@team-agent/installer 0.5.92 → 0.5.97

package/crates/team-agent/src/cli/mod.rs

@@ -516,10 +516,11 @@ pub mod lifecycle_port {
     /// (经同一帧 list_targets pane→session)。state 无任何锚 → 退命名判据 + spare_fallback event。
     fn anchor_sessions_from_state(
         state: &Value,
+        endpoint: Option<&str>,
         pane_targets: &[crate::transport::PaneInfo],
         event_log: &crate::event_log::EventLog,
     ) -> std::collections::BTreeSet<String> {
-        let anchor_pane_ids = collect_state_leader_anchor_pane_ids(state);
+        let anchor_pane_ids = collect_state_leader_anchor_pane_ids(state, endpoint);
         if anchor_pane_ids.is_empty() {
             // 无锚(state 损坏/未记)→ 退纯命名前缀判据(下游 sessions_to_kill 仍 spare 前缀)。
             let _ = event_log.write(
@@ -549,9 +550,7 @@ pub mod lifecycle_port {
     /// E12 下沉纯函数:bare-shutdown socket 拆除决策。
     #[derive(Debug, Clone, PartialEq, Eq)]
     pub(crate) enum KillDecision {
-        /// socket 独享(无 spare、无外来 session)→ 可整 server 拆除。
-        KillServerExclusive,
-        /// 有 spare(leader 锚/前缀)或非独享 → 逐 session kill,绝不 kill-server。
+        /// Ownership-gated cleanup always uses per-session kills; never whole-server teardown.
         KillIndividually {
             to_kill: Vec<crate::transport::SessionName>,
             spared: Vec<crate::transport::SessionName>,
@@ -560,8 +559,7 @@ pub mod lifecycle_port {
 
     ///
     /// E12 纯决策(单测下沉):spare = `anchor_sessions` ∪ `team-agent-leader-*` 前缀(并集,锚优先)。
-    /// 全部 session 都不 spare 且非空 → `KillServerExclusive`(独享 socket 兜底);否则逐 session
-    /// kill 非 spare 的(共享 socket / leader 在 → 绝不整 server 拆)。空 session 集 → 逐 kill(no-op)。
+    /// Always returns a per-session plan; ownership markers are checked by the caller.
     pub(crate) fn sessions_to_kill(
         sessions: &[crate::transport::SessionName],
         anchor_sessions: &std::collections::BTreeSet<String>,
@@ -572,12 +570,9 @@ pub mod lifecycle_port {
         };
         let spared: Vec<_> = sessions.iter().filter(|s| is_spared(s)).cloned().collect();
         let to_kill: Vec<_> = sessions.iter().filter(|s| !is_spared(s)).cloned().collect();
-        // 独享 = 非空 + 无 spare(socket 上每个 session 都是要 kill 的我方 session)。
-        if spared.is_empty() && !sessions.is_empty() {
-            KillDecision::KillServerExclusive
-        } else {
-            KillDecision::KillIndividually { to_kill, spared }
-        }
+        // Fail closed: even an apparently exclusive socket is not proof of
+        // ownership, so whole-server teardown is never selected.
+        KillDecision::KillIndividually { to_kill, spared }
     }
 
     #[derive(Debug, Default)]
@@ -615,118 +610,183 @@ pub mod lifecycle_port {
     }
 
     fn bare_shutdown_socket_cleanup(
+        workspace: &Path,
         transport: &dyn crate::transport::Transport,
         state: &Value,
         event_log: &crate::event_log::EventLog,
     ) -> ShutdownSocketCleanup {
-        // E12 (P0): the leader terminal lives on this socket by design. A bare shutdown must
-        // NOT `kill-server` it away. spare = state-anchor sessions ∪ `team-agent-leader-*`
-        // prefix sessions (union; cr E12 ①). kill_server only when the socket is exclusively
-        // ours (no spare + no foreign session); shared socket → kill our sessions individually
-        // (cr E12 ②). All spare derivation comes from ONE snapshot (list_targets + state) —
-        // no independent ps/tmux re-derivation (N39).
-        let pane_targets = transport.list_targets().unwrap_or_default();
-        let sessions = socket_session_names_from_targets(&pane_targets);
-        if !state_uses_external_leader(state) {
-            return managed_leader_socket_cleanup(transport, state, &sessions, event_log);
-        }
-        let anchor_sessions = anchor_sessions_from_state(state, &pane_targets, event_log);
-        match sessions_to_kill(&sessions, &anchor_sessions) {
-            KillDecision::KillServerExclusive => {
-                if state.get("tmux_socket_source").and_then(Value::as_str) == Some("leader_env") {
-                    let _ = event_log.write(
-                        "shutdown.kill_server_skipped_shared_socket",
-                        json!({
-                            "reason": "leader_env_tmux_socket",
-                            "spared_sessions": [],
-                            "killed_sessions": sessions.iter().map(|s| s.as_str()).collect::<Vec<_>>(),
-                        }),
-                    );
-                    let mut error = None;
-                    for session in &sessions {
-                        let targets = vec![session.as_str().to_string()];
-                        if let Err(audit_error) = crate::kill_audit::pre_kill_audit(
-                            event_log,
-                            transport,
-                            "shutdown.shared_socket_individual",
-                            crate::kill_audit::KILL_SESSION,
-                            &targets,
-                        ) {
-                            error.get_or_insert_with(|| {
-                                format!("pre-kill audit failed: {audit_error}")
-                            });
-                            continue;
-                        }
-                        if let Err(err) = transport.kill_session(session) {
-                            if !tmux_absent_error(&err.to_string()) {
-                                error.get_or_insert_with(|| err.to_string());
-                            }
-                        }
+        // A bare shutdown is scoped to sessions registered in state.  Never turn a
+        // socket-wide enumeration into an authorization decision: a shared/default
+        // tmux socket may contain unrelated teams and the absence of a marker is
+        // deliberately treated as Unknown.
+        let mut candidates = Vec::new();
+        if let Some(session) = state
+            .get("session_name")
+            .and_then(Value::as_str)
+            .filter(|name| !name.is_empty())
+        {
+            candidates.push(crate::transport::SessionName::new(session));
+        }
+        if let Some(teams) = state.get("teams").and_then(Value::as_object) {
+            for team in teams.values() {
+                if let Some(session) = team
+                    .get("session_name")
+                    .and_then(Value::as_str)
+                    .filter(|name| !name.is_empty())
+                {
+                    let session = crate::transport::SessionName::new(session);
+                    if !candidates.iter().any(|existing| existing == &session) {
+                        candidates.push(session);
                     }
-                    return ShutdownSocketCleanup {
-                        killed_sessions: sessions,
-                        spared_sessions: Vec::new(),
-                        error,
-                    };
-                }
-                let targets = sessions
-                    .iter()
-                    .map(|session| session.as_str().to_string())
-                    .collect::<Vec<_>>();
-                let error = match crate::kill_audit::pre_kill_audit(
-                    event_log,
-                    transport,
-                    "shutdown.exclusive_socket",
-                    crate::kill_audit::KILL_SERVER,
-                    &targets,
-                ) {
-                    Ok(()) => transport.kill_server().err().map(|error| error.to_string()),
-                    Err(error) => Some(format!("pre-kill audit failed: {error}")),
-                };
-                ShutdownSocketCleanup {
-                    killed_sessions: sessions,
-                    spared_sessions: Vec::new(),
-                    error,
                 }
             }
-            KillDecision::KillIndividually { to_kill, spared } => {
-                if !spared.is_empty() || to_kill.len() != sessions.len() {
-                    // shared socket / leader spared → never whole-server teardown.
-                    let _ = event_log.write(
-                        "shutdown.kill_server_skipped_shared_socket",
-                        json!({
-                            "spared_sessions": spared.iter().map(|s| s.as_str()).collect::<Vec<_>>(),
-                            "killed_sessions": to_kill.iter().map(|s| s.as_str()).collect::<Vec<_>>(),
-                        }),
-                    );
+        }
+        let targets = match transport.list_targets() {
+            Ok(targets) => targets,
+            Err(error) => {
+                return ShutdownSocketCleanup {
+                    error: Some(format!("session ownership probe failed: {error}")),
+                    ..ShutdownSocketCleanup::default()
                 }
-                let mut error = None;
-                for session in &to_kill {
-                    let targets = vec![session.as_str().to_string()];
-                    if let Err(audit_error) = crate::kill_audit::pre_kill_audit(
+            }
+        };
+        let protected_sessions = anchor_sessions_from_state(
+            state,
+            transport.tmux_endpoint().as_deref(),
+            &targets,
+            event_log,
+        );
+        let mut result = ShutdownSocketCleanup::default();
+        for session in candidates {
+            if protected_sessions.contains(session.as_str()) {
+                push_unique_session(&mut result.spared_sessions, session);
+                continue;
+            }
+            match session_ownership(workspace, state, transport, &session, &targets) {
+                SessionOwnership::Owned => {
+                    let target = vec![session.as_str().to_string()];
+                    if let Err(error) = crate::kill_audit::pre_kill_audit_scoped(
                         event_log,
                         transport,
-                        "shutdown.selected_session",
+                        "shutdown.state_owned_session",
                         crate::kill_audit::KILL_SESSION,
-                        &targets,
+                        &target,
+                        workspace,
+                        state_for_session(state, session.as_str())
+                            .and_then(|identity| identity.get("team_key").and_then(Value::as_str)),
+                        state_for_session(state, session.as_str()).and_then(state_generation),
+                        "session_marker+state_session+pane_path",
+                        "positive workspace/team ownership",
                     ) {
-                        error
-                            .get_or_insert_with(|| format!("pre-kill audit failed: {audit_error}"));
-                        continue;
-                    }
-                    if let Err(err) = transport.kill_session(session) {
-                        if !tmux_absent_error(&err.to_string()) {
-                            error.get_or_insert_with(|| err.to_string());
+                        result.error.get_or_insert_with(|| format!("pre-kill audit failed: {error}"));
+                    } else if let Err(error) = transport.kill_session(&session) {
+                        if !tmux_absent_error(&error.to_string()) {
+                            result.error.get_or_insert_with(|| error.to_string());
                         }
+                    } else {
+                        result.killed_sessions.push(session);
                     }
                 }
-                ShutdownSocketCleanup {
-                    killed_sessions: to_kill,
-                    spared_sessions: spared,
-                    error,
+                SessionOwnership::Foreign => {
+                    push_unique_session(&mut result.spared_sessions, session);
                 }
+                SessionOwnership::Unknown => {
+                    push_unique_session(&mut result.spared_sessions, session);
+                    result.error.get_or_insert_with(|| "session ownership is unknown; destructive cleanup refused".to_string());
+                }
+                SessionOwnership::Gone => {}
             }
         }
+        result
+    }
+
+    #[derive(Debug, Clone, Copy, PartialEq, Eq)]
+    enum SessionOwnership {
+        Owned,
+        Foreign,
+        Unknown,
+        Gone,
+    }
+
+    fn state_generation(state: &Value) -> Option<&str> {
+        state
+            .get("generation")
+            .and_then(Value::as_str)
+            .or_else(|| {
+                state
+                    .get("agents")
+                    .and_then(Value::as_object)
+                    .and_then(|agents| {
+                        agents.values().find_map(|agent| {
+                            agent.get("spawned_at").and_then(Value::as_str)
+                        })
+                    })
+            })
+    }
+
+    fn state_for_session<'a>(state: &'a Value, session: &str) -> Option<&'a Value> {
+        if state.get("session_name").and_then(Value::as_str) == Some(session) {
+            return Some(state);
+        }
+        state
+            .get("teams")
+            .and_then(Value::as_object)
+            .and_then(|teams| teams.values().find_map(|team| state_for_session(team, session)))
+    }
+
+    fn session_ownership(
+        workspace: &Path,
+        state: &Value,
+        transport: &dyn crate::transport::Transport,
+        session: &crate::transport::SessionName,
+        targets: &[crate::transport::PaneInfo],
+    ) -> SessionOwnership {
+        let identity_state = state_for_session(state, session.as_str()).unwrap_or(state);
+        let session_targets = targets
+            .iter()
+            .filter(|target| target.session.as_str() == session.as_str())
+            .collect::<Vec<_>>();
+        if session_targets.is_empty() {
+            return SessionOwnership::Gone;
+        }
+        let owner = match transport.session_owner(session) {
+            Ok(Some(owner)) => owner,
+            Ok(None) | Err(_) => return SessionOwnership::Unknown,
+        };
+        let canonical = workspace
+            .canonicalize()
+            .unwrap_or_else(|_| workspace.to_path_buf())
+            .to_string_lossy()
+            .to_string();
+        if owner.workspace != canonical {
+            return SessionOwnership::Foreign;
+        }
+        let Some(expected_team) = identity_state
+            .get("team_key")
+            .or_else(|| identity_state.get("active_team_key"))
+            .and_then(Value::as_str)
+            .filter(|team| !team.is_empty())
+        else {
+            return SessionOwnership::Unknown;
+        };
+        if owner.team != expected_team {
+            return SessionOwnership::Foreign;
+        }
+        let Some(expected_generation) = state_generation(identity_state)
+            .filter(|generation| !generation.is_empty()) else {
+            return SessionOwnership::Unknown;
+        };
+        if owner.generation != expected_generation {
+            return SessionOwnership::Foreign;
+        }
+        if session_targets.iter().any(|target| {
+            target.current_path.as_ref().map_or(true, |path| {
+                !path_is_under(path.as_path(), workspace)
+            })
+        }) {
+            return SessionOwnership::Unknown;
+        }
+        SessionOwnership::Owned
     }
 
     #[cfg(test)]
@@ -737,6 +797,43 @@ pub mod lifecycle_port {
         use crate::transport::test_support::OfflineTransport;
         use crate::transport::{PaneId, PaneInfo, SessionName};
 
+        #[test]
+        fn leader_process_protection_does_not_alias_panes_across_endpoints() {
+            let state = json!({
+                "team_owner": {"pane_id": "%2"},
+                "leader_receiver": {"pane_id": "%2", "tmux_socket": "/tmp/s4-host-only.sock"}
+            });
+            let table = [(10, 1), (20, 10), (30, 10), (31, 30)]
+                .into_iter()
+                .map(|(pid, ppid)| ProcessInfo { pid, ppid, pgid: None, session: None, command: String::new() })
+                .collect::<Vec<_>>();
+            for (endpoint, pane_pid, should_protect) in [
+                ("/tmp/s4-worker-only.sock", 20, false),
+                ("/tmp/s4-host-only.sock", 30, true),
+            ] {
+                let transport = OfflineTransport::new()
+                    .with_tmux_endpoint(endpoint)
+                    .with_targets(vec![PaneInfo {
+                        pane_id: PaneId::new("%2"),
+                        session: SessionName::new("ordinary-session"),
+                        window_index: None,
+                        window_name: None,
+                        pane_index: None,
+                        tty: None,
+                        current_command: None,
+                        current_path: None,
+                        active: true,
+                        pane_pid: Some(pane_pid),
+                        leader_env: BTreeMap::new(),
+                    }]);
+                let mut protected = ShutdownProtection::default();
+                extend_protection_with_leader_panes(&mut protected, &transport, &state, &table);
+                assert_eq!(protected.pids.contains(&pane_pid), should_protect);
+                assert_eq!(protected.pids.contains(&31), should_protect, "leader descendants");
+                assert_eq!(protected.pids.contains(&10), should_protect, "leader's server");
+            }
+        }
+
         #[test]
         fn audit_write_failure_skips_destructive_kill() {
             let root =
@@ -762,10 +859,12 @@ pub mod lifecycle_port {
                     leader_env: BTreeMap::new(),
                 }]);
             let cleanup = bare_shutdown_socket_cleanup(
+                Path::new("/tmp/team-agent-test"),
                 &transport,
                 &json!({
                     "is_external_leader": true,
                     "tmux_socket_source": "leader_env",
+                    "session_name": "team-audit-fail",
                 }),
                 &event_log,
             );
@@ -777,7 +876,7 @@ pub mod lifecycle_port {
             assert!(cleanup
                 .error
                 .as_deref()
-                .is_some_and(|error| error.contains("pre-kill audit failed")));
+                .is_some_and(|error| error.contains("ownership is unknown")));
             let _ = std::fs::remove_dir_all(root);
         }
     }
@@ -786,106 +885,6 @@ pub mod lifecycle_port {
         crate::state::projection::state_is_external_leader(state)
     }
 
-    fn managed_leader_socket_cleanup(
-        transport: &dyn crate::transport::Transport,
-        state: &Value,
-        sessions: &[crate::transport::SessionName],
-        event_log: &crate::event_log::EventLog,
-    ) -> ShutdownSocketCleanup {
-        let target = state
-            .get("session_name")
-            .and_then(Value::as_str)
-            .filter(|session| !session.is_empty())
-            .map(crate::transport::SessionName::new);
-        // E49 (0.3.24 P0, shutdown kills leader CLI): for managed leader topology
-        // the target session may carry the leader anchor pane. The pre-fix code
-        // pushed it into `to_kill` and then issued `kill_session` — which ended
-        // the leader pane (= leader CLI). When the target session has a live
-        // anchor pane in `list_targets`, spare it instead.
-        let leader_anchor_ids = collect_state_leader_anchor_pane_ids(state);
-        let live_targets = transport.list_targets().unwrap_or_default();
-        let mut to_kill = Vec::new();
-        let mut target_spared_for_anchor: Option<crate::transport::SessionName> = None;
-        if let Some(target) = target {
-            let target_has_anchor = live_targets.iter().any(|t| {
-                t.session.as_str() == target.as_str()
-                    && leader_anchor_ids.contains(t.pane_id.as_str())
-            });
-            if target_has_anchor {
-                target_spared_for_anchor = Some(target);
-            } else if sessions.is_empty()
-                || sessions
-                    .iter()
-                    .any(|session| session.as_str() == target.as_str())
-            {
-                to_kill.push(target);
-            }
-        }
-        let mut spared = sessions
-            .iter()
-            .filter(|session| {
-                !to_kill
-                    .iter()
-                    .any(|target| target.as_str() == session.as_str())
-            })
-            .cloned()
-            .collect::<Vec<_>>();
-        if let Some(anchored) = target_spared_for_anchor {
-            if !spared.iter().any(|s| s.as_str() == anchored.as_str()) {
-                spared.push(anchored);
-            }
-        }
-        let _ = event_log.write(
-            "shutdown.kill_server_skipped_managed_leader",
-            json!({
-                "reason": "managed_leader_topology",
-                "spared_sessions": spared.iter().map(|s| s.as_str()).collect::<Vec<_>>(),
-                "killed_sessions": to_kill.iter().map(|s| s.as_str()).collect::<Vec<_>>(),
-            }),
-        );
-        let mut error = None;
-        for session in &to_kill {
-            // 0.3.28 Step 6 (warn-only invariant): the kill list MUST NOT
-            // contain a leader session. `sessions_to_kill` already excludes
-            // leader-prefixed sessions, but the managed-leader cleanup path
-            // (this function) computes its own `to_kill` from `target`. Any
-            // leader-prefixed entry here is a topology violation introduced
-            // somewhere upstream. Log loudly and skip the kill.
-            if session
-                .as_str()
-                .starts_with(crate::leader::LEADER_SESSION_PREFIX)
-            {
-                eprintln!(
-                    "team_agent::layout shutdown_invariant_violation kind=KillListContainsLeaderSession \
-                     session=`{}` action=skipping_kill (post-Step-9 will hard-fail)",
-                    session.as_str()
-                );
-                continue;
-            }
-            let targets = vec![session.as_str().to_string()];
-            if let Err(audit_error) = crate::kill_audit::pre_kill_audit(
-                event_log,
-                transport,
-                "shutdown.managed_worker_session",
-                crate::kill_audit::KILL_SESSION,
-                &targets,
-            ) {
-                error.get_or_insert_with(|| format!("pre-kill audit failed: {audit_error}"));
-                continue;
-            }
-            if let Err(err) = transport.kill_session(session) {
-                if !tmux_absent_error(&err.to_string()) {
-                    error.get_or_insert_with(|| err.to_string());
-                }
-            }
-        }
-        ShutdownSocketCleanup {
-            killed_sessions: to_kill,
-            spared_sessions: spared,
-            error,
-        }
-    }
-
     fn owned_shutdown_endpoint(
         workspace: &Path,
         state: &Value,
@@ -970,40 +969,15 @@ pub mod lifecycle_port {
             );
             return cleanup;
         }
-        if let Err(error) = crate::kill_audit::pre_kill_audit(
-            event_log,
-            transport,
-            "shutdown.empty_owned_endpoint",
-            crate::kill_audit::KILL_SERVER,
-            &[],
-        ) {
-            cleanup.error = Some(format!("pre-kill audit failed: {error}"));
-            return cleanup;
-        }
-        if let Err(error) = transport.kill_server() {
-            cleanup.error = Some(error.to_string());
-        }
-        if let Some(path) = endpoint.socket_file.as_ref() {
-            if path.exists() {
-                match std::fs::remove_file(path) {
-                    Ok(()) => cleanup.removed_files.push(path.display().to_string()),
-                    Err(error) if error.kind() == std::io::ErrorKind::NotFound => {}
-                    Err(error) => {
-                        cleanup.error.get_or_insert_with(|| error.to_string());
-                    }
-                }
-            }
-            if path.exists() {
-                cleanup.residual_files.push(path.display().to_string());
-            }
-        }
+        // Endpoint ownership is not sufficient authority to tear down a tmux
+        // server: an endpoint can be shared after launch.  Leave both the
+        // server and its socket file intact; only explicitly owned sessions
+        // are removed above.
         let _ = event_log.write(
-            "shutdown.owned_endpoint_cleanup",
+            "shutdown.owned_endpoint_cleanup_skipped",
             json!({
                 "endpoint": endpoint.endpoint,
-                "removed_files": cleanup.removed_files.clone(),
-                "residual_files": cleanup.residual_files.clone(),
-                "error": cleanup.error.clone(),
+                "reason": "server_and_socket_cleanup_disabled_fail_closed",
             }),
         );
         cleanup
@@ -1081,10 +1055,18 @@ pub mod lifecycle_port {
         root_pids.extend(pane_pids);
         root_pids.sort_unstable();
         root_pids.dedup();
-        let root_pgids = process_pgids(&root_pids, &protected, &entry_table);
         deadline.check("reap_process_tree")?;
-        reap_process_tree(&root_pids, &protected, &entry_table);
-        reap_process_groups(&root_pgids, &protected);
+        let process_event_log = crate::event_log::EventLog::new(&run_workspace);
+        reap_process_tree(
+            &run_workspace,
+            &state,
+            reap_scope,
+            &root_pids,
+            &protected,
+            &entry_table,
+            &process_event_log,
+        );
+
         let mut kill_error: Option<String> = None;
         let mut killed_sessions = Vec::new();
         let mut spared_sessions = Vec::new();
@@ -1100,57 +1082,114 @@ pub mod lifecycle_port {
             // session per `list_targets`. When managed-leader, switch to per-pane
             // cleanup: kill the workers individually and SPARE the leader's pane.
             //
-            // External leader (is_external_leader=true) keeps the unconditional
-            // kill_session — the team session is a disposable worker session in
-            // that topology, the leader pane lives elsewhere.
-            let leader_anchor_ids = collect_state_leader_anchor_pane_ids(&state);
-            let live_targets_now = transport.list_targets().unwrap_or_default();
-            let session_has_leader_anchor = live_targets_now.iter().any(|target| {
-                target.session.as_str() == session.as_str()
-                    && leader_anchor_ids.contains(target.pane_id.as_str())
-            });
-            if crate::state::projection::state_is_managed_leader(&state)
-                && session_has_leader_anchor
-            {
-                let _ = event_log_write_session_spared(&run_workspace, session, &leader_anchor_ids);
-                push_unique_session(&mut spared_sessions, session.clone());
-                let worker_panes = collect_session_worker_panes(
+            // External and managed leaders both pass the same positive ownership
+            // gate; topology only decides whether a verified session is reduced
+            // to worker panes or removed as a whole.
+            let leader_anchor_ids = collect_state_leader_anchor_pane_ids(
+                &state,
+                transport.tmux_endpoint().as_deref(),
+            );
+            let (live_targets_now, live_targets_known) = match transport.list_targets() {
+                Ok(targets) => (targets, true),
+                Err(error) => {
+                    push_unique_session(&mut spared_sessions, session.clone());
+                    kill_error.get_or_insert_with(|| format!("session ownership probe failed: {error}"));
+                    (Vec::new(), false)
+                }
+            };
+            let ownership = if live_targets_known {
+                session_ownership(
+                    &run_workspace,
                     &state,
-                    session.as_str(),
-                    &leader_anchor_ids,
+                    transport,
+                    session,
                     &live_targets_now,
-                );
-                for pane in &worker_panes {
-                    if let Err(error) = transport.kill_pane(pane) {
-                        if !tmux_absent_error(&error.to_string()) {
-                            kill_error.get_or_insert_with(|| error.to_string());
-                        }
-                    }
+                )
+            } else {
+                SessionOwnership::Unknown
+            };
+            if ownership != SessionOwnership::Owned {
+                push_unique_session(&mut spared_sessions, session.clone());
+                if ownership == SessionOwnership::Unknown {
+                    kill_error.get_or_insert_with(|| "session ownership is unknown; destructive cleanup refused".to_string());
                 }
-                let _ =
-                    crate::lifecycle::display::close_team_display_backends(&run_workspace, session);
             } else {
-                let targets = vec![session.as_str().to_string()];
-                let event_log = crate::event_log::EventLog::new(&run_workspace);
-                if let Err(error) = crate::kill_audit::pre_kill_audit(
-                    &event_log,
-                    transport,
-                    "shutdown.team_session",
-                    crate::kill_audit::KILL_SESSION,
-                    &targets,
-                ) {
-                    kill_error = Some(format!("pre-kill audit failed: {error}"));
-                } else if let Err(error) = transport.kill_session(session) {
-                    if tmux_absent_error(&error.to_string()) {
-                        push_unique_session(&mut killed_sessions, session.clone());
-                    } else {
-                        kill_error = Some(error.to_string());
+                let session_has_leader_anchor = live_targets_now.iter().any(|target| {
+                    target.session.as_str() == session.as_str()
+                        && leader_anchor_ids.contains(target.pane_id.as_str())
+                });
+                if crate::state::projection::state_is_managed_leader(&state)
+                    && session_has_leader_anchor
+                {
+                    let _ = event_log_write_session_spared(&run_workspace, session, &leader_anchor_ids);
+                    push_unique_session(&mut spared_sessions, session.clone());
+                    let worker_panes = collect_session_worker_panes(
+                        &state,
+                        session.as_str(),
+                        &leader_anchor_ids,
+                        &live_targets_now,
+                    );
+                    let event_log = crate::event_log::EventLog::new(&run_workspace);
+                    let identity_state = state_for_session(&state, session.as_str()).unwrap_or(&state);
+                    for pane in &worker_panes {
+                        let pane_target = vec![pane.as_str().to_string()];
+                        let audit = crate::kill_audit::pre_kill_audit_scoped(
+                            &event_log,
+                            transport,
+                            "shutdown.managed_worker_pane",
+                            "kill-pane",
+                            &pane_target,
+                            &run_workspace,
+                            identity_state.get("team_key").and_then(Value::as_str),
+                            state_generation(identity_state),
+                            "session_marker+state_worker_pane+pane_path",
+                            "positive workspace/team ownership",
+                        );
+                        if let Err(error) = audit {
+                            kill_error.get_or_insert_with(|| format!("pre-kill audit failed: {error}"));
+                            continue;
+                        }
+                        if let Err(error) = transport.kill_pane(pane) {
+                            if !tmux_absent_error(&error.to_string()) {
+                                kill_error.get_or_insert_with(|| error.to_string());
+                            }
+                        }
                     }
+                    let _ = crate::lifecycle::display::close_team_display_backends(
+                        &run_workspace,
+                        session,
+                    );
                 } else {
-                    push_unique_session(&mut killed_sessions, session.clone());
+                    let targets = vec![session.as_str().to_string()];
+                    let event_log = crate::event_log::EventLog::new(&run_workspace);
+                    if let Err(error) = crate::kill_audit::pre_kill_audit_scoped(
+                        &event_log,
+                        transport,
+                        "shutdown.team_session",
+                        crate::kill_audit::KILL_SESSION,
+                        &targets,
+                        &run_workspace,
+                        state_for_session(&state, session.as_str())
+                            .and_then(|identity| identity.get("team_key").and_then(Value::as_str)),
+                        state_for_session(&state, session.as_str()).and_then(state_generation),
+                        "session_marker+state_session+pane_path",
+                        "positive workspace/team ownership",
+                    ) {
+                        kill_error = Some(format!("pre-kill audit failed: {error}"));
+                    } else if let Err(error) = transport.kill_session(session) {
+                        if tmux_absent_error(&error.to_string()) {
+                            push_unique_session(&mut killed_sessions, session.clone());
+                        } else {
+                            kill_error = Some(error.to_string());
+                        }
+                    } else {
+                        push_unique_session(&mut killed_sessions, session.clone());
+                    }
+                    let _ = crate::lifecycle::display::close_team_display_backends(
+                        &run_workspace,
+                        session,
+                    );
                 }
-                let _ =
-                    crate::lifecycle::display::close_team_display_backends(&run_workspace, session);
             }
         }
         deadline.check("reap_workspace_residuals")?;
@@ -1158,7 +1197,6 @@ pub mod lifecycle_port {
             &run_workspace,
             &state,
             &root_pids,
-            &root_pgids,
             transport,
             reap_scope,
             &mut probe_degraded,
@@ -1166,7 +1204,7 @@ pub mod lifecycle_port {
         if team.is_none() {
             deadline.check("shared_socket_cleanup")?;
             let event_log = crate::event_log::EventLog::new(&run_workspace);
-            let cleanup = bare_shutdown_socket_cleanup(transport, &state, &event_log);
+            let cleanup = bare_shutdown_socket_cleanup(&run_workspace, transport, &state, &event_log);
             for session in cleanup.killed_sessions {
                 push_unique_session(&mut killed_sessions, session);
             }
@@ -1175,6 +1213,7 @@ pub mod lifecycle_port {
                 kill_error.get_or_insert(error);
             }
         }
+        spared_sessions.retain(|session| transport.has_session(session).unwrap_or(true));
         deadline.check("session_residuals")?;
         let (session_residuals, session_residual_error) = session_residuals_after_reap_many(
             transport,
@@ -1208,7 +1247,6 @@ pub mod lifecycle_port {
             &run_workspace,
             &state,
             &root_pids,
-            &root_pgids,
             &protected,
             reap_scope,
             &verify_table,
@@ -1838,12 +1876,25 @@ pub mod lifecycle_port {
     /// pids; Gap 37 escalation order TERM -> grace -> KILL preserved), then a single
     /// bounded wait for the whole union. kill/wait sets derive from the SAME snapshot
     /// as the protected set (N39).
-    fn reap_process_tree(root_pids: &[u32], protected: &ShutdownProtection, table: &[ProcessInfo]) {
+    fn reap_process_tree(
+        workspace: &Path,
+        state: &Value,
+        scope: ShutdownReapScope,
+        root_pids: &[u32],
+        protected: &ShutdownProtection,
+        table: &[ProcessInfo],
+        event_log: &crate::event_log::EventLog,
+    ) {
         let mut pids = Vec::new();
         let mut seen = std::collections::BTreeSet::new();
+        let spawn_cwds = state_spawn_cwds(state, scope);
         for root in root_pids {
             for pid in process_tree_from_table(*root, table) {
-                if !protected.contains_pid(pid) && seen.insert(pid) {
+                let owned = table
+                    .iter()
+                    .find(|process| process.pid == pid)
+                    .is_some_and(|process| process_matches_workspace(process, workspace, &spawn_cwds));
+                if owned && !protected.contains_pid(pid) && seen.insert(pid) {
                     pids.push(pid);
                 }
             }
@@ -1851,50 +1902,58 @@ pub mod lifecycle_port {
         if pids.is_empty() {
             return;
         }
-        for pid in pids.iter().rev() {
-            let _ = crate::platform::process::terminate_pid(
-                *pid,
-                crate::platform::process::SignalKind::TerminateGraceful,
-            );
+        let audit_targets = pids.iter().map(u32::to_string).collect::<Vec<_>>();
+        if event_log
+            .write(
+                "transport.pre_kill_audit",
+                json!({
+                    "phase": "pre_call",
+                    "caller": "shutdown.process_tree",
+                    "caller_pid": std::process::id(),
+                    "action": "signal-pid",
+                    "endpoint": Value::Null,
+                    "targets": audit_targets,
+                    "workspace": workspace.to_string_lossy(),
+                    "team": state.get("team_key").and_then(Value::as_str),
+                    "generation": state_generation(state),
+                    "owner_evidence": "argv+cwd+state_pid_tree",
+                    "reason": "positive per-pid process ownership",
+                }),
+            )
+            .is_err()
+        {
+            return;
         }
-        std::thread::sleep(std::time::Duration::from_millis(150));
         for pid in pids.iter().rev() {
             let _ = crate::platform::process::terminate_pid(
                 *pid,
-                crate::platform::process::SignalKind::TerminateForce,
-            );
-        }
-        wait_for_processes_gone(&pids, std::time::Duration::from_secs(1));
-    }
-
-    fn reap_process_groups(pgids: &[u32], protected: &ShutdownProtection) {
-        // 0.5.x Windows portability Batch 3: routes group termination
-        // through `platform::process::terminate_group`. Unix keeps
-        // `kill(-pgid, SIGTERM|SIGKILL)` semantics byte-for-byte;
-        // Windows returns AlreadyGone (no pgid concept — Job Object
-        // teardown is the shim-side concern per design §Route B).
-        // The `pgid_t <= 1 || protected.contains_pgid(...)` filter
-        // stays in place so pgid 0/1 (kernel/init) and the caller's
-        // own group are never targeted.
-        for pgid in pgids {
-            if *pgid <= 1 || protected.contains_pgid(*pgid) {
-                continue;
-            }
-            let _ = crate::platform::process::terminate_group(
-                *pgid,
                 crate::platform::process::SignalKind::TerminateGraceful,
             );
         }
         std::thread::sleep(std::time::Duration::from_millis(150));
-        for pgid in pgids {
-            if *pgid <= 1 || protected.contains_pgid(*pgid) {
-                continue;
+        // Re-probe identity before escalation.  If the probe fails, refuse the
+        // force signal rather than risking a recycled PID.
+        if let Ok(live_table) = probed_process_table() {
+            let spawn_cwds = state_spawn_cwds(state, scope);
+            for pid in pids.iter().rev() {
+                let Some(before) = table.iter().find(|process| process.pid == *pid) else {
+                    continue;
+                };
+                let Some(after) = live_table.iter().find(|process| process.pid == *pid) else {
+                    continue;
+                };
+                if before.ppid == after.ppid
+                    && before.command == after.command
+                    && process_matches_workspace(after, workspace, &spawn_cwds)
+                {
+                    let _ = crate::platform::process::terminate_pid(
+                        *pid,
+                        crate::platform::process::SignalKind::TerminateForce,
+                    );
+                }
             }
-            let _ = crate::platform::process::terminate_group(
-                *pgid,
-                crate::platform::process::SignalKind::TerminateForce,
-            );
         }
+        wait_for_processes_gone(&pids, std::time::Duration::from_secs(1));
     }
 
     /// PERF-6 C-①-2 + C-②-5: every residual round fetches ONE fresh snapshot (reap
@@ -1904,7 +1963,6 @@ pub mod lifecycle_port {
         workspace: &Path,
         state: &Value,
         root_pids: &[u32],
-        root_pgids: &[u32],
         transport: &dyn crate::transport::Transport,
         scope: ShutdownReapScope,
         probe_degraded: &mut bool,
@@ -1917,7 +1975,6 @@ pub mod lifecycle_port {
                 workspace,
                 state,
                 root_pids,
-                root_pgids,
                 &protected,
                 scope,
                 &round_table,
@@ -1929,12 +1986,15 @@ pub mod lifecycle_port {
                 .iter()
                 .map(|process| process.pid)
                 .collect::<Vec<_>>();
-            reap_process_tree(&residual_pids, &protected, &round_table);
-            let pgids = residuals
-                .iter()
-                .filter_map(|process| process.pgid)
-                .collect::<Vec<_>>();
-            reap_process_groups(&pgids, &protected);
+            reap_process_tree(
+                workspace,
+                state,
+                scope,
+                &residual_pids,
+                &protected,
+                &round_table,
+                &crate::event_log::EventLog::new(workspace),
+            );
             std::thread::sleep(std::time::Duration::from_millis(100));
         }
     }
@@ -2045,12 +2105,13 @@ pub mod lifecycle_port {
     ///   → 任一 team 的 shutdown 都不杀任何 team 的 leader 锚 pane
     pub fn collect_state_leader_anchor_pane_ids(
         state: &Value,
+        endpoint: Option<&str>,
     ) -> std::collections::BTreeSet<String> {
         let mut out = std::collections::BTreeSet::new();
-        push_anchor_pane_id(state, &mut out);
+        push_anchor_pane_id(state, endpoint, &mut out);
         if let Some(teams) = state.get("teams").and_then(Value::as_object) {
             for (_, team_state) in teams {
-                push_anchor_pane_id(team_state, &mut out);
+                push_anchor_pane_id(team_state, endpoint, &mut out);
             }
         }
         out
@@ -2116,9 +2177,9 @@ pub mod lifecycle_port {
                 if leader_anchor_ids.contains(pane_id) {
                     continue;
                 }
-                let in_live = live_targets
-                    .iter()
-                    .any(|info| info.pane_id.as_str() == pane_id);
+                let in_live = live_targets.iter().any(|info| {
+                    info.pane_id.as_str() == pane_id && info.session.as_str() == session
+                });
                 if in_live && seen.insert(pane_id.to_string()) {
                     out.push(crate::transport::PaneId::new(pane_id));
                     continue;
@@ -2170,17 +2231,40 @@ pub mod lifecycle_port {
         Ok(())
     }
 
-    /// 单帧扫 team_owner.pane_id + leader_receiver.pane_id → BTreeSet 累加。
-    fn push_anchor_pane_id(state: &Value, out: &mut std::collections::BTreeSet<String>) {
-        for key in &["team_owner", "leader_receiver"] {
-            if let Some(pane_id) = state
-                .get(*key)
-                .and_then(|v| v.get("pane_id"))
-                .and_then(Value::as_str)
+    /// Pane IDs are server-local. Exclude proven foreign-endpoint anchors, but
+    /// retain legacy anchors when their endpoint (or the transport's) is unknown.
+    fn push_anchor_pane_id(
+        state: &Value,
+        endpoint: Option<&str>,
+        out: &mut std::collections::BTreeSet<String>,
+    ) {
+        for (key, peer_key) in [("team_owner", "leader_receiver"), ("leader_receiver", "team_owner")] {
+            let Some(anchor) = state.get(key) else { continue };
+            let Some(pane_id) = anchor.get("pane_id").and_then(Value::as_str).filter(|s| !s.is_empty()) else {
+                continue;
+            };
+            // team_owner often records only a pane ID; its receiver supplies
+            // the endpoint only when it describes that exact same anchor.
+            let anchor_endpoint = anchor.get("tmux_socket").and_then(Value::as_str)
                 .filter(|s| !s.is_empty())
-            {
-                out.insert(pane_id.to_string());
+                .or_else(|| {
+                    state.get(peer_key)
+                        .filter(|peer| peer.get("pane_id").and_then(Value::as_str) == Some(pane_id))
+                        .and_then(|peer| peer.get("tmux_socket"))
+                        .and_then(Value::as_str)
+                        .filter(|s| !s.is_empty())
+                });
+            if let (Some(anchor_path), Some(transport_path)) = (
+                anchor_endpoint.and_then(socket_file_for_endpoint),
+                endpoint.filter(|s| !s.is_empty()).and_then(socket_file_for_endpoint),
+            ) {
+                let anchor_path = anchor_path.canonicalize().unwrap_or(anchor_path);
+                let transport_path = transport_path.canonicalize().unwrap_or(transport_path);
+                if anchor_path != transport_path {
+                    continue;
+                }
             }
+            out.insert(pane_id.to_string());
         }
     }
 
@@ -2291,8 +2375,10 @@ pub mod lifecycle_port {
         );
         // Source 2: state.json team_owner / leader_receiver 真锚 pane_id(top-level +
         // teams[*]),per-workspace socket 命中。
-        let anchor_pane_ids: std::collections::BTreeSet<String> =
-            collect_state_leader_anchor_pane_ids(state);
+        let anchor_pane_ids = collect_state_leader_anchor_pane_ids(
+            state,
+            transport.tmux_endpoint().as_deref(),
+        );
         leader_pane_pids.extend(
             pane_targets
                 .iter()
@@ -2308,6 +2394,7 @@ pub mod lifecycle_port {
         // 的 list_targets 找 anchor pane_id → pane_pid → 进入 process_tree 保护。
         // 不在 state 中的 socket 不查(MUST-17 不撒宽 / 不主动枚举全机器 sockets)。
         for socket_endpoint in collect_state_recorded_tmux_sockets(state) {
+            let anchor_pane_ids = collect_state_leader_anchor_pane_ids(state, Some(&socket_endpoint));
             let cross_backend =
                 crate::tmux_backend::TmuxBackend::for_tmux_endpoint(&socket_endpoint);
             let cross_panes =
@@ -2386,44 +2473,16 @@ pub mod lifecycle_port {
         }
     }
 
-    fn process_pgids(
-        pids: &[u32],
-        protected: &ShutdownProtection,
-        table: &[ProcessInfo],
-    ) -> Vec<u32> {
-        let mut pgids = pids
-            .iter()
-            .filter_map(|pid| table.iter().find(|process| process.pid == *pid))
-            .filter_map(|process| process.pgid)
-            .filter(|pgid| {
-                // 0.5.x Windows portability Batch 4: `libc::pid_t` was
-                // used here as a signed-integer conversion gate to
-                // reject values > INT_MAX. Replace with the equivalent
-                // `i32::try_from` (pgid_t is `c_int` on every Unix we
-                // support). Windows has no pgid concept so `pgids`
-                // is empty in practice — the filter is dead code on
-                // Windows but must still compile.
-                i32::try_from(*pgid)
-                    .map(|pgid_int| pgid_int > 1 && !protected.contains_pgid(*pgid))
-                    .unwrap_or(false)
-            })
-            .collect::<Vec<_>>();
-        pgids.sort_unstable();
-        pgids.dedup();
-        pgids
-    }
-
     fn process_residuals(
         workspace: &Path,
         state: &Value,
         root_pids: &[u32],
-        root_pgids: &[u32],
         protected: &ShutdownProtection,
         scope: ShutdownReapScope,
         table: &[ProcessInfo],
     ) -> Vec<Value> {
         let mut residuals = matched_processes(
-            workspace, state, root_pids, root_pgids, protected, scope, table,
+            workspace, state, root_pids, protected, scope, table,
         );
         let mut seen = residuals
             .iter()
@@ -2461,7 +2520,6 @@ pub mod lifecycle_port {
         workspace: &Path,
         state: &Value,
         root_pids: &[u32],
-        root_pgids: &[u32],
         protected: &ShutdownProtection,
         scope: ShutdownReapScope,
         table: &[ProcessInfo],
@@ -2471,33 +2529,14 @@ pub mod lifecycle_port {
             .flat_map(|pid| process_tree_from_table(*pid, table))
             .filter(|pid| !protected.contains_pid(*pid))
             .collect::<std::collections::BTreeSet<_>>();
-        let root_pgids = root_pgids
-            .iter()
-            .copied()
-            .collect::<std::collections::BTreeSet<_>>();
         let spawn_cwds = state_spawn_cwds(state, scope);
-        let workspace_text = workspace.to_string_lossy().to_string();
-        let mut cwd_probe_budget = 3_usize;
-        let mut out = Vec::new();
-        for process in table {
-            if protected.contains_pid(process.pid) {
-                continue;
-            }
-            let matches_workspace = scope == ShutdownReapScope::Workspace
-                && process_matches_workspace(
-                    process,
-                    &workspace_text,
-                    &spawn_cwds,
-                    &mut cwd_probe_budget,
-                );
-            if matches_workspace
-                || root_tree.contains(&process.pid)
-                || process.pgid.is_some_and(|pgid| root_pgids.contains(&pgid))
-            {
-                out.push(process.clone());
-            }
-        }
-        out
+        table
+            .iter()
+            .filter(|process| !protected.contains_pid(process.pid))
+            .filter(|process| root_tree.contains(&process.pid))
+            .filter(|process| process_matches_workspace(process, workspace, &spawn_cwds))
+            .cloned()
+            .collect()
     }
 
     fn process_tree_from_table(root_pid: u32, table: &[ProcessInfo]) -> Vec<u32> {
@@ -2550,59 +2589,21 @@ pub mod lifecycle_port {
 
     fn process_matches_workspace(
         process: &ProcessInfo,
-        workspace_text: &str,
-        spawn_cwds: &[PathBuf],
-        cwd_probe_budget: &mut usize,
+        workspace: &Path,
+        _spawn_cwds: &[PathBuf],
     ) -> bool {
-        let command = process.command.as_str();
-        if command.contains("mcp-server")
-            && command.contains("--workspace")
-            && command.contains(workspace_text)
-        {
-            return true;
-        }
-        if command.contains(workspace_text) {
-            return true;
-        }
-        if spawn_cwds.is_empty() || *cwd_probe_budget == 0 {
-            return false;
-        }
-        *cwd_probe_budget -= 1;
-        let Some(cwd) = process_cwd(process.pid) else {
-            return false;
-        };
-        spawn_cwds
-            .iter()
-            .any(|spawn_cwd| path_is_under(&cwd, spawn_cwd))
-    }
-
-    fn process_cwd(pid: u32) -> Option<PathBuf> {
-        let proc_cwd = PathBuf::from(format!("/proc/{pid}/cwd"));
-        if let Ok(path) = std::fs::read_link(proc_cwd) {
-            return Some(path);
-        }
-        if crate::os_probe::probe_timed_out() {
-            return None;
-        }
-        let output = crate::os_probe::bounded_command_output_with_probe(
-            std::process::Command::new("lsof").args([
-                "-a",
-                "-p",
-                &pid.to_string(),
-                "-d",
-                "cwd",
-                "-Fn",
-            ]),
-            "lsof_cwd",
-            Some(pid),
-        )
-        .ok()?;
-        if !output.status.success() {
-            return None;
-        }
-        String::from_utf8_lossy(&output.stdout)
-            .lines()
-            .find_map(|line| line.strip_prefix('n').map(PathBuf::from))
+        let workspace = workspace
+            .canonicalize()
+            .unwrap_or_else(|_| workspace.to_path_buf());
+        crate::platform::argv::argv_tokens(process.pid).is_some_and(|tokens| {
+            tokens.windows(2).any(|pair| {
+                pair[0] == "--workspace"
+                    && Path::new(&pair[1])
+                        .canonicalize()
+                        .map(|path| path == workspace)
+                        .unwrap_or(false)
+            })
+        })
     }
 
     fn path_is_under(path: &Path, root: &Path) -> bool {
@@ -4488,6 +4489,7 @@ pub mod lifecycle_port {
         match status {
             crate::coordinator::StopOutcome::Missing => "missing",
             crate::coordinator::StopOutcome::InvalidPidRemoved => "invalid_pid_removed",
+            crate::coordinator::StopOutcome::NotOwned => "not_owned",
             crate::coordinator::StopOutcome::KillFailed => "kill_failed",
             crate::coordinator::StopOutcome::Stopped => "stopped",
         }
@@ -4823,17 +4825,15 @@ pub mod diagnose_port {
             && profile_smoke_value.get("status").and_then(Value::as_str)
                 == Some("legacy_team_invalid");
         let effective_smoke_ok = profile_smoke_ok || legacy_only_failure;
+        // Coordinator and runtime topology are facts of the unified pipeline,
+        // not of the static environment probe.  The caller supplies the
+        // selected runtime workspace and performs one read-only health read.
         let ok = workspace_valid && (team_context || workspace_has_entries) && effective_smoke_ok;
-        let health = crate::coordinator::coordinator_health(
-            &crate::coordinator::WorkspacePath::new(workspace.to_path_buf()),
-        );
-        let coordinator_ok = health.ok;
-        let coordinator_error = health.metadata_mismatch_reason.clone();
         let state = crate::state::persist::load_runtime_state(workspace).unwrap_or(Value::Null);
         let grok_slot = crate::cli::grok_slot_report(workspace, &state).to_json();
         let grok_slot_ok = grok_slot.get("consistent").and_then(Value::as_bool) == Some(true)
             && grok_slot.get("readable").and_then(Value::as_bool) == Some(true);
-        let ok = ok && grok_slot_ok && coordinator_ok;
+        let ok = ok && grok_slot_ok;
         Ok(json!({
             "tmux": {
                 "installed": tmux_installed,
@@ -4841,14 +4841,14 @@ pub mod diagnose_port {
             },
             "workspace": workspace.to_string_lossy().to_string(),
             "workspace_is_git_repo": workspace.join(".git").exists(),
-            "providers": {},
+            "providers": crate::cli::diagnose::provider_doctor_checks(),
             "mcp": {
                 "server_command": which_path("team_orchestrator"),
                 "local_module": true,
             },
             "secret_scan": secret_scan(workspace),
             "profile_smoke": profile_smoke_value,
-            "coordinator": coordinator_health_value(health),
+            "coordinator": coordinator_not_required_value(),
             "grok_slot": grok_slot,
             "ok": ok,
             "error": if ok {
@@ -4860,10 +4860,6 @@ pub mod diagnose_port {
                     .unwrap_or_else(|| json!("grok_slot_mismatch"))
             } else if !profile_smoke_ok && !legacy_only_failure {
                 json!("profile_smoke_failed")
-            } else if !coordinator_ok {
-                coordinator_error
-                    .map(Value::String)
-                    .unwrap_or_else(|| json!("coordinator_unavailable"))
             } else if workspace_valid {
                 json!("workspace has no Team Agent spec or runtime context")
             } else {
@@ -5055,11 +5051,12 @@ pub mod diagnose_port {
                 continue;
             };
             let rel = path.strip_prefix(root).unwrap_or(path);
+            // Keep only a locating rule/path/line tuple.  Never include even
+            // redacted assignment text in the structured report.
             findings.push(json!({
                 "path": rel.to_string_lossy().to_string(),
                 "line": idx.saturating_add(1),
                 "rule": "api_key_assignment",
-                "match_excerpt": format!("{key}=<redacted>"),
             }));
         }
     }
@@ -5224,7 +5221,23 @@ pub mod diagnose_port {
         })
     }
 
-    fn coordinator_health_value(health: crate::coordinator::HealthReport) -> Value {
+    pub(crate) fn coordinator_not_required_value() -> Value {
+        json!({
+            "ok": true,
+            "status": "not_required",
+            "pid": Value::Null,
+            "metadata": Value::Null,
+            "metadata_ok": true,
+            "metadata_mismatch_reason": Value::Null,
+            "binary_path": Value::Null,
+            "binary_version": Value::Null,
+            "schema_ok": true,
+            "schema_error": Value::Null,
+            "schema": {"message_store_schema_version": Value::Null},
+        })
+    }
+
+    pub(crate) fn coordinator_health_value(health: crate::coordinator::HealthReport) -> Value {
         let expose_binary_drift = health.service_available && !health.metadata_ok;
         let binary_identity_relation = health.binary_identity_relation.as_str();
         let mut value = json!({