feat(maestro): apply the kill's worktree rule when a session exits on its own

When the liveness watcher finds the tmux session gone, it now settles
the worktree through the same probe_worktree/record_retention helpers
the kill uses: pristine trees are removed, anything else is retained as
exited-with-unsaved-work before anything is torn down. The dead session
drops its worktree reference once settled, so dismissing it neither
refuses nor retains twice. If the probe or store write fails, nothing
is destroyed and the tree stays attached to the record.

WORKTREES.md: written down the settle-then-drop and nothing-destroyed-
when-unrecordable rules alongside the existing exit rule.

crates/maestro-cli/src/cli.rs

@@ -125,6 +125,11 @@ pub struct AttachArgs {
 pub struct KillArgs {
     /// Session ID to kill.
     pub session_id: Uuid,
+    /// End the session even when its worktree holds unsaved work (#81). The
+    /// worktree is retained — never removed — and enters the retained store;
+    /// the output says where the work went.
+    #[arg(long)]
+    pub force: bool,
 }
 
 fn parse_kv(s: &str) -> Result<(String, String), String> {
@@ -256,6 +261,32 @@ mod tests {
         }
     }
 
+    #[test]
+    fn parses_kill_with_force() {
+        let id = "01234567-89ab-cdef-0123-456789abcdef";
+        let cli = Cli::try_parse_from(["maestro", "kill", "--force", id]).unwrap();
+        match cli.command {
+            Command::Kill(a) => {
+                assert!(a.force);
+                assert_eq!(a.session_id.to_string(), id);
+            }
+            _ => panic!("wrong subcommand"),
+        }
+    }
+
+    #[test]
+    fn kill_force_defaults_false() {
+        let id = "01234567-89ab-cdef-0123-456789abcdef";
+        let cli = Cli::try_parse_from(["maestro", "kill", id]).unwrap();
+        match cli.command {
+            Command::Kill(a) => {
+                assert!(!a.force);
+                assert_eq!(a.session_id.to_string(), id);
+            }
+            _ => panic!("wrong subcommand"),
+        }
+    }
+
     #[test]
     fn rejects_env_without_equals() {
         let err = Cli::try_parse_from([

crates/maestro-cli/src/dispatch.rs

@@ -62,8 +62,15 @@ pub async fn dispatch(
             run_attach_url(&daemon.attach_url(args.session_id)?, detach).await?;
         }
         Command::Kill(args) => {
-            run_kill(&daemon, &args).await?;
+            let outcome = run_kill(&daemon, &args).await?;
             println!("killed {}", args.session_id);
+            if let Some(w) = outcome.retained {
+                println!(
+                    "worktree retained at {} (branch {}, force-killed)",
+                    w.path.display(),
+                    w.branch
+                );
+            }
         }
         Command::Worktrees(args) => match args.command {
             WorktreesCommand::List(list_args) => {

crates/maestro-cli/src/sessions.rs

@@ -59,9 +59,14 @@ pub async fn run_spawn(client: &DaemonClient, args: SpawnArgs) -> Result<Session
     client.create_session(&body).await
 }
 
-/// `maestro kill`
-pub async fn run_kill(client: &DaemonClient, args: &KillArgs) -> Result<(), ClientError> {
-    client.kill_session(args.session_id).await
+/// `maestro kill`. A forced kill (#81) retains the guarded worktree instead
+/// of removing it; the outcome names what was kept so dispatch can say where
+/// the work went.
+pub async fn run_kill(
+    client: &DaemonClient,
+    args: &KillArgs,
+) -> Result<maestro_protocol::KillOutcome, ClientError> {
+    client.kill_session(args.session_id, args.force).await
 }
 
 #[allow(clippy::write_literal)]

crates/maestro-cli/tests/against_mock_daemon.rs

@@ -76,7 +76,7 @@ async fn killing_an_unknown_session_reports_the_daemon_reason() {
 
     let err = mock
         .client()
-        .kill_session(Uuid::new_v4())
+        .kill_session(Uuid::new_v4(), false)
         .await
         .expect_err("should fail");
 
@@ -102,6 +102,9 @@ async fn spawn_round_trips_through_the_daemon() {
     assert_eq!(created.command, "bash");
     assert_eq!(mock.session_count(), 1);
 
-    mock.client().kill_session(created.id).await.expect("kill");
+    mock.client()
+        .kill_session(created.id, false)
+        .await
+        .expect("kill");
     assert_eq!(mock.session_count(), 0);
 }

crates/maestro-daemon/src/cli.rs

@@ -56,6 +56,9 @@ async fn run(config_override: Option<PathBuf>) -> Result<(), Box<dyn Error>> {
         Some(p) => Tmux::new().with_tmpdir(p),
         None => Tmux::new(),
     };
+    let retained = std::sync::Arc::new(crate::retained_store::RetainedStore::new(
+        config.retained_worktree_dir(),
+    ));
     let manager = SessionManager::new_with_replay(
         std::sync::Arc::new(tmux),
         sidecar,
@@ -66,7 +69,10 @@ async fn run(config_override: Option<PathBuf>) -> Result<(), Box<dyn Error>> {
     .with_spawn_history(crate::spawn_history::SpawnHistory::new(
         config.spawn_history_path(),
         config.spawn_history_days,
-    ));
+    ))
+    // One store instance shared with the API handlers: forced kills (#81)
+    // record here, `GET /api/worktrees` (#80) reads back from here.
+    .with_retained_store((*retained).clone());
 
     let recovered = manager.recover().await?;
     tracing::info!(recovered, "adopted existing tmux sessions");
@@ -85,9 +91,7 @@ async fn run(config_override: Option<PathBuf>) -> Result<(), Box<dyn Error>> {
         config.home_dir.clone(),
         config.spawn_presets.clone(),
         projects,
-        std::sync::Arc::new(crate::retained_store::RetainedStore::new(
-            config.retained_worktree_dir(),
-        )),
+        retained,
     ));
 
     let listener = TcpListener::bind(config.bind_addr).await?;

crates/maestro-daemon/src/local_api.rs

@@ -101,6 +101,8 @@ fn session_error_status(code: &str) -> StatusCode {
         "invalid_env" => StatusCode::BAD_REQUEST,
         // A guard refusal is the daemon doing its job, not a failure (#79).
         "kill_guarded" => StatusCode::CONFLICT,
+        // The retention record failed before anything was torn down (#81).
+        "retain_failed" => StatusCode::INTERNAL_SERVER_ERROR,
         _ => StatusCode::BAD_GATEWAY,
     }
 }
@@ -179,9 +181,29 @@ async fn create_session(
 /// The full error info is passed through rather than rebuilt from
 /// code + message: a `kill_guarded` refusal (#79) carries the unsaved-work
 /// breakdown clients render as "3 uncommitted files, 2 unpushed commits".
-async fn kill_session(State(s): State<LocalApiState>, Path(id): Path<Uuid>) -> Response {
-    match s.manager.kill(id).await {
-        Ok(()) => StatusCode::NO_CONTENT.into_response(),
+///
+/// With `force=true` (#81) the refusal is overridden and the worktree
+/// retained; the answer is `200` with a [`KillOutcome`] naming what was kept.
+/// Without it the historical `204` stands.
+#[derive(Debug, Deserialize)]
+struct KillQuery {
+    #[serde(default)]
+    force: bool,
+}
+
+async fn kill_session(
+    State(s): State<LocalApiState>,
+    Path(id): Path<Uuid>,
+    Query(q): Query<KillQuery>,
+) -> Response {
+    match s.manager.kill(id, q.force).await {
+        Ok(outcome) => {
+            if q.force {
+                (StatusCode::OK, Json(outcome)).into_response()
+            } else {
+                StatusCode::NO_CONTENT.into_response()
+            }
+        }
         Err(e) => {
             let info = e.to_error_info();
             let status = session_error_status(&info.code);

crates/maestro-daemon/src/session.rs

@@ -12,8 +12,8 @@ use std::time::Duration;
 
 use chrono::Utc;
 use maestro_protocol::{
-    ActivityState, CatalogEvent, CreateSessionRequest, ErrorInfo, Session, SessionStatus,
-    WorktreeSpec,
+    ActivityState, CatalogEvent, CreateSessionRequest, ErrorInfo, KillOutcome, RetainedReason,
+    RetainedWorktree, Session, SessionStatus, WorktreeSpec,
 };
 use tokio::sync::{broadcast, Mutex};
 use uuid::Uuid;
@@ -44,6 +44,11 @@ pub struct SessionRecord {
     /// `base_repo` from the original `WorktreeSpec`. Needed because `git
     /// worktree remove` is invoked from the base repo.
     pub worktree_base_repo: Option<PathBuf>,
+    /// Branch the worktree was created for, recorded at spawn time so a
+    /// forced kill (#81) can name it in the retained store without asking
+    /// git. Sessions adopted from a sidecar have no branch on record and
+    /// fall back to a probe.
+    pub worktree_branch: Option<String>,
     pub bridge: PtyBridge,
 }
 
@@ -66,6 +71,11 @@ pub enum SessionError {
     /// so it stays the operator's handle on their own unfinished work.
     #[error("kill refused: the session's worktree holds unsaved work ({0})")]
     KillGuarded(maestro_protocol::InTheWay),
+    /// A forced kill (#81) records the retained worktree *before* tearing
+    /// anything down; if that write fails nothing was destroyed, and the
+    /// error says so rather than silently dropping unrecorded work.
+    #[error("could not record the retained worktree: {0}")]
+    RetainFailed(String),
 }
 
 impl SessionError {
@@ -78,6 +88,7 @@ impl SessionError {
             SessionError::NotFound(_) => "not_found",
             SessionError::InvalidEnv(_) => "invalid_env",
             SessionError::KillGuarded(_) => "kill_guarded",
+            SessionError::RetainFailed(_) => "retain_failed",
         };
         let unsaved_work = match self {
             SessionError::KillGuarded(b) => Some(*b),
@@ -110,6 +121,9 @@ struct Inner {
     /// `None` when not wired (tests, or before `with_spawn_history`); a present
     /// but disabled (`0`-day) history no-ops internally.
     spawn_history: Option<SpawnHistory>,
+    /// Retained-worktree store (#78). `None` when not wired (tests); a forced
+    /// kill (#81) without one fails rather than destroying unrecorded work.
+    retained: Option<crate::retained_store::RetainedStore>,
     sessions: Mutex<HashMap<Uuid, SessionRecord>>,
     events: broadcast::Sender<CatalogEvent>,
 }
@@ -148,6 +162,7 @@ impl SessionManager {
                 fifo_dir,
                 replay_bytes,
                 spawn_history: None,
+                retained: None,
                 sessions: Mutex::new(HashMap::new()),
                 events,
             }),
@@ -170,6 +185,16 @@ impl SessionManager {
         self
     }
 
+    /// Attach the retained-worktree store (#78). A forced kill (#81) needs it
+    /// to record what it retained; without one, forcing a guarded kill fails
+    /// instead of destroying unrecorded work.
+    pub fn with_retained_store(mut self, store: crate::retained_store::RetainedStore) -> Self {
+        let inner = Arc::get_mut(&mut self.inner)
+            .expect("with_retained_store must be called before the SessionManager is cloned");
+        inner.retained = Some(store);
+        self
+    }
+
     /// Recent working dirs for the spawn form's picker (#24), newest-first and
     /// deduped by dir. Empty when no history is wired or it's disabled.
     pub fn recent_dirs(&self) -> Vec<RecentDir> {
@@ -340,6 +365,7 @@ impl SessionManager {
                     fifo_path,
                     worktree_path,
                     worktree_base_repo,
+                    worktree_branch: req.worktree.as_ref().map(|w| w.branch.clone()),
                     bridge,
                 },
             );
@@ -391,11 +417,17 @@ impl SessionManager {
     /// work, so ending a session that still matters becomes a deliberate
     /// `--force`, which retains rather than destroys (WORKTREES.md).
     ///
+    /// With `force` (#81) the refusal — and nothing else — is overridden:
+    /// the session ends and the guarded worktree is retained into the store
+    /// with reason `force-killed`, never removed. The outcome names what was
+    /// kept so a client can say where the work went. A tree whose guard is
+    /// clear never needed the override; force removes it as usual.
+    ///
     /// This is also how an already-dead session is dismissed. tmux is only
     /// asked to kill something it still has — calling `kill-session` on a
     /// session tmux already lost fails, and that failure is not interesting.
-    pub async fn kill(&self, id: Uuid) -> Result<(), SessionError> {
-        let (worktree, alive) = {
+    pub async fn kill(&self, id: Uuid, force: bool) -> Result<KillOutcome, SessionError> {
+        let (worktree, alive, known_branch) = {
             let sessions = self.inner.sessions.lock().await;
             match sessions.get(&id) {
                 None => return Err(SessionError::NotFound(id)),
@@ -405,26 +437,48 @@ impl SessionManager {
                         .clone()
                         .zip(record.worktree_path.clone()),
                     record.session.status == SessionStatus::Running,
+                    record.worktree_branch.clone(),
                 ),
             }
         };
-        // The guard runs before anything is torn down: refusing must leave the
-        // session running, exactly as the operator left it.
-        if let Some((_, path)) = worktree {
-            let probe = tmux_blocking(move || crate::worktree_probe::in_the_way(&path))
-                .await
-                .map_err(|e| SessionError::Worktree(crate::worktree::WorktreeError::from(e)))?;
-            if !probe.is_clear() {
-                return Err(SessionError::KillGuarded(probe));
-            }
+        let probe = match &worktree {
+            Some(wt) => Some(probe_worktree(wt).await?),
+            None => None,
+        };
+        let guarded = probe.as_ref().is_some_and(|p| !p.is_clear());
+        if guarded && !force {
+            return Err(SessionError::KillGuarded(probe.expect("guarded ⇒ probed")));
         }
+
+        let mut retained = None;
+        if guarded {
+            let entry = record_retention(
+                &self.inner,
+                worktree.as_ref().expect("guarded ⇒ worktree"),
+                known_branch,
+                RetainedReason::ForceKilled,
+            )
+            .await?;
+            retained = Some((entry, probe.expect("guarded ⇒ probed")));
+        }
+
         let removed = {
             let mut sessions = self.inner.sessions.lock().await;
             sessions.remove(&id)
         };
         let record = removed.ok_or(SessionError::NotFound(id))?;
-        cleanup_record(&self.inner, id, record, alive).await;
-        Ok(())
+        cleanup_record(&self.inner, id, record, alive, retained.is_some()).await;
+        Ok(KillOutcome {
+            retained: retained.map(|(entry, in_the_way)| RetainedWorktree {
+                id: entry.id,
+                path: entry.path,
+                base_repo: entry.base_repo,
+                branch: entry.branch,
+                reason: entry.reason,
+                retained_at: entry.retained_at,
+                in_the_way,
+            }),
+        })
     }
 
     /// Spawn a background task that polls tmux every `interval` and reaps any
@@ -511,6 +565,9 @@ impl SessionManager {
                     // directory blocks that branch forever (#54 review).
                     worktree_path: session.worktree.as_ref().map(|w| w.path.clone()),
                     worktree_base_repo: session.worktree.as_ref().map(|w| w.base_repo.clone()),
+                    // The sidecar carries no branch; a forced kill of a
+                    // recovered session probes git for it at retain time.
+                    worktree_branch: None,
                     bridge,
                 },
             );
@@ -587,6 +644,49 @@ where
         .expect("tmux blocking task panicked")
 }
 
+/// What would be lost if this session's worktree were removed right now.
+///
+/// The single [`crate::worktree_probe::in_the_way`] call site behind both
+/// endings of a session — the kill guard (#79) and the natural-exit
+/// classifier (#82) — so neither can grow its own answer. Runs on the
+/// blocking pool; callers invoke it before anything is torn down, because
+/// refusing (#79), recording a retention (#81) and classifying an exit (#82)
+/// must all see the tree as it stands.
+async fn probe_worktree(
+    (_, path): &(PathBuf, PathBuf),
+) -> Result<maestro_protocol::InTheWay, SessionError> {
+    let path = path.clone();
+    tmux_blocking(move || crate::worktree_probe::in_the_way(&path))
+        .await
+        .map_err(|e| SessionError::Worktree(crate::worktree::WorktreeError::from(e)))
+}
+
+/// Record a guarded worktree into the retained store with `reason`, BEFORE
+/// the caller tears anything down: a failed write then leaves every artifact
+/// untouched instead of destroying work nothing records anymore. Shared by
+/// the forced kill (`ForceKilled`, #81) and the natural exit
+/// (`ExitedWithUnsavedWork`, #82) so the two paths cannot drift apart.
+async fn record_retention(
+    inner: &Arc<Inner>,
+    (base_repo, path): &(PathBuf, PathBuf),
+    known_branch: Option<String>,
+    reason: RetainedReason,
+) -> Result<crate::retained_store::RetainedWorktreeRecord, SessionError> {
+    let store = inner
+        .retained
+        .clone()
+        .ok_or_else(|| SessionError::RetainFailed("no retained store is wired".into()))?;
+    // Sessions adopted from a sidecar carry no branch; ask the tree itself.
+    let branch = known_branch
+        .or_else(|| crate::worktree_probe::current_branch(path))
+        .unwrap_or_default();
+    let base_repo = base_repo.clone();
+    let path = path.clone();
+    tmux_blocking(move || store.retain(&path, &base_repo, &branch, reason))
+        .await
+        .map_err(|e| SessionError::RetainFailed(e.to_string()))
+}
+
 /// One pass of the liveness watcher: diff the in-memory session set against
 /// the tmux-reported one, and reap anything we still track but tmux has lost.
 async fn run_liveness_pass(inner: &Arc<Inner>) {
@@ -710,17 +810,23 @@ async fn run_activity_pass(inner: &Arc<Inner>) {
 }
 
 /// Mark a session whose tmux session has disappeared as [`SessionStatus::Dead`]
-/// and announce it, keeping everything else intact.
+/// and settle its worktree per the natural-exit rule (#82).
 ///
 /// The record, its [`PtyBridge`] and its sidecar all survive, so a client can
 /// still attach and read the replay buffer — the last thing a crashed agent
-/// printed. Artifacts are only released when the operator dismisses the session
-/// via [`SessionManager::kill`].
+/// printed. The worktree, though, is settled right away: a pristine tree is
+/// removed, anything else is retained into the store with reason
+/// `exited-with-unsaved-work` (#82) — nobody is waiting for an answer here,
+/// so there is nothing to refuse. Once settled, the session drops its
+/// worktree reference so a later dismissal neither re-probes nor retains
+/// twice. When settling is impossible (probe or store failure) nothing is
+/// destroyed: the tree stays attached to the dead record for a later forced
+/// dismissal (#81).
 ///
 /// Dead sessions do not survive a daemon restart: [`SessionManager::recover`]
 /// adopts only what tmux still holds.
 async fn mark_session_dead(inner: &Arc<Inner>, id: Uuid) {
-    let updated = {
+    let flipped = {
         let mut sessions = inner.sessions.lock().await;
         match sessions.get_mut(&id) {
             Some(record) if record.session.status == SessionStatus::Running => {
@@ -728,19 +834,111 @@ async fn mark_session_dead(inner: &Arc<Inner>, id: Uuid) {
                 // Release anyone attached; the scrollback survives for a later
                 // attach.
                 record.bridge.seal();
-                Some(record.session.clone())
+                let session = record.session.clone();
+                let worktree = record
+                    .worktree_base_repo
+                    .clone()
+                    .zip(record.worktree_path.clone());
+                let branch = record.worktree_branch.clone();
+                Some((session, worktree, branch))
             }
             _ => None,
         }
     };
-    let Some(session) = updated else {
+    let Some((mut session, worktree, branch)) = flipped else {
         return;
     };
     tracing::info!(%id, "session vanished from tmux — marking dead");
+
+    if let Some(wt) = &worktree {
+        if settle_worktree_on_exit(inner, id, wt, branch).await {
+            session.worktree = None;
+            let mut sessions = inner.sessions.lock().await;
+            if let Some(record) = sessions.get_mut(&id) {
+                record.session.worktree = None;
+                record.worktree_path = None;
+                record.worktree_base_repo = None;
+                record.worktree_branch = None;
+            }
+        }
+    }
+
     let _ = inner.sidecar.write(&session);
     let _ = inner.events.send(CatalogEvent::SessionUpdated { session });
 }
 
+/// Apply the exit rule (#82) to a vanished session's worktree: probe what
+/// would be lost through the kill's own predicate, then remove a pristine
+/// tree or retain anything else under `exited-with-unsaved-work` — recorded
+/// before anything is torn down, exactly as the forced kill does (#81). A
+/// removal that fails falls back to retaining, since whatever blocked it is
+/// by definition something worth keeping.
+///
+/// Returns whether the worktree was fully accounted for. When it wasn't,
+/// nothing was destroyed and the caller keeps the reference on the record.
+async fn settle_worktree_on_exit(
+    inner: &Arc<Inner>,
+    id: Uuid,
+    (base_repo, path): &(PathBuf, PathBuf),
+    known_branch: Option<String>,
+) -> bool {
+    let probe = match probe_worktree(&(base_repo.clone(), path.clone())).await {
+        Ok(probe) => probe,
+        Err(e) => {
+            tracing::error!(
+                %id,
+                path = %path.display(),
+                error = ?e,
+                "natural exit: could not probe the worktree; leaving it untouched"
+            );
+            return false;
+        }
+    };
+    if probe.is_clear() {
+        match inner.worktree_mgr.remove(base_repo, path) {
+            Ok(()) => {
+                tracing::info!(%id, path = %path.display(), "natural exit: pristine worktree removed");
+                return true;
+            }
+            Err(e) => {
+                tracing::warn!(
+                    %id,
+                    path = %path.display(),
+                    error = ?e,
+                    "natural exit: removal failed; falling back to retention"
+                );
+            }
+        }
+    }
+    match record_retention(
+        inner,
+        &(base_repo.clone(), path.clone()),
+        known_branch,
+        RetainedReason::ExitedWithUnsavedWork,
+    )
+    .await
+    {
+        Ok(entry) => {
+            tracing::info!(
+                %id,
+                path = %path.display(),
+                entry = %entry.id,
+                "natural exit: worktree retained with unsaved work"
+            );
+            true
+        }
+        Err(e) => {
+            tracing::error!(
+                %id,
+                path = %path.display(),
+                error = ?e,
+                "natural exit: could not record the retained worktree; leaving it untouched"
+            );
+            false
+        }
+    }
+}
+
 /// Tear down a session's artifacts after it has already been removed from the
 /// live map. Shared by [`SessionManager::kill`] and [`reap_dead_session`],
 /// which previously hand-rolled the same five-step choreography and only
@@ -750,7 +948,17 @@ async fn mark_session_dead(inner: &Arc<Inner>, id: Uuid) {
 /// gone from the authoritative in-memory map, so a stray FIFO or sidecar can't
 /// resurrect it. `kill_tmux` is `true` for an explicit kill and `false` for the
 /// liveness reaper (tmux already lost the session).
-async fn cleanup_record(inner: &Arc<Inner>, id: Uuid, record: SessionRecord, kill_tmux: bool) {
+///
+/// `retain_worktree` is `true` for a forced kill (#81) that retained the
+/// worktree instead of removing it: the store entry is already written, so
+/// only the removal is skipped.
+async fn cleanup_record(
+    inner: &Arc<Inner>,
+    id: Uuid,
+    record: SessionRecord,
+    kill_tmux: bool,
+    retain_worktree: bool,
+) {
     if kill_tmux {
         let tmux = inner.tmux.clone();
         let _ = tmux_blocking(move || tmux.kill_session(id)).await;
@@ -759,11 +967,18 @@ async fn cleanup_record(inner: &Arc<Inner>, id: Uuid, record: SessionRecord, kil
     if let Err(e) = inner.sidecar.delete(id) {
         tracing::warn!(error = ?e, %id, "cleanup: failed to delete sidecar");
     }
-    // The kill already passed the unsaved-work guard (#79), so this removal
-    // is unconditional; the manager's own nested-worktree refusal is all that
-    // can still leave a tree on disk, and it says where.
     if let (Some(base), Some(path)) = (&record.worktree_base_repo, &record.worktree_path) {
-        if let Err(e) = inner.worktree_mgr.remove(base, path) {
+        if retain_worktree {
+            tracing::info!(
+                %id,
+                path = %path.display(),
+                "forced kill: worktree retained in the store"
+            );
+        } else if let Err(e) = inner.worktree_mgr.remove(base, path) {
+            // The kill already passed the unsaved-work guard (#79), so this
+            // removal is unconditional; the manager's own nested-worktree
+            // refusal is all that can still leave a tree on disk, and it
+            // says where.
             tracing::warn!(
                 %id,
                 path = %path.display(),

crates/maestro-daemon/src/worktree_probe.rs

@@ -90,6 +90,14 @@ pub fn reclaim(base_repo: &Path, path: &Path) -> Result<(), ProbeError> {
     .map(|_| ())
 }
 
+/// Branch checked out at `path`. Sessions adopted from a sidecar carry no
+/// branch on record, so a forced kill (#81) asks the tree itself when it
+/// needs one for the retained store. `None` when git cannot answer.
+pub fn current_branch(path: &Path) -> Option<String> {
+    let out = run_git(path, ["rev-parse", "--abbrev-ref", "HEAD"]).ok()?;
+    Some(String::from_utf8_lossy(&out.stdout).trim().to_string())
+}
+
 fn run_git<I, S>(dir: &Path, args: I) -> Result<std::process::Output, ProbeError>
 where
     I: IntoIterator<Item = S>,

crates/maestro-daemon/tests/attach.rs

@@ -122,7 +122,7 @@ async fn pty_round_trip_through_real_tmux() {
         "marker not seen in {} bytes",
         got.len()
     );
-    let _ = h.manager.kill(id).await;
+    let _ = h.manager.kill(id, false).await;
 }
 
 #[tokio::test]
@@ -170,7 +170,7 @@ async fn multi_attach_broadcasts_to_every_client() {
         "client B missed marker"
     );
 
-    let _ = h.manager.kill(id).await;
+    let _ = h.manager.kill(id, false).await;
 }
 
 #[tokio::test]
@@ -219,7 +219,7 @@ async fn sends_close_frame_when_session_is_killed() {
     // give the broadcast subscriber a moment to register before killing.
     tokio::time::sleep(Duration::from_millis(100)).await;
 
-    h.manager.kill(id).await.unwrap();
+    h.manager.kill(id, false).await.unwrap();
 
     let mut saw_close = false;
     let deadline = std::time::Instant::now() + Duration::from_secs(2);
@@ -267,7 +267,7 @@ async fn ping_text_frame_gets_pong() {
         }
     }
     assert!(saw_pong, "no Pong frame received");
-    let _ = h.manager.kill(id).await;
+    let _ = h.manager.kill(id, false).await;
 }
 
 #[tokio::test]

crates/maestro-daemon/tests/common/mod.rs

@@ -176,6 +176,11 @@ impl DaemonFixture {
             manager =
                 manager.with_spawn_history(SpawnHistory::new(Self::history_path(&state_dir), days));
         }
+        // Same wiring as production: the manager records retained worktrees
+        // into the same dir the API reads back from (#78/#81).
+        let retained =
+            maestro_daemon::retained_store::RetainedStore::new(Self::retained_path(&state_dir));
+        manager = manager.with_retained_store(retained.clone());
         Self {
             tmux,
             manager,
@@ -187,6 +192,10 @@ impl DaemonFixture {
         state_dir.path().join("spawn-history.jsonl")
     }
 
+    fn retained_path(state_dir: &TempDir) -> PathBuf {
+        state_dir.path().join("retained-worktrees")
+    }
+
     pub fn spawn_history_path(&self) -> PathBuf {
         Self::history_path(&self.state_dir)
     }

crates/maestro-daemon/tests/local_api.rs

@@ -25,6 +25,13 @@ struct Harness {
     _fx: DaemonFixture,
 }
 
+impl Harness {
+    /// Store over the same dir the manager records into.
+    fn fx_retained(&self) -> maestro_daemon::retained_store::RetainedStore {
+        self._fx.retained_store()
+    }
+}
+
 async fn spawn(hostname: &str) -> Harness {
     spawn_with(
         hostname,
@@ -386,6 +393,62 @@ async fn delete_a_session_whose_worktree_holds_unsaved_work_returns_409_with_the
     assert!(h.manager.has(session.id).await, "the session keeps running");
 }
 
+/// `?force=true` on the wire (#81): the refusal is overridden, the session
+/// ends, and the answer is 200 with the retained worktree named — path,
+/// branch, reason — while the directory survives on disk.
+#[tokio::test]
+async fn delete_with_force_ends_the_session_and_answers_what_was_retained() {
+    let h = spawn("naps-laptop").await;
+    let repo = tempfile::TempDir::new().unwrap();
+    common::init_repo(repo.path());
+    let wt_path = repo
+        .path()
+        .join("worktrees")
+        .join("feat")
+        .join("guarded")
+        .to_path_buf();
+    let session = h
+        .manager
+        .create(CreateSessionRequest {
+            cwd: repo.path().to_path_buf(),
+            command: "sleep 60".into(),
+            env: HashMap::new(),
+            metadata: HashMap::new(),
+            worktree: Some(WorktreeSpec {
+                base_repo: repo.path().to_path_buf(),
+                branch: "feat/guarded".into(),
+                cleanup_on_kill: true,
+                create_branch: true,
+            }),
+        })
+        .await
+        .unwrap();
+    std::fs::write(wt_path.join("README"), b"unsaved edits\n").unwrap();
+
+    let resp = reqwest::Client::new()
+        .delete(url(h.addr, &format!("/api/sessions/{}", session.id)))
+        .query(&[("force", "true")])
+        .send()
+        .await
+        .unwrap();
+    assert_eq!(resp.status(), reqwest::StatusCode::OK);
+    let body: maestro_protocol::KillOutcome = resp.json().await.unwrap();
+    let retained = body.retained.expect("force names what was kept");
+    assert_eq!(retained.path, wt_path);
+    assert_eq!(retained.branch, "feat/guarded");
+    assert_eq!(
+        retained.reason,
+        maestro_protocol::RetainedReason::ForceKilled
+    );
+
+    // The tree would have blocked; force must not have removed it.
+    assert!(wt_path.exists(), "a forced kill never removes the worktree");
+    // And the store entry exists, so reclaim can find it (#80).
+    let all = h.fx_retained().load_all().unwrap();
+    assert_eq!(all.len(), 1);
+    assert_eq!(all[0].id, retained.id);
+}
+
 #[tokio::test]
 async fn attach_ws_streams_pty_bytes() {
     let h = spawn("naps-laptop").await;
@@ -476,7 +539,7 @@ async fn events_ws_sends_a_snapshot_then_streams_deltas() {
         other => panic!("expected SessionAdded, got {other:?}"),
     }
 
-    h.manager.kill(created.id).await.expect("kill");
+    h.manager.kill(created.id, false).await.expect("kill");
 
     // Activity ticks may interleave; the removal is what we are waiting for.
     let deadline = tokio::time::Instant::now() + Duration::from_secs(5);

crates/maestro-daemon/tests/session_decisions.rs

@@ -211,7 +211,7 @@ async fn killing_a_dead_session_dismisses_it_without_touching_tmux() {
     fx.manager.start_liveness_watcher(Duration::from_millis(25));
     wait_for_status(&fx.manager, s.id, SessionStatus::Dead).await;
 
-    fx.manager.kill(s.id).await.expect("dismiss");
+    fx.manager.kill(s.id, false).await.expect("dismiss");
 
     assert!(fx.manager.snapshot().await.is_empty());
     assert!(
@@ -371,7 +371,7 @@ async fn killing_an_unknown_session_is_reported_as_not_found() {
     let fx = Fx::new();
     let err = fx
         .manager
-        .kill(uuid::Uuid::new_v4())
+        .kill(uuid::Uuid::new_v4(), false)
         .await
         .expect_err("should fail");
     assert_eq!(err.to_error_info().code, "not_found");

crates/maestro-daemon/tests/session_manager.rs

@@ -8,7 +8,9 @@ use std::process::Command;
 use std::sync::Arc;
 use std::time::Duration;
 
-use common::{init_bare_repo, init_repo, push_all_to_origin, wait_for_async, DaemonFixture};
+use common::{
+    init_bare_repo, init_repo, push_all_to_origin, wait_for_async, DaemonFixture, TmuxFixture,
+};
 use maestro_daemon::SessionError;
 use maestro_protocol::CreateSessionRequest;
 use maestro_protocol::{CatalogEvent, SessionStatus, WorktreeSpec};
@@ -44,7 +46,7 @@ async fn create_then_kill_round_trip() {
     // FIFO should also exist
     assert!(fx.fifo_dir().join(format!("{}.fifo", session.id)).exists());
 
-    fx.manager.kill(session.id).await.expect("kill");
+    fx.manager.kill(session.id, false).await.expect("kill");
     assert!(!fx.manager.has(session.id).await);
     assert!(!fx
         .sidecar_dir()
@@ -79,7 +81,7 @@ async fn create_appends_to_spawn_history() {
     assert_eq!(recents[0].cwd, cwd.path());
     assert_eq!(recents[0].last_command, "sleep 30");
 
-    fx.manager.kill(session.id).await.expect("kill");
+    fx.manager.kill(session.id, false).await.expect("kill");
     // History is durable — it survives the kill that deletes the sidecar.
     assert_eq!(fx.manager.recent_dirs().len(), 1);
 }
@@ -185,7 +187,7 @@ async fn create_captures_initial_output_in_replay_buffer() {
         "first-line marker never reached the replay buffer — pipe-pane raced past the child's initial output",
     );
 
-    fx.manager.kill(s.id).await.unwrap();
+    fx.manager.kill(s.id, false).await.unwrap();
 }
 
 /// Regression for NAP-228: after `recover`, PTY output must reach new
@@ -251,7 +253,7 @@ async fn recover_re_pipes_pane_so_output_reaches_subscribers() {
         String::from_utf8_lossy(&buf),
     );
 
-    revived.kill(s.id).await.unwrap();
+    revived.kill(s.id, false).await.unwrap();
 }
 
 #[tokio::test]
@@ -300,7 +302,7 @@ async fn pty_broadcast_delivers_bytes_to_subscribers() {
     assert!(buf_a.windows(want.len()).any(|w| w == want), "A missed");
     assert!(buf_b.windows(want.len()).any(|w| w == want), "B missed");
 
-    fx.manager.kill(s.id).await.unwrap();
+    fx.manager.kill(s.id, false).await.unwrap();
     let _ = (cwd, PathBuf::from("/"));
 }
 
@@ -359,7 +361,7 @@ async fn liveness_watcher_marks_dead_sessions_and_keeps_them_until_dismissed() {
     );
 
     // Dismissing it is what finally releases the artifacts.
-    fx.manager.kill(s.id).await.expect("dismiss");
+    fx.manager.kill(s.id, false).await.expect("dismiss");
     assert!(!fx.manager.has(s.id).await);
     assert!(!fx.sidecar_dir().join(format!("{}.json", s.id)).exists());
     assert!(!fx.fifo_dir().join(format!("{}.fifo", s.id)).exists());
@@ -398,7 +400,7 @@ async fn kill_removes_worktree_even_when_spec_says_keep() {
     init_bare_repo(origin.path());
     push_all_to_origin(repo.path(), origin.path());
 
-    fx.manager.kill(s.id).await.expect("kill");
+    fx.manager.kill(s.id, false).await.expect("kill");
     assert!(!wt_path.exists(), "worktree must be removed on kill");
 }
 
@@ -432,7 +434,11 @@ async fn killing_refuses_a_worktree_with_unsaved_work_and_keeps_the_session() {
     std::fs::write(wt_path.join("README"), b"unsaved edits\n").unwrap();
     std::fs::write(wt_path.join("notes.txt"), b"untracked\n").unwrap();
 
-    let err = fx.manager.kill(s.id).await.expect_err("kill must refuse");
+    let err = fx
+        .manager
+        .kill(s.id, false)
+        .await
+        .expect_err("kill must refuse");
     let SessionError::KillGuarded(breakdown) = err else {
         panic!("expected KillGuarded, got {err:?}");
     };
@@ -456,7 +462,10 @@ async fn killing_refuses_a_worktree_with_unsaved_work_and_keeps_the_session() {
     let origin = TempDir::new().unwrap();
     init_bare_repo(origin.path());
     push_all_to_origin(repo.path(), origin.path());
-    fx.manager.kill(s.id).await.expect("kill after saving");
+    fx.manager
+        .kill(s.id, false)
+        .await
+        .expect("kill after saving");
     assert!(!fx.manager.has(s.id).await);
 }
 
@@ -487,7 +496,11 @@ async fn killing_refuses_until_the_branch_is_pushed() {
         .await
         .expect("create");
 
-    let err = fx.manager.kill(s.id).await.expect_err("unpushed blocks");
+    let err = fx
+        .manager
+        .kill(s.id, false)
+        .await
+        .expect_err("unpushed blocks");
     let SessionError::KillGuarded(breakdown) = err else {
         panic!("expected KillGuarded, got {err:?}");
     };
@@ -499,11 +512,118 @@ async fn killing_refuses_until_the_branch_is_pushed() {
     init_bare_repo(origin.path());
     push_all_to_origin(repo.path(), origin.path());
 
-    fx.manager.kill(s.id).await.expect("pushed branch kills");
+    fx.manager
+        .kill(s.id, false)
+        .await
+        .expect("pushed branch kills");
     assert!(!wt_path.exists());
     assert!(!fx.manager.has(s.id).await);
 }
 
+/// `--force` overrides the refusal and nothing else (#81): the session ends,
+/// but the guarded worktree is RETAINED — never removed — and enters the
+/// retained store with reason `force-killed`. The outcome names what was
+/// kept, so a client can say where the work went.
+#[tokio::test]
+async fn force_killing_a_guarded_worktree_retains_it_in_the_store() {
+    let fx = DaemonFixture::new();
+    let repo = TempDir::new().unwrap();
+    init_repo(repo.path());
+    let wt_path = repo.path().join("worktrees").join("feat").join("forced");
+
+    let s = fx
+        .manager
+        .create(CreateSessionRequest {
+            cwd: repo.path().to_path_buf(),
+            command: "sleep 30".into(),
+            env: HashMap::new(),
+            metadata: HashMap::new(),
+            worktree: Some(WorktreeSpec {
+                base_repo: repo.path().to_path_buf(),
+                branch: "feat/forced".into(),
+                cleanup_on_kill: true,
+                create_branch: true,
+            }),
+        })
+        .await
+        .expect("create");
+    std::fs::write(wt_path.join("README"), b"unsaved edits\n").unwrap();
+
+    // Without force the same kill is refused (#79).
+    assert!(fx.manager.kill(s.id, false).await.is_err());
+
+    let outcome = fx
+        .manager
+        .kill(s.id, true)
+        .await
+        .expect("force overrides the refusal");
+
+    // The tree would have blocked, so force must NOT have removed it...
+    assert!(
+        wt_path.exists(),
+        "a forced kill never removes a tree that would have blocked"
+    );
+    // ...but the session itself still ends.
+    assert!(!fx.manager.has(s.id).await);
+    assert!(!fx.sidecar_dir().join(format!("{}.json", s.id)).exists());
+
+    let retained = outcome.retained.expect("the outcome names what was kept");
+    assert_eq!(retained.path, wt_path);
+    assert_eq!(retained.branch, "feat/forced");
+    assert_eq!(
+        retained.reason,
+        maestro_protocol::RetainedReason::ForceKilled
+    );
+    assert!(
+        retained.in_the_way.uncommitted_files >= 1,
+        "the outcome carries what blocked the unforced kill"
+    );
+
+    // The store holds the same entry, so reclaim can find it later (#80).
+    let all = fx.retained_store().load_all().expect("load_all");
+    assert_eq!(all.len(), 1);
+    assert_eq!(all[0].id, retained.id);
+    assert_eq!(all[0].path, wt_path);
+    assert_eq!(all[0].branch, "feat/forced");
+}
+
+/// Force overrides the *refusal*, not ordinary cleanup (#81): a tree whose
+/// guard is clear never needed the override, so a forced kill removes it as
+/// usual and retains nothing.
+#[tokio::test]
+async fn force_on_a_clear_worktree_removes_it_as_usual() {
+    let fx = DaemonFixture::new();
+    let repo = TempDir::new().unwrap();
+    init_repo(repo.path());
+    let wt_path = repo.path().join("worktrees").join("feat").join("clear");
+
+    let s = fx
+        .manager
+        .create(CreateSessionRequest {
+            cwd: repo.path().to_path_buf(),
+            command: "sleep 30".into(),
+            env: HashMap::new(),
+            metadata: HashMap::new(),
+            worktree: Some(WorktreeSpec {
+                base_repo: repo.path().to_path_buf(),
+                branch: "feat/clear".into(),
+                cleanup_on_kill: true,
+                create_branch: true,
+            }),
+        })
+        .await
+        .expect("create");
+
+    let origin = TempDir::new().unwrap();
+    init_bare_repo(origin.path());
+    push_all_to_origin(repo.path(), origin.path());
+
+    let outcome = fx.manager.kill(s.id, true).await.expect("kill");
+    assert!(!wt_path.exists(), "a clear tree needs no override");
+    assert_eq!(outcome.retained, None);
+    assert!(fx.retained_store().load_all().unwrap().is_empty());
+}
+
 /// Ignored files NEVER block (#79): build output like `target/` exists in
 /// every Rust checkout, and an ignored-only worktree stays killable.
 #[tokio::test]
@@ -543,13 +663,198 @@ async fn killing_an_ignored_only_worktree_succeeds() {
     push_all_to_origin(repo.path(), origin.path());
 
     fx.manager
-        .kill(s.id)
+        .kill(s.id, false)
         .await
         .expect("ignored files never block");
     assert!(!wt_path.exists());
     assert!(!fx.manager.has(s.id).await);
 }
 
+/// Natural exit (#82): a session that ends on its own never reaches the kill
+/// path, so the watcher applies the kill's rule itself. A worktree holding
+/// unsaved work is RETAINED into the store with reason
+/// `exited-with-unsaved-work`; the session stays in the catalog marked dead
+/// until dismissed — and dismissing it afterwards neither refuses nor
+/// retains a second time.
+#[tokio::test]
+async fn natural_exit_of_a_guarded_worktree_retains_it_in_the_store() {
+    let fx = DaemonFixture::new();
+    let repo = TempDir::new().unwrap();
+    init_repo(repo.path());
+    let wt_path = repo.path().join("worktrees").join("feat").join("exited");
+
+    let s = fx
+        .manager
+        .create(CreateSessionRequest {
+            cwd: repo.path().to_path_buf(),
+            command: "sleep 30".into(),
+            env: HashMap::new(),
+            metadata: HashMap::new(),
+            worktree: Some(WorktreeSpec {
+                base_repo: repo.path().to_path_buf(),
+                branch: "feat/exited".into(),
+                cleanup_on_kill: true,
+                create_branch: true,
+            }),
+        })
+        .await
+        .expect("create");
+    std::fs::write(wt_path.join("README"), b"unsaved edits\n").unwrap();
+
+    // The command finishes / the operator types exit: tmux loses the session.
+    fx.tmux.tmux.kill_session(s.id).expect("kill tmux directly");
+    fx.manager.start_liveness_watcher(Duration::from_millis(50));
+    wait_for_async(Duration::from_secs(3), || async {
+        fx.manager.status(s.id).await == Some(SessionStatus::Dead)
+    })
+    .await;
+
+    // The tree would have blocked a kill, so the exit kept it...
+    assert!(
+        wt_path.exists(),
+        "a dirty worktree must be retained on natural exit"
+    );
+    // ...in the store, under the exit reason.
+    let all = fx.retained_store().load_all().expect("load_all");
+    assert_eq!(all.len(), 1);
+    assert_eq!(all[0].path, wt_path);
+    assert_eq!(all[0].branch, "feat/exited");
+    assert_eq!(
+        all[0].reason,
+        maestro_protocol::RetainedReason::ExitedWithUnsavedWork
+    );
+
+    // The session itself stays until dismissed; its worktree reference is
+    // settled, so dismissal is neither refused nor retained twice.
+    // (`Dead` is observable a moment before the settle step finishes.)
+    wait_for_async(Duration::from_secs(3), || async {
+        fx.manager.snapshot().await[0].worktree.is_none()
+    })
+    .await;
+    let snap = fx.manager.snapshot().await;
+    assert_eq!(snap.len(), 1);
+    assert_eq!(snap[0].worktree, None);
+
+    fx.manager.kill(s.id, false).await.expect("dismiss");
+    assert!(!fx.manager.has(s.id).await);
+    assert!(
+        wt_path.exists(),
+        "dismissal must not remove what was retained"
+    );
+    assert_eq!(
+        fx.retained_store().load_all().unwrap().len(),
+        1,
+        "dismissal must not record a second entry for the same tree"
+    );
+}
+
+/// Natural exit (#82), pristine side of the same rule: nothing in the tree,
+/// nothing to keep — the worktree is removed and no store entry appears.
+#[tokio::test]
+async fn natural_exit_of_a_pristine_worktree_removes_it() {
+    let fx = DaemonFixture::new();
+    let repo = TempDir::new().unwrap();
+    init_repo(repo.path());
+    let wt_path = repo.path().join("worktrees").join("feat").join("clean");
+
+    let s = fx
+        .manager
+        .create(CreateSessionRequest {
+            cwd: repo.path().to_path_buf(),
+            command: "sleep 30".into(),
+            env: HashMap::new(),
+            metadata: HashMap::new(),
+            worktree: Some(WorktreeSpec {
+                base_repo: repo.path().to_path_buf(),
+                branch: "feat/clean".into(),
+                cleanup_on_kill: true,
+                create_branch: true,
+            }),
+        })
+        .await
+        .expect("create");
+
+    // Push everything so the branch has no unpushed commits: fully pristine.
+    let origin = TempDir::new().unwrap();
+    init_bare_repo(origin.path());
+    push_all_to_origin(repo.path(), origin.path());
+
+    fx.tmux.tmux.kill_session(s.id).expect("kill tmux directly");
+    fx.manager.start_liveness_watcher(Duration::from_millis(50));
+    wait_for_async(Duration::from_secs(3), || async {
+        fx.manager.status(s.id).await == Some(SessionStatus::Dead)
+    })
+    .await;
+
+    assert!(
+        !wt_path.exists(),
+        "a pristine worktree leaves nothing behind"
+    );
+    assert!(
+        fx.retained_store().load_all().unwrap().is_empty(),
+        "nothing was retained"
+    );
+    assert!(
+        fx.manager.has(s.id).await,
+        "the session stays until dismissed"
+    );
+}
+
+/// When nothing can be recorded — here: no retained store wired — the exit
+/// destroys nothing. The tree stays on disk AND stays attached to the dead
+/// record, where dismissing it later still goes through #81's forced-kill
+/// retention. Losing the store must not lose the work.
+#[tokio::test]
+async fn natural_exit_without_a_retained_store_destroys_nothing() {
+    let tmux = TmuxFixture::new();
+    let state_dir = TempDir::new().unwrap();
+    let manager = maestro_daemon::SessionManager::new(
+        Arc::new(tmux.tmux.clone()),
+        maestro_daemon::sidecar::SidecarStore::new(state_dir.path().join("sessions")),
+        maestro_daemon::worktree::WorktreeManager::new(None),
+        state_dir.path().join("fifos"),
+    );
+    // Deliberately NO with_retained_store.
+
+    let repo = TempDir::new().unwrap();
+    init_repo(repo.path());
+    let wt_path = repo.path().join("worktrees").join("feat").join("nostore");
+
+    let s = manager
+        .create(CreateSessionRequest {
+            cwd: repo.path().to_path_buf(),
+            command: "sleep 30".into(),
+            env: HashMap::new(),
+            metadata: HashMap::new(),
+            worktree: Some(WorktreeSpec {
+                base_repo: repo.path().to_path_buf(),
+                branch: "feat/nostore".into(),
+                cleanup_on_kill: true,
+                create_branch: true,
+            }),
+        })
+        .await
+        .expect("create");
+    std::fs::write(wt_path.join("README"), b"unsaved edits\n").unwrap();
+
+    tmux.tmux.kill_session(s.id).expect("kill tmux directly");
+    manager.start_liveness_watcher(Duration::from_millis(50));
+    wait_for_async(Duration::from_secs(3), || async {
+        manager.status(s.id).await == Some(SessionStatus::Dead)
+    })
+    .await;
+
+    assert!(
+        wt_path.exists(),
+        "nothing may be destroyed when it cannot be recorded"
+    );
+    let snap = manager.snapshot().await;
+    assert!(
+        snap[0].worktree.is_some(),
+        "the tree must stay attached to the record so a later forced kill can retain it"
+    );
+}
+
 fn run_git_in(dir: &std::path::Path, args: &[&str]) {
     let out = Command::new("git")
         .arg("-C")
@@ -587,5 +892,5 @@ async fn liveness_watcher_leaves_running_sessions_alone() {
     tokio::time::sleep(Duration::from_millis(400)).await;
     assert!(fx.manager.has(s.id).await, "live session got reaped");
 
-    fx.manager.kill(s.id).await.unwrap();
+    fx.manager.kill(s.id, false).await.unwrap();
 }

crates/maestro-e2e/src/client.rs

@@ -61,7 +61,7 @@ impl TestClient {
     }
 
     pub async fn kill_session(&self, id: Uuid) -> Result<(), ClientError> {
-        Ok(self.daemon.kill_session(id).await?)
+        Ok(self.daemon.kill_session(id, false).await.map(|_| ())?)
     }
 
     pub async fn list_projects(&self) -> Result<Vec<Project>, ClientError> {

crates/maestro-protocol/src/client.rs

@@ -16,7 +16,7 @@ use uuid::Uuid;
 use crate::routes::{self, UrlError};
 use crate::{
     ApiError, BranchInfo, CreateSessionRequest, DaemonInfo, DirListing, ErrorInfo, GitRoot,
-    Project, RecentDirs, RetainedWorktree, Session,
+    KillOutcome, Project, RecentDirs, RetainedWorktree, Session,
 };
 
 /// Time budget for establishing a TCP connection to the daemon.
@@ -114,12 +114,24 @@ impl DaemonClient {
         self.post(routes::SESSIONS, req).await
     }
 
-    pub async fn kill_session(&self, id: Uuid) -> Result<(), ClientError> {
+    /// Kill a session. With `force`, the daemon overrides the kill refusal
+    /// (#79) and retains the worktree instead of removing it (#81); the
+    /// outcome then names what was kept and where.
+    pub async fn kill_session(&self, id: Uuid, force: bool) -> Result<KillOutcome, ClientError> {
+        let query: &[(&str, &str)] = if force {
+            &[(routes::FORCE_QUERY, "true")]
+        } else {
+            &[]
+        };
         let resp = self
-            .send(Method::DELETE, &routes::session(id), &[], None::<&()>)
+            .send(Method::DELETE, &routes::session(id), query, None::<&()>)
             .await?;
-        check(resp).await?;
-        Ok(())
+        let resp = check(resp).await?;
+        // An unforced kill keeps its historical `204 No Content`.
+        if resp.status() == reqwest::StatusCode::NO_CONTENT {
+            return Ok(KillOutcome { retained: None });
+        }
+        json(resp).await
     }
 
     pub async fn daemon_info(&self) -> Result<DaemonInfo, ClientError> {

crates/maestro-protocol/src/lib.rs

@@ -195,6 +195,19 @@ pub struct RetainedWorktree {
     pub in_the_way: InTheWay,
 }
 
+/// Body of a successful forced `DELETE /api/sessions/:id?force=true` (#81).
+/// A forced kill ends the session and retains its worktree instead of
+/// removing it; this names what was kept, where, and why, so a client can say
+/// where the work went rather than leaving the operator to wonder.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct KillOutcome {
+    /// Set when the worktree was retained with reason `force-killed`; `None`
+    /// when there was nothing to retain — no worktree at all, or one whose
+    /// guard was clear, which a forced kill removes as usual.
+    #[serde(default)]
+    pub retained: Option<RetainedWorktree>,
+}
+
 impl std::fmt::Display for InTheWay {
     /// "3 uncommitted files, 2 unpushed commits" — zero parts are omitted;
     /// all-zero renders "no unsaved work". This is the phrasing every client
@@ -679,6 +692,22 @@ mod tests {
         );
     }
 
+    #[test]
+    fn kill_outcome_round_trips_a_retained_worktree() {
+        let outcome = KillOutcome {
+            retained: Some(sample_retained_worktree()),
+        };
+        assert_eq!(outcome, round_trip(&outcome));
+    }
+
+    /// A kill with nothing to retain — no worktree, or a clean one removed as
+    /// usual — decodes from an absent field.
+    #[test]
+    fn kill_outcome_defaults_to_nothing_retained() {
+        let outcome: KillOutcome = serde_json::from_value(serde_json::json!({})).unwrap();
+        assert_eq!(outcome.retained, None);
+    }
+
     #[test]
     fn catalog_event_round_trips_and_tags() {
         let session = sample_session();

crates/maestro-protocol/src/routes.rs

@@ -38,6 +38,10 @@ pub const PATH_QUERY: &str = "path";
 /// Name of the query parameter [`NAMES`] takes.
 pub const COUNT_QUERY: &str = "count";
 
+/// Name of the boolean query parameter `DELETE /api/sessions/:id` takes (#81).
+/// `force=true` overrides the kill refusal and retains the worktree.
+pub const FORCE_QUERY: &str = "force";
+
 /// `DELETE /api/sessions/:id`.
 pub fn session(id: Uuid) -> String {
     format!("{SESSIONS}/{id}")

crates/maestro-tui/src/app.rs

@@ -733,8 +733,14 @@ impl App {
     }
 
     async fn submit_kill(&mut self, sid: Uuid) {
-        match self.daemon.kill_session(sid).await {
-            Ok(()) => self.release_session(sid),
+        match self.daemon.kill_session(sid, false).await {
+            Ok(outcome) => {
+                if let Some(w) = outcome.retained {
+                    self.state
+                        .toast(format!("worktree retained: {}", w.path.display()));
+                }
+                self.release_session(sid)
+            }
             Err(e) => self.state.toast(format!("kill failed: {e}")),
         }
     }

docs/WORKTREES.md

@@ -75,11 +75,16 @@ That is only safe because force retains.
 
 `--force` overrides the refusal and nothing else. It ends the session and
 **retains** the worktree, which enters the retained store with reason
-`force-killed`. Force never removes a tree that would have blocked.
+`force-killed`. Force never removes a tree that would have blocked. A tree
+whose guard is clear never needed the override, so a forced kill removes it as
+usual — force changes what happens to blocked work, not ordinary cleanup.
 
-Ending a session and destroying its work are two separate destructive acts. Each
-is chosen on its own, and the second one — reclaim — is chosen while looking at
-exactly what would be lost.
+Ending a session and destroying its work are two separate destructive acts.
+Each is chosen on its own, and the second one — reclaim — is chosen while
+looking at exactly what would be lost.
+
+A forced kill answers with what was retained and why, so a client can say where
+the work went rather than leaving the operator to wonder.
 
 ### Exiting on its own
 
@@ -92,6 +97,17 @@ Nobody is waiting for an answer here, so there is nothing to refuse; the choice
 only between removing and retaining. It uses the same predicate as the kill, from
 the same place, so the two paths cannot drift apart.
 
+The session itself stays in the catalog marked dead until it is dismissed — the
+replay buffer of a crashed agent outlives it — but its worktree is settled at
+once. Once settled, the session drops its worktree reference, so dismissing it
+afterwards neither refuses (there is no guard left to trip) nor records a second
+store entry.
+
+If settling is impossible — the probe fails, or the store cannot take the entry —
+nothing is destroyed. The tree stays on disk *and* stays attached to the dead
+record, where dismissing it later still goes through `--force`'s
+retain-before-teardown path. Losing the store must never mean losing the work.
+
 ## The retained store
 
 A `state_dir`-backed record of worktrees the daemon created that no longer have a