_RETRIES - 1);
5194        respond(
5195            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5196            &mut talk,
5197            &talks,
5198            &cfg,
5199            &text,
5200        )
5201        .await
5202        .expect("respond must survive a write failure its own retries can outlast");
5203
5204        assert_eq!(talk.turns.len(), 2);
5205        assert_eq!(talk.turns[1].who, Who::Agent);
5206        assert_eq!(talk.turns[1].body, "go ahead");
5207        let on_disk = talks.get(&talk.id).expect("get");
5208        assert_eq!(
5209            on_disk.turns, talk.turns,
5210            "the reply must reach disk despite the early write failures"
5211        );
5212    }
5213
5214    /// When the write-then-rename never recovers - standing in for a disk
5215    /// that stays unwritable rather than a reader that eventually lets go -
5216    /// the reply must not disappear without a trace the way it did in the
5217    /// real incident this repository saw: no error on the phone, no note in
5218    /// the transcript, and the turn simply gone from `talks/<id>.json`.
5219    #[tokio::test]
5220    async fn a_persistent_write_failure_while_saving_the_reply_is_never_silent() {
5221        let (tmp, talks) = store();
5222        let spec = mock_agent(tmp.path(), REPLY, env("go ahead"));
5223        let cfg = config(spec);
5224        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5225
5226        let text = record(&mut talk, &talks, "check the tests", Vec::new()).expect("record");
5227        // Exactly enough forced failures to exhaust the reply's own retries;
5228        // the shorter note that replaces it then saves cleanly, which is the
5229        // common case this exercises - a large write racing something,
5230        // followed by a small one that does not.
5231        failpoint::force_put_failures(PUT_RETRIES);
5232        let err = respond(
5233            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5234            &mut talk,
5235            &talks,
5236            &cfg,
5237            &text,
5238        )
5239        .await
5240        .expect_err("a reply that cannot be saved must be reported, not swallowed");
5241        assert!(err.to_string().contains("could not be saved"), "{err}");
5242
5243        let on_disk = talks.get(&talk.id).expect("get");
5244        assert_eq!(
5245            on_disk.turns.len(),
5246            2,
5247            "the operator turn plus a visible note"
5248        );
5249        assert_eq!(on_disk.turns[0].body, "check the tests");
5250        let note = &on_disk.turns[1];
5251        assert_eq!(note.who, Who::Agent);
5252        assert!(note.body.starts_with(MAGI_NOTE), "{}", note.body);
5253        assert!(
5254            note.body.contains("could not be saved"),
5255            "the operator must be told the reply is missing, not left staring \
5256             at a gap with no explanation: {}",
5257            note.body
5258        );
5259        assert_eq!(
5260            talk.turns, on_disk.turns,
5261            "the in-memory talk must match what actually landed on disk"
5262        );
5263
5264        // The generated answer itself must still be recoverable, not merely
5265        // reported as lost.
5266        let artifacts = talks.artifacts_of(&talk.id);
5267        let stash = std::fs::read_dir(&artifacts)
5268            .expect("artifacts dir")
5269            .filter_map(|e| e.ok())
5270            .find(|e| e.file_name().to_string_lossy().ends_with("-lost.txt"))
5271            .expect("a stash file for the lost reply");
5272        let stashed = std::fs::read_to_string(stash.path()).expect("read stash");
5273        assert_eq!(stashed, "go ahead");
5274
5275        // Losing the reply must not also lose the seat. The CLI took a turn
5276        // and consumed this seat's session id; if the note's write left the
5277        // record claiming otherwise, the next turn would re-open a session
5278        // the CLI is already holding - the `20260907-011805-fb57` desync -
5279        // and would re-send the whole briefing besides. Both decisions read
5280        // the seat straight off disk (`agent::has_session` and `turn`'s own
5281        // `seat.turns == 0` branch), so this is the field that has to match.
5282        assert_eq!(
5283            on_disk.seat.turns, 1,
5284            "the note's write must carry the turn the CLI actually took"
5285        );
5286        assert_eq!(
5287            on_disk.seat.claude_session, talk.seat.claude_session,
5288            "the session id handed to the CLI must survive the failed reply"
5289        );
5290        assert_eq!(on_disk.seat.captured_session, talk.seat.captured_session);
5291        assert!(
5292            agent::has_session(AgentKind::Command, &on_disk.seat, cfg.graph.sessions),
5293            "the next turn must resume, not open the same session id twice"
5294        );
5295    }
5296
5297    /// Even the note can fail to save, if the disk stays unwritable for long
5298    /// enough. `respond` must still report the failure rather than pretend
5299    /// the turn succeeded, and must not leave the in-memory `talk` claiming
5300    /// a turn that never reached disk.
5301    #[tokio::test]
5302    async fn a_write_failure_that_also_loses_the_note_still_reports_it() {
5303        let (tmp, talks) = store();
5304        let spec = mock_agent(tmp.path(), REPLY, env("go ahead"));
5305        let cfg = config(spec);
5306        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5307
5308        let text = record(&mut talk, &talks, "check the tests", Vec::new()).expect("record");
5309        // Enough forced failures to exhaust the retries for both the reply
5310        // and the note that would have replaced it.
5311        failpoint::force_put_failures(PUT_RETRIES * 2);
5312        let err = respond(
5313            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5314            &mut talk,
5315            &talks,
5316            &cfg,
5317            &text,
5318        )
5319        .await
5320        .expect_err("neither the reply nor the note could be saved");
5321        assert!(err.to_string().contains("could not be saved"), "{err}");
5322
5323        assert_eq!(talk.turns.len(), 1, "only the operator's own turn");
5324        let on_disk = talks.get(&talk.id).expect("get");
5325        assert_eq!(on_disk.turns.len(), 1);
5326
5327        // Nothing at all reached disk, so the seat could not either: the CLI
5328        // took a turn the record does not know about. That is pinned here as
5329        // the known cost of a file that cannot be written twice over, not as
5330        // something this branch could do better - the only way to record the
5331        // seat is the write that just failed. It is also the point where
5332        // this meets `20260907-011805-fb57`: a turn taken before some later
5333        // write lands would re-open a session id the CLI already holds. The
5334        // in-memory seat keeps the truth the CLI reported, which is why it is
5335        // not wound back to match.
5336        assert_eq!(
5337            on_disk.seat.turns, 0,
5338            "an unwritable file cannot record the turn the CLI took"
5339        );
5340        assert_eq!(
5341            talk.seat.turns, 1,
5342            "the in-memory seat still reports the turn the CLI actually took"
5343        );
5344        assert_eq!(
5345            on_disk.seat.claude_session, talk.seat.claude_session,
5346            "the session id was minted at `begin` and never changes here"
5347        );
5348    }
5349
5350    /// An attachment lets the operator send an otherwise-empty message, and
5351    /// its absolute path is what actually reaches the agent's prompt - here
5352    /// on the very first turn, where it has to share the briefing.
5353    #[tokio::test]
5354    async fn attachments_reach_the_prompt_and_an_empty_body_is_still_a_turn() {
5355        let (tmp, talks) = store();
5356        let spec = mock_agent(tmp.path(), ECHO, BTreeMap::new());
5357        let cfg = config(spec);
5358        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5359
5360        let att = talks
5361            .put_attachment(
5362                &talk.id,
5363                "image/png",
5364                "screenshot.png",
5365                b"pretend-png-bytes",
5366            )
5367            .expect("put attachment");
5368
5369        say(
5370            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5371            &mut talk,
5372            &talks,
5373            &cfg,
5374            "",
5375            vec![att.clone()],
5376        )
5377        .await
5378        .expect("an empty body with an attachment is still a turn");
5379
5380        let operator_turn = &talk.turns[0];
5381        assert_eq!(operator_turn.who, Who::Operator);
5382        assert_eq!(operator_turn.body, "");
5383        assert_eq!(operator_turn.attachments, vec![att.clone()]);
5384
5385        let prompt = &talk.turns[1].body;
5386        let expected_path = talks
5387            .attachments_dir(&talk.id)
5388            .join(format!("{}.png", att.id));
5389        assert!(
5390            prompt.contains(&expected_path.display().to_string()),
5391            "the agent must be told the attachment's absolute path: {prompt}"
5392        );
5393        assert!(prompt.contains("image/png"), "and its mime: {prompt}");
5394    }
5395
5396    /// See `chat`'s test of the same name: `run::home()` returns a bare
5397    /// relative `PathBuf` verbatim when `MAGI_HOME` is set to a relative
5398    /// path, so a `Talks` store built on it has a relative `root` too. That
5399    /// is fine for this store's own I/O, which runs in this process against
5400    /// this process's cwd, but `attachment_path` hands its result to a
5401    /// *different* process invoked with `cwd: &talk.repo` - an uncorrected
5402    /// relative path would resolve against the repository instead of
5403    /// wherever the attachment actually landed.
5404    #[test]
5405    fn attachment_path_is_absolute_even_when_the_store_root_is_relative() {
5406        let talks = Talks::at(PathBuf::from("relative-talks-root-for-this-test"));
5407        let att = Attachment {
5408            id: "0".repeat(32),
5409            name: "shot.png".to_owned(),
5410            mime: "image/png".to_owned(),
5411            bytes: 3,
5412        };
5413        let path = talks
5414            .attachment_path("some-talk-id", &att)
5415            .expect("a supported mime always yields a path");
5416        assert!(
5417            path.is_absolute(),
5418            "must be absolute even off a relative store root: {}",
5419            path.display()
5420        );
5421    }
5422
5423    #[tokio::test]
5424    async fn a_turn_past_the_configured_talk_timeout_is_reported_with_that_timeout() {
5425        // `[graph] timeout_talk` must be the number this module actually
5426        // waits, not a leftover hardcoded fifteen minutes - so the mock
5427        // sleeps past a deliberately tiny override and the failure note is
5428        // checked against that same override, not the old default.
5429        let (tmp, talks) = store();
5430        let slow = mock_agent(
5431            tmp.path(),
5432            "#!/bin/sh\ncat >/dev/null\nsleep 2\n",
5433            BTreeMap::new(),
5434        );
5435        let mut cfg = config(slow);
5436        cfg.graph.timeout_talk = 1;
5437        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5438
5439        let err = say(
5440            &talks.claim_turn(&talk.id).unwrap().unwrap(),
5441            &mut talk,
5442            &talks,
5443            &cfg,
5444            "check the tests",
5445            Vec::new(),
5446        )
5447        .await
5448        .expect_err("a turn that never answers is an error");
5449        assert!(
5450            err.to_string().contains("did not answer within 1s"),
5451            "{err}"
5452        );
5453
5454        let on_disk = talks.get(&talk.id).expect("get");
5455        let note = on_disk.turns.last().expect("a note turn was recorded");
5456        assert!(
5457            note.body.contains("did not answer within 1s"),
5458            "the transcript must show the configured timeout: {}",
5459            note.body
5460        );
5461    }
5462
5463    #[test]
5464    fn closing_is_idempotent_and_a_closed_talk_takes_no_more_turns() {
5465        let (tmp, talks) = store();
5466        let spec = mock_agent(tmp.path(), REPLY, env("hi"));
5467        let cfg = config(spec);
5468        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5469
5470        close(&mut talk, &talks).expect("close");
5471        assert_eq!(talk.status, TalkStatus::Closed);
5472        close(&mut talk, &talks).expect("closing twice is not an error");
5473
5474        let err =
5475            record(&mut talk, &talks, "still there?", Vec::new()).expect_err("closed talks refuse");
5476        assert!(err.to_string().contains("closed"));
5477        let _ = &cfg; // config kept only to build the agent above
5478    }
5479
5480    #[tokio::test]
5481    async fn a_close_that_lands_while_a_turn_is_in_flight_is_not_undone_by_the_reply() {
5482        let (tmp, talks) = store();
5483        let spec = mock_agent(tmp.path(), REPLY, env("here you go"));
5484        let cfg = config(spec);
5485        // The in-flight turn's own handle: loaded once, the way a spawned
5486        // background task in `web::talk_say` holds one for the whole turn.
5487        let mut in_flight = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5488
5489        // The operator closes the conversation through a *different* handle
5490        // while the turn above is still running - exactly what a close typed
5491        // on the phone while an agent is mid-answer looks like.
5492        let mut closed_elsewhere = talks.get(&in_flight.id).expect("reread");
5493        close(&mut closed_elsewhere, &talks).expect("close");
5494        assert_eq!(
5495            talks.get(&in_flight.id).expect("reread").status,
5496            TalkStatus::Closed,
5497            "the close landed on disk before the turn finished"
5498        );
5499
5500        // The turn's own handle still says `open` - it was loaded before the
5501        // close - and finishing it must not resurrect the conversation the
5502        // operator already ended.
5503        assert_eq!(in_flight.status, TalkStatus::Open);
5504        respond(
5505            &talks.claim_turn(&in_flight.id).unwrap().unwrap(),
5506            &mut in_flight,
5507            &talks,
5508            &cfg,
5509            "one more question",
5510        )
5511        .await
5512        .expect("the turn itself still completes");
5513
5514        let on_disk = talks.get(&in_flight.id).expect("reread");
5515        assert_eq!(
5516            on_disk.status,
5517            TalkStatus::Closed,
5518            "a close must stick even when a turn that started before it finishes after it"
5519        );
5520        // The reply is not lost either: a turn already in flight when the
5521        // operator closed still gets its answer recorded.
5522        assert!(
5523            on_disk.turns.iter().any(|t| t.body == "here you go"),
5524            "the in-flight turn's own reply is still recorded: {:?}",
5525            on_disk.turns
5526        );
5527    }
5528
5529    #[test]
5530    fn a_close_that_lands_before_record_is_called_is_not_undone_by_it() {
5531        let (tmp, talks) = store();
5532        let spec = mock_agent(tmp.path(), REPLY, env("hi"));
5533        let cfg = config(spec);
5534        // The handle `web::talk_say` would have read before awaiting config
5535        // discovery, then carried across that await into `record`.
5536        let mut stale = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5537
5538        // The operator closes the conversation through a *different* handle
5539        // in the gap between that read and the call to `record` below.
5540        let mut closed_elsewhere = talks.get(&stale.id).expect("reread");
5541        close(&mut closed_elsewhere, &talks).expect("close");
5542        assert_eq!(
5543            talks.get(&stale.id).expect("reread").status,
5544            TalkStatus::Closed,
5545            "the close landed on disk before record was called"
5546        );
5547
5548        // The stale handle still says `open` - it was loaded before the
5549        // close - so a `record` that trusted it would append a turn and
5550        // write the conversation back open, undoing the close.
5551        assert_eq!(stale.status, TalkStatus::Open);
5552        let err = record(&mut stale, &talks, "still there?", Vec::new())
5553            .expect_err("a close that landed first must be honored, not overwritten");
5554        assert!(err.to_string().contains("closed"));
5555
5556        let on_disk = talks.get(&stale.id).expect("reread");
5557        assert_eq!(
5558            on_disk.status,
5559            TalkStatus::Closed,
5560            "record must not resurrect a conversation closed while its snapshot was stale"
5561        );
5562        assert!(
5563            on_disk.turns.is_empty(),
5564            "the rejected turn must not have been appended: {:?}",
5565            on_disk.turns
5566        );
5567        let _ = &cfg; // config kept only to build the agent above
5568    }
5569
5570    #[test]
5571    fn consultation_queue_does_not_wait_on_either_conversation_lock() {
5572        for evidence in [true, false] {
5573            let (tmp, talks) = store();
5574            let cfg = config(mock_agent(tmp.path(), REPLY, env("hi")));
5575            let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).unwrap();
5576            close(&mut talk, &talks).unwrap();
5577            let answer = evidence.then(|| talks.answer_guard().unwrap());
5578            let file = (!evidence).then(|| {
5579                TurnLock::take(&talks.root.join(".store.turn"))
5580                    .unwrap()
5581                    .unwrap()
5582            });
5583            assert!(
5584                try_queue_consult(&talks, &talk.id, "consultation", None)
5585                    .unwrap()
5586                    .is_none()
5587            );
5588            let saved = talks.get(&talk.id).unwrap();
5589            assert_eq!(saved.status, TalkStatus::Closed);
5590            assert!(saved.pending_messages.is_empty());
5591            drop(file);
5592            drop(answer);
5593            assert!(
5594                try_queue_consult(&talks, &talk.id, "consultation", None)
5595                    .unwrap()
5596                    .is_some()
5597            );
5598            let saved = talks.get(&talk.id).unwrap();
5599            assert_eq!(saved.status, TalkStatus::Open);
5600            assert_eq!(saved.pending, "consultation");
5601            assert_eq!(saved.pending_messages.len(), 1);
5602            assert_eq!(saved.pending_messages[0].sender, Some(Sender::Magi));
5603        }
5604    }
5605
5606    #[test]
5607    fn evidence_try_lock_does_not_wait_for_either_store_handle() {
5608        let (tmp, talks) = store();
5609        let guard = talks.answer_guard().unwrap();
5610        // The same handle contends on the mutex, another on the file lock.
5611        assert!(talks.try_answer_guard().unwrap().is_none());
5612        let other = Talks::at(tmp.path().join("talks"));
5613        assert!(other.try_answer_guard().unwrap().is_none());
5614        drop(guard);
5615        assert!(other.try_answer_guard().unwrap().is_some());
5616    }
5617
5618    #[test]
5619    fn owner_retraction_waits_until_the_authorized_question_is_persisted() {
5620        use crate::ask::{Answer, Question, Questions};
5621        use std::sync::mpsc;
5622
5623        let (tmp, talks) = store();
5624        let cfg = config(mock_agent(tmp.path(), REPLY, env("hi")));
5625        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).unwrap();
5626        let questions = Questions::at(tmp.path().join("questions"));
5627        let mut q = Question::new(
5628            "run".into(),
5629            crate::land::APPROVAL_NODE.into(),
5630            "land".into(),
5631            "Merge?".into(),
5632            String::new(),
5633            vec!["merge".into(), "hold".into()],
5634        );
5635        questions.put(&mut q).unwrap();
5636        let instruction = format!("{}: hold", q.id);
5637        record(&mut talk, &talks, &instruction, vec![]).unwrap();
5638        let (started_tx, started_rx) = mpsc::channel();
5639        let (finished_tx, finished_rx) = mpsc::channel();
5640        std::thread::scope(|scope| {
5641            let mut writer = None;
5642            let (saved, guard) = questions
5643                .update(&q.id, |q| {
5644                    let guard = talks.try_answer_guard()?.context("evidence is busy")?;
5645                    q.answer_authority = crate::consult::authorize_answer(
5646                        q,
5647                        &talks,
5648                        &questions,
5649                        &talk.id,
5650                        crate::queue::CHAT_NODE,
5651                        "hold",
5652                        Some(&instruction),
5653                    )?;
5654                    assert!(q.answer_authority.is_some());
5655                    let talks = talks.clone();
5656                    let mut snapshot = talk.clone();
5657                    writer = Some(scope.spawn(move || {
5658                        started_tx.send(()).unwrap();
5659                        queue(&mut snapshot, &talks, "wait", vec![]).unwrap();
5660                        finished_tx.send(()).unwrap();
5661                    }));
5662                    started_rx.recv().unwrap();
5663                    assert!(matches!(
5664                        finished_rx.recv_timeout(Duration::from_millis(100)),
5665                        Err(mpsc::RecvTimeoutError::Timeout)
5666                    ));
5667                    q.answer(Answer::Choice("hold".into()))?;
5668                    // Match the CLI: return this guard through the atomic write.
5669                    Ok(guard)
5670                })
5671                .unwrap();
5672            assert_eq!(saved.resolution().as_deref(), Some("hold"));
5673            assert_eq!(
5674                questions.get(&q.id).unwrap().resolution().as_deref(),
5675                Some("hold")
5676            );
5677            assert!(matches!(
5678                finished_rx.recv_timeout(Duration::from_millis(100)),
5679                Err(mpsc::RecvTimeoutError::Timeout)
5680            ));
5681            drop(guard);
5682            writer.unwrap().join().unwrap();
5683            finished_rx.recv().unwrap();
5684        });
5685        assert_eq!(talks.get(&talk.id).unwrap().pending, "wait");
5686    }
5687
5688    #[test]
5689    fn close_blocks_on_records_guard_rather_than_interleaving_with_it() {
5690        let (tmp, talks) = store();
5691        let spec = mock_agent(tmp.path(), REPLY, env("hi"));
5692        let cfg = config(spec);
5693        let talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5694
5695        // Hold the same guard `record`'s read-modify-write section holds for
5696        // the whole of its own read-then-write, standing in for `record`
5697        // being paused between its read and its `put`.
5698        let held = talks.guard().unwrap();
5699
5700        let talks2 = talks.clone();
5701        let id = talk.id.clone();
5702        let closing = std::thread::spawn(move || {
5703            let mut talk = talks2.get(&id).expect("get");
5704            close(&mut talk, &talks2).expect("close");
5705        });
5706
5707        std::thread::sleep(Duration::from_millis(50));
5708        assert!(
5709            !closing.is_finished(),
5710            "close must wait for the guard, not read and write while it is held - \
5711             a re-read alone narrows this window without closing it"
5712        );
5713
5714        drop(held);
5715        closing.join().expect("close thread panicked");
5716
5717        assert_eq!(
5718            talks.get(&talk.id).expect("reread").status,
5719            TalkStatus::Closed,
5720            "once the guard is free, close still lands"
5721        );
5722        let _ = &cfg; // config kept only to build the agent above
5723    }
5724
5725    #[test]
5726    fn reopening_a_closed_talk_lets_it_take_turns_again_and_reopening_twice_is_not_an_error() {
5727        let (tmp, talks) = store();
5728        let spec = mock_agent(tmp.path(), REPLY, env("hi"));
5729        let cfg = config(spec);
5730        let mut talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5731
5732        close(&mut talk, &talks).expect("close");
5733        assert_eq!(talk.status, TalkStatus::Closed);
5734
5735        reopen(&mut talk, &talks).expect("reopen");
5736        assert_eq!(talk.status, TalkStatus::Open);
5737        assert_eq!(
5738            talks.get(&talk.id).expect("reread").status,
5739            TalkStatus::Open
5740        );
5741
5742        // Idempotent: reopening an already-open talk is not an error.
5743        reopen(&mut talk, &talks).expect("reopening an open talk is not an error");
5744        assert_eq!(talk.status, TalkStatus::Open);
5745
5746        record(&mut talk, &talks, "one more thing", Vec::new())
5747            .expect("a reopened talk takes turns again");
5748        let _ = &cfg; // config kept only to build the agent above
5749    }
5750
5751    #[test]
5752    fn removing_a_talk_deletes_its_record_and_artifacts_and_refuses_an_unknown_id() {
5753        let (tmp, talks) = store();
5754        let spec = mock_agent(tmp.path(), REPLY, env("hi"));
5755        let cfg = config(spec);
5756        let talk = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5757
5758        let artifacts = talks.artifacts_of(&talk.id);
5759        std::fs::create_dir_all(&artifacts).expect("create artifacts dir");
5760        std::fs::write(artifacts.join("turn-1.txt"), "hello").expect("write artifact");
5761
5762        talks.remove(&talk.id).expect("remove");
5763        assert!(!talks.path_of(&talk.id).is_file(), "the record is gone");
5764        assert!(!artifacts.is_dir(), "the artifacts directory is gone");
5765        assert!(
5766            talks.get(&talk.id).is_err(),
5767            "a removed talk cannot be read back"
5768        );
5769
5770        let err = talks
5771            .remove("nonexistent-id")
5772            .expect_err("unknown id refused");
5773        assert!(err.to_string().contains("no talk matches"), "{err}");
5774        let _ = &cfg; // config kept only to build the agent above
5775    }
5776
5777    #[tokio::test]
5778    async fn a_delete_that_lands_while_a_turn_is_in_flight_is_not_undone_by_the_reply() {
5779        let (tmp, talks) = store();
5780        let spec = mock_agent(tmp.path(), REPLY, env("here you go"));
5781        let cfg = config(spec);
5782        // The in-flight turn's own handle, loaded before the delete lands -
5783        // the same shape as the matching close test above.
5784        let mut in_flight = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5785
5786        talks.remove(&in_flight.id).expect("remove");
5787        assert!(
5788            talks.get(&in_flight.id).is_err(),
5789            "the delete landed on disk before the turn finished"
5790        );
5791
5792        // The turn's own handle has no way to know the record is gone -
5793        // finishing it must not write the file back into existence.
5794        respond(
5795            &talks.claim_turn(&in_flight.id).unwrap().unwrap(),
5796            &mut in_flight,
5797            &talks,
5798            &cfg,
5799            "one more question",
5800        )
5801        .await
5802        .expect("the turn itself still completes rather than erroring");
5803
5804        assert!(
5805            talks.get(&in_flight.id).is_err(),
5806            "a delete must stick even when a turn that started before it finishes after it"
5807        );
5808    }
5809
5810    #[test]
5811    fn a_delete_that_lands_before_record_is_called_is_not_undone_by_it() {
5812        let (tmp, talks) = store();
5813        let spec = mock_agent(tmp.path(), REPLY, env("hi"));
5814        let cfg = config(spec);
5815        // The handle `web::talk_say` would have read before awaiting config
5816        // discovery, then carried across that await into `record`.
5817        let mut stale = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5818
5819        talks.remove(&stale.id).expect("remove");
5820
5821        // The stale handle has no way to know the record is gone - a
5822        // `record` that trusted it would append a turn and write the
5823        // conversation back into existence.
5824        let err = record(&mut stale, &talks, "still there?", Vec::new())
5825            .expect_err("a delete that landed first must be honored, not overwritten");
5826        assert!(err.to_string().contains("deleted"), "{err}");
5827
5828        assert!(
5829            talks.get(&stale.id).is_err(),
5830            "record must not resurrect a conversation deleted while its snapshot was stale"
5831        );
5832        let _ = &cfg; // config kept only to build the agent above
5833    }
5834
5835    #[test]
5836    fn a_delete_that_lands_before_close_is_called_is_not_undone_by_it() {
5837        let (tmp, talks) = store();
5838        let spec = mock_agent(tmp.path(), REPLY, env("hi"));
5839        let cfg = config(spec);
5840        // `web::talk_close` loads `talk` and calls `close` right after - this
5841        // stands in for a delete landing in that gap.
5842        let mut stale = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5843
5844        talks.remove(&stale.id).expect("remove");
5845
5846        // The stale handle has no way to know the record is gone - a `close`
5847        // that fell back to it would write the conversation back into
5848        // existence, closed.
5849        let err = close(&mut stale, &talks)
5850            .expect_err("a delete that landed first must be honored, not overwritten");
5851        assert!(err.to_string().contains("deleted"), "{err}");
5852
5853        assert!(
5854            talks.get(&stale.id).is_err(),
5855            "close must not resurrect a conversation deleted while its snapshot was stale"
5856        );
5857        let _ = &cfg; // config kept only to build the agent above
5858    }
5859
5860    #[test]
5861    fn a_delete_that_lands_before_reopen_is_called_is_not_undone_by_it() {
5862        let (tmp, talks) = store();
5863        let spec = mock_agent(tmp.path(), REPLY, env("hi"));
5864        let cfg = config(spec);
5865        // `web::talk_reopen` loads `talk` and calls `reopen` right after -
5866        // this stands in for a delete landing in that gap.
5867        let mut stale = begin(&talks, &cfg, tmp.path().to_owned(), None).expect("begin");
5868        close(&mut stale, &talks).expect("close");
5869
5870        talks.remove(&stale.id).expect("remove");
5871
5872        // The stale handle has no way to know the record is gone - a
5873        // `reopen` that fell back to it would write the conversation back
5874        // into existence, open.
5875        let err = reopen(&mut stale, &talks)
5876            .expect_err("a delete that landed first must be honored, not overwritten");
5877        assert!(err.to_string().contains("deleted"), "{err}");
5878
5879        assert!(
5880            talks.get(&stale.id).is_err(),
5881            "reopen must not resurrect a conversation deleted while its snapshot was stale"
5882        );
5883        let _ = &cfg; // config kept only to build the agent above
5884    }
5885
5886    #[test]
5887    fn list_puts_open_talks_before_closed_ones() {
5888        let (tmp, talks) = store();
5889        let make = |id: &str, status: TalkStatus| {
5890            let mut t = Talk {
5891                recommendations: Default::default(),
5892                posts_read_through: None,
5893                pending_breaks: None,
5894                pending_messages: Vec::new(),
5895                schema: SCHEMA,
5896                id: id.to_owned(),
5897                repo: tmp.path().to_owned(),
5898                agent: "mock".to_owned(),
5899                status,
5900                turns: Vec::new(),
5901                pending: String::new(),
5902                pending_attachments: Vec::new(),
5903                fallback: false,
5904                persona: String::new(),
5905                persona_dirty: false,
5906                implementers: 1,
5907                implementers_dirty: false,
5908                notified: Vec::new(),
5909                completion_notices: Vec::new(),
5910                created_at: Timestamp::now(),
5911                updated_at: Timestamp::now(),
5912                seat: SeatState::new(SEAT, "mock", 7),
5913            };
5914            talks.put(&mut t).expect("put");
5915        };
5916        make("20260901-000000-0001", TalkStatus::Open);
5917        make("20260902-000000-0002", TalkStatus::Open);
5918        make("20260903-000000-0003", TalkStatus::Closed);
5919
5920        let ids: Vec<String> = talks.list().into_iter().map(|t| t.id).collect();
5921        assert_eq!(
5922            ids,
5923            [
5924                "20260902-000000-0002",
5925                "20260901-000000-0001",
5926                "20260903-000000-0003"
5927            ]
5928        );
5929        assert_eq!(talks.count_open(), 2);
5930    }
5931
5932    #[test]
5933    fn tasks_of_finds_only_this_talks_own_tasks() {
5934        let dir = tempfile::tempdir().expect("tempdir");
5935        let queue = Queue::at(dir.path().join("queue"));
5936
5937        let mut mine = Task::new(
5938            "rework the loader".to_owned(),
5939            "rework the loader".to_owned(),
5940            PathBuf::from("/repo"),
5941            Source::Agent {
5942                run: "20260904-014455-ab12".to_owned(),
5943                node: "chat".to_owned(),
5944            },
5945        );
5946        queue.put(&mut mine).expect("put mine");
5947
5948        let mut theirs = Task::new(
5949            "unrelated".to_owned(),
5950            "unrelated".to_owned(),
5951            PathBuf::from("/repo"),
5952            Source::Agent {
5953                run: "20260904-090000-zz99".to_owned(),
5954                node: "implement".to_owned(),
5955            },
5956        );
5957        queue.put(&mut theirs).expect("put theirs");
5958
5959        let mut human = Task::new(
5960            "typed by hand".to_owned(),
5961            "typed by hand".to_owned(),
5962            PathBuf::from("/repo"),
5963            Source::Human,
5964        );
5965        queue.put(&mut human).expect("put human");
5966
5967        let found = tasks_of(&queue, "20260904-014455-ab12");
5968        assert_eq!(found.len(), 1);
5969        assert_eq!(found[0].id, mine.id);
5970    }
5971
5972    #[test]
5973    fn the_briefing_names_solo_task_add() {
5974        let brief = briefing(Path::new("/repo"), "en", false);
5975        assert!(brief.contains("magi task add --solo"));
5976        assert!(brief.contains("/repo"));
5977        assert!(!brief.contains("Hold this conversation in"));
5978    }
5979
5980    /// Talk fixes `repo` at the directory the conversation was opened in, so
5981    /// an agent asked to change some other checkout has no path to it unless
5982    /// the briefing itself says `--repo` can take a short name - see
5983    /// `resolve_repo_by_name` in `src/main.rs`, which is what actually
5984    /// resolves it.
5985    #[test]
5986    fn the_briefing_explains_targeting_a_different_repository_by_name() {
5987        let brief = briefing(Path::new("/repo"), "en", false);
5988        assert!(brief.contains("--repo does not have to be a full path"));
5989        assert!(brief.contains("owner/repo"));
5990        assert!(brief.contains("magi repos"));
5991        assert!(brief.contains("ask the operator"));
5992    }
5993
5994    #[test]
5995    fn the_briefing_tells_the_assistant_to_pass_images_with_attach() {
5996        let brief = briefing(Path::new("/repo"), "en", false);
5997        assert!(brief.contains("--attach <path>"), "{brief}");
5998        assert!(brief.contains("deleting this conversation"), "{brief}");
5999    }
6000
6001    #[test]
6002    fn the_briefing_names_the_language_when_it_is_not_english() {
6003        let brief = briefing(Path::new("/repo"), "Japanese", false);
6004        assert!(brief.contains("Hold this conversation in Japanese"));
6005    }
6006
6007    #[test]
6008    fn the_briefing_forbids_writes_unless_the_repository_opted_in() {
6009        let read_only = briefing(Path::new("/repo"), "en", false);
6010        assert!(read_only.contains("Do not write files"));
6011        assert!(!read_only.contains("allow_write"));
6012
6013        let writable = briefing(Path::new("/repo"), "en", true);
6014        assert!(!writable.contains("Do not write files"));
6015        assert!(writable.contains("allow_write = true"));
6016        // Still names the queue for anything past a small named edit, and
6017        // still tells the agent to report what it changed.
6018        assert!(writable.contains("magi task add --solo"));
6019        assert!(writable.contains("say plainly what you"));
6020    }
6021
6022    #[test]
6023    fn the_briefing_says_the_chat_cannot_push_and_tasks_report_back() {
6024        let b = briefing_for(Path::new("/r"), "en", false, None, None, 1);
6025        assert!(
6026            b.contains("cannot send the operator a message on your own"),
6027            "{b}"
6028        );
6029        assert!(b.contains(crate::prompt::CHAT_TASK_NOTICE_HEADING));
6030        assert!(b.contains("magi talk watch --timeout 900"));
6031        assert!(b.contains("nor a successfully registered watch"));
6032    }
6033
6034    #[test]
6035    fn a_post_is_replayed_until_the_agent_answers_again() {
6036        let (_tmp, talks) = store();
6037        let mut talk = ctx_talk("mock", Vec::new());
6038        talks.put(&mut talk).expect("put");
6039        post(&talks, &talk.id, "  found it  ", "20260902-000000-aaaa").expect("post");
6040        let mut talk = talks.get(&talk.id).expect("get");
6041        let block = posts_since_last_reply(&talk).expect("block");
6042        assert!(block.starts_with("# Posted by a task"), "{block}");
6043        assert!(block.contains("found it"));
6044        talk.turns.push(reply("ok", None));
6045        assert!(posts_since_last_reply(&talk).is_none());
6046        assert!(post(&talks, &talk.id, "  ", "r").is_err());
6047        close(&mut talk, &talks).expect("close");
6048        assert!(post(&talks, &talk.id, "late", "r").is_err());
6049    }
6050
6051    #[test]
6052    fn queue_once_accepts_a_key_a_single_time_even_after_a_drain() {
6053        let (_tmp, talks) = store();
6054        let mut talk = ctx_talk("mock", Vec::new());
6055        talks.put(&mut talk).expect("put");
6056        assert!(queue_once(&mut talk, &talks, "one", "k").expect("first"));
6057        drain(&mut talk, &talks).expect("drain");
6058        assert!(!queue_once(&mut talk, &talks, "one", "k").expect("again"));
6059        assert!(talks.get(&talk.id).expect("get").pending.is_empty());
6060    }
6061    #[test]
6062    fn sender_mixed_notifications_survive_reload_drain_and_reply() {
6063        let (_tmp, talks) = store();
6064        let mut talk = ctx_talk("mock", Vec::new());
6065        talks.put(&mut talk).unwrap();
6066        queue(&mut talk, &talks, "owner", Vec::new()).unwrap();
6067        assert!(queue_once(&mut talk, &talks, "notice", "n").unwrap());
6068        let mut talk = talks.get(&talk.id).unwrap();
6069        assert_eq!(talk.pending_messages[0].sender, Some(Sender::Operator));
6070        assert_eq!(talk.pending_messages[1].sender, Some(Sender::Magi));
6071        assert!(
6072            edit_pending_message(&mut talk, &talks, "forged", "owner\n\nnotice", &[], Some(1))
6073                .is_err()
6074        );
6075        drain(&mut talk, &talks).unwrap();
6076        let mut talk = talks.get(&talk.id).unwrap();
6077        let turn = talk.turns.last().unwrap();
6078        assert_eq!(turn.who, Who::Operator);
6079        assert_eq!(sender(turn, turn.messages.first()), Sender::Operator);
6080        assert_eq!(sender(turn, turn.messages.get(1)), Sender::Magi);
6081        let reply = resolve_reply(
6082            &talk,
6083            Some(ReplyTarget {
6084                turn: 0,
6085                message: 1,
6086            }),
6087        )
6088        .unwrap()
6089        .unwrap();
6090        assert_eq!(reply.sender, Some(Sender::Magi));
6091        assert!(!queue_once(&mut talk, &talks, "notice", "n").unwrap());
6092    }
6093
6094    #[test]
6095    fn sender_legacy_operator_prose_is_never_migrated_by_heading() {
6096        let mut turn = reply("# Task finished\n\nmagi: operator typed this", None);
6097        turn.who = Who::Operator;
6098        assert_eq!(sender(&turn, None), Sender::Operator);
6099        turn.who = Who::Agent;
6100        turn.body = "magi: failed".into();
6101        assert_eq!(sender(&turn, None), Sender::Magi);
6102        turn.body = "persona reply".into();
6103        assert_eq!(sender(&turn, None), Sender::Persona);
6104        turn.posted_by = Some("task".into());
6105        assert_eq!(sender(&turn, None), Sender::Magi);
6106        turn.posted_by = None;
6107        turn.body = "magi: actual persona words".into();
6108        turn.sender = Some(Sender::Persona);
6109        assert_eq!(sender(&turn, None), Sender::Persona);
6110    }
6111}