@team-agent/installer 0.2.11 → 0.3.1

package/crates/team-agent/src/lifecycle/tests/main_preserved.rs

@@ -0,0 +1,265 @@
+use super::*;
+use super::agent_ops::lanea_team_ws;
+use super::lane_ops::{LaneSpawns, LaneTransport};
+use super::launch_spawn::{
+    restart_ws_two_resumable_workers, seed_healthy_coordinator, DELEG_ROLE_ALPHA, DELEG_ROLE_BRAVO,
+};
+
+type RecordedSpawns = LaneSpawns;
+
+// ═════════════════════════════════════════════════════════════════════════
+// BUG-1 [highest · regression] — respawn into a DEAD tmux session/`-L` server
+// must RE-CREATE the session (new-session / spawn_first), NOT add a window to a
+// server that no longer exists (new-window / spawn_into → "no server running" →
+// orphaned agent). `reset --discard-session` kills the agent's last window; tmux
+// then destroys the whole session AND its CP-1 `-L` server. The respawn that
+// follows must therefore new-session.
+//
+// Golden runtime.py:1017 `_tmux_start_command_for_agent_window`:
+//     if _tmux_session_exists(session_name):  # live `tmux has-session -t <name>`
+//         return ["tmux","new-window",...], "new-window"
+//     return ["tmux","new-session","-d",...], "new-session"
+// The new-session-vs-new-window choice is driven by a LIVE `has-session` probe,
+// NOT by whether state["session_name"] is a non-empty string.
+//
+// Today `start_agent_with_transport` computes `into_existing_session =
+// session_name_present(&state)` (a pure state-string check; restart.rs:129) and
+// NEVER consults `transport.has_session(...)` — so a discarded/dead session
+// (state string still set) takes the spawn_into branch. The socket fix
+// (0f886a9/4435f9a) covered status/shutdown but missed this lifecycle respawn path.
+// ═════════════════════════════════════════════════════════════════════════
+/// Recording transport with a CONFIGURABLE `has_session` — models the real
+/// `tmux has-session` probe the respawn spawn-decision MUST consult. Records
+/// every spawn (kind + argv) like `RecordingLaunchTransport`.
+struct SessionProbeRecordingTransport {
+    spawns: RecordedSpawns,
+    session_exists: bool,
+}
+impl crate::transport::Transport for SessionProbeRecordingTransport {
+    fn kind(&self) -> crate::transport::BackendKind {
+        crate::transport::BackendKind::Tmux
+    }
+    fn spawn_first(&self, session: &crate::transport::SessionName, window: &crate::transport::WindowName, argv: &[String], _cwd: &std::path::Path, _env: &std::collections::BTreeMap<String, String>) -> Result<crate::transport::SpawnResult, crate::transport::TransportError> {
+        self.spawns.lock().unwrap().push(("spawn_first".to_string(), argv.to_vec()));
+        Ok(crate::transport::SpawnResult { pane_id: crate::transport::PaneId::new("%0"), session: session.clone(), window: window.clone(), child_pid: None })
+    }
+    fn spawn_into(&self, session: &crate::transport::SessionName, window: &crate::transport::WindowName, argv: &[String], _cwd: &std::path::Path, _env: &std::collections::BTreeMap<String, String>) -> Result<crate::transport::SpawnResult, crate::transport::TransportError> {
+        self.spawns.lock().unwrap().push(("spawn_into".to_string(), argv.to_vec()));
+        Ok(crate::transport::SpawnResult { pane_id: crate::transport::PaneId::new(format!("%{}", window.as_str())), session: session.clone(), window: window.clone(), child_pid: None })
+    }
+    fn inject(&self, _t: &crate::transport::Target, _p: &crate::transport::InjectPayload, _s: crate::transport::Key, _b: bool) -> Result<crate::transport::InjectReport, crate::transport::TransportError> {
+        unimplemented!("not reached by start_agent respawn")
+    }
+    fn send_keys(&self, _t: &crate::transport::Target, _k: &[crate::transport::Key]) -> Result<(), crate::transport::TransportError> {
+        unimplemented!("not reached")
+    }
+    fn capture(&self, _t: &crate::transport::Target, _r: crate::transport::CaptureRange) -> Result<crate::transport::CapturedText, crate::transport::TransportError> {
+        unimplemented!("not reached")
+    }
+    fn query(&self, _t: &crate::transport::Target, _f: crate::transport::PaneField) -> Result<Option<String>, crate::transport::TransportError> {
+        unimplemented!("not reached")
+    }
+    fn liveness(&self, _p: &crate::transport::PaneId) -> Result<crate::model::enums::PaneLiveness, crate::transport::TransportError> {
+        unimplemented!("not reached")
+    }
+    fn list_targets(&self) -> Result<Vec<crate::transport::PaneInfo>, crate::transport::TransportError> {
+        unimplemented!("not reached")
+    }
+    fn has_session(&self, _s: &crate::transport::SessionName) -> Result<bool, crate::transport::TransportError> {
+        Ok(self.session_exists)
+    }
+    fn list_windows(&self, _s: &crate::transport::SessionName) -> Result<Vec<crate::transport::WindowName>, crate::transport::TransportError> {
+        // a dead server has no windows; a live one has the agent window.
+        Ok(if self.session_exists { vec![crate::transport::WindowName::new("alpha")] } else { Vec::new() })
+    }
+    fn set_session_env(&self, _s: &crate::transport::SessionName, _k: &str, _v: &str) -> Result<crate::transport::SetEnvOutcome, crate::transport::TransportError> {
+        unimplemented!("not reached")
+    }
+    fn kill_session(&self, _s: &crate::transport::SessionName) -> Result<(), crate::transport::TransportError> {
+        unimplemented!("not reached")
+    }
+    fn kill_window(&self, _t: &crate::transport::Target) -> Result<(), crate::transport::TransportError> {
+        Ok(())
+    }
+    fn attach_session(&self, _s: &crate::transport::SessionName) -> Result<crate::transport::AttachOutcome, crate::transport::TransportError> {
+        unimplemented!("not reached")
+    }
+}
+fn respawn_ws_one_resumable_worker() -> PathBuf {
+    let ws = temp_ws().join("respawn_dead_session");
+    std::fs::create_dir_all(&ws).unwrap();
+    crate::state::persist::save_runtime_state(
+        &ws,
+        &json!({
+            "session_name": "team-sa",
+            "agents": {"alpha": {"status": "running", "provider": "codex", "session_id": "sess-a", "first_send_at": "2026-05-27T10:00:00+00:00"}}
+        }),
+    )
+    .unwrap();
+    seed_healthy_coordinator(&ws);
+    ws
+}
+// RED — the respawn-time spawn decision must consult the LIVE session probe. With
+// `has_session=false` (the `-L` server destroyed by --discard-session killing the
+// last window), the respawn MUST new-session (spawn_first). Today `into_existing_session
+// = session_name_present(&state)` is true (state string still "team-sa") so the code
+// records spawn_into against a dead server -> assertion fails -> RED. NOT a panic:
+// the transport implements has_session, so the only failure is the wrong spawn kind.
+#[test]
+fn start_agent_respawn_into_dead_session_uses_new_session_not_new_window() {
+    let ws = respawn_ws_one_resumable_worker();
+    let spawns = std::sync::Arc::new(std::sync::Mutex::new(Vec::new()));
+    // session_exists=false models the dead CP-1 `-L` server after --discard-session.
+    let transport = SessionProbeRecordingTransport {
+        spawns: std::sync::Arc::clone(&spawns),
+        session_exists: false,
+    };
+    let _ = start_agent_with_transport(&ws, &AgentId::new("alpha"), false, false, false, None, &transport);
+    let recorded = spawns.lock().unwrap().clone();
+    assert_eq!(recorded.len(), 1, "exactly one respawn for alpha; got {recorded:?}");
+    assert_eq!(
+        recorded[0].0, "spawn_first",
+        "golden runtime.py:1017: when `tmux has-session` is FALSE (dead -L server after \
+         --discard-session destroyed the last window), the respawn MUST new-session \
+         (spawn_first), NOT new-window (spawn_into) — spawn_into a dead server crashes with \
+         'no server running' and orphans the agent. The decision must come from \
+         transport.has_session(), not session_name_present(&state); got {recorded:?}"
+    );
+}
+// ═════════════════════════════════════════════════════════════════════════
+// reset/remove CANONICAL-WORKSPACE regression-locks (bug-A class: e17096c/0f886a9/4435f9a).
+// rt-host-a @ c262bf7: `reset`/`remove --workspace <teamdir/.team>` resolved to the SUBPATH's own
+// .team/runtime instead of the LIVE parent run-workspace -> they operated on a detached context
+// (live roster/windows untouched; respawn went nowhere -> post-reset dispatch died). `start` dodged it
+// by being invoked with $ws. FIX (porter-a): reset/remove canonicalize the run-workspace
+// (lifecycle_paths -> run_workspace = canonical_run_workspace(input)), read spec from the teamdir, and
+// reset reuses start's live-session-probe respawn. These LOCK the canonical resolution so it can't
+// re-regress.
+// ═════════════════════════════════════════════════════════════════════════
+// item ① — reset --discard-session must REBUILD the worker's window via the SAME respawn path as start
+// (reset_agent_at_paths -> start_agent_with_transport). Into a LIVE session (has_session=true) the
+// rebuild is a new-window (spawn_into). The regression had reset operate on a detached runtime so the
+// window was never rebuilt. OS-safe: SessionProbeRecordingTransport (no real tmux) + seeded coordinator.
+#[test]
+fn reset_agent_discard_session_rebuilds_window_via_start_respawn() {
+    let ws = restart_ws_two_resumable_workers(); // compiled spec + state(alpha,bravo running) + seeded coordinator
+    let spawns = std::sync::Arc::new(std::sync::Mutex::new(Vec::new()));
+    // live session present (other workers keep it alive) but alpha's window was discarded.
+    let transport = SessionProbeRecordingTransport {
+        spawns: std::sync::Arc::clone(&spawns),
+        session_exists: true,
+    };
+    let _ = reset_agent_with_transport(&ws, &AgentId::new("alpha"), true, false, None, &transport);
+    let recorded = spawns.lock().unwrap().clone();
+    assert!(
+        !recorded.is_empty(),
+        "reset --discard-session must RESPAWN alpha's window (reuse start's respawn path); the bug-A \
+         regression operated on a detached runtime so nothing was rebuilt -> ZERO spawns. got {recorded:?}"
+    );
+    assert_eq!(
+        recorded.last().unwrap().0, "spawn_into",
+        "into a LIVE session (has_session=true) the window rebuild is a new-window (spawn_into); got {recorded:?}"
+    );
+}
+// item ② — remove invoked with a NON-canonical input (the workspace's own `.team` subpath) must resolve
+// the CANONICAL live run-workspace and apply the removal to the LIVE roster — not the subpath's detached
+// runtime. canonical_run_workspace($ws/.team) == $ws. The regression made remove "succeed" against a
+// detached context, leaving the live roster [alpha,bravo] intact. (lanea_team_ws re-points routing to
+// bravo so removing alpha validates cleanly.)
+#[test]
+fn remove_agent_via_team_subpath_applies_to_canonical_live_roster() {
+    let ws = lanea_team_ws("stopped"); // canonical workspace: spec + state roster [alpha,bravo]
+    let subpath = ws.join(".team"); // non-canonical input; canonicalizes back to $ws
+    let tx = LaneTransport::new("team-laneateam", &[]);
+    let outcome = remove_agent_with_transport(&subpath, &aid("alpha"), true, true, None, &tx);
+    assert!(
+        matches!(outcome, Ok(RemoveAgentOutcome::Removed { .. })),
+        "remove via the .team subpath must resolve the canonical live workspace and succeed; got {outcome:?}"
+    );
+    let state = crate::state::persist::load_runtime_state(&ws).expect("load LIVE $ws state");
+    assert!(
+        state.get("agents").and_then(serde_json::Value::as_object).is_some_and(|a| !a.contains_key("alpha")),
+        "remove must apply to the LIVE $ws roster (canonical resolution), removing alpha; the bug-A \
+         regression operated on the subpath's detached runtime, leaving the live roster intact. state={state}"
+    );
+}
+/// bug-B [✦ DIVERGENCE] fixture — a NORMAL compiled 2-agent team with routing/tasks LEFT EXACTLY AS
+/// COMPILED (route-alpha + route-bravo, default_assignee=alpha, tasks[0].assignee=alpha). Unlike
+/// lanea_team_ws / lanea_ws_agents (which deliberately re-point refs away so a remove validates), this
+/// is the REAL "remove a normal worker" case — the one golden cannot handle (every compiled agent is
+/// routed, compiler.py:57). Both workers seeded `stopped` (no tmux); LaneTransport reports no windows.
+fn bugb_routed_team_ws() -> PathBuf {
+    let ws = temp_ws().join("bugb_routed");
+    std::fs::create_dir_all(ws.join("agents")).unwrap();
+    std::fs::write(ws.join("TEAM.md"), "---\nname: laneateam\nobjective: bug-B routed-remove probe.\nprovider: codex\n---\n\nteam.\n").unwrap();
+    std::fs::write(ws.join("agents").join("alpha.md"), DELEG_ROLE_ALPHA).unwrap();
+    std::fs::write(ws.join("agents").join("bravo.md"), DELEG_ROLE_BRAVO).unwrap();
+    let spec = crate::compiler::compile_team(&ws).expect("compile bug-B team");
+    let yaml = crate::model::yaml::dumps(&spec);
+    // sanity: the UNMODIFIED compiled spec really routes bravo — that route-bravo rule is the dangling
+    // ref the cascade-prune must drop (and which makes golden's validate_spec raise + rollback).
+    assert!(yaml.contains("assign_to: \"bravo\""), "fixture: compiled spec must carry route-bravo (assign_to bravo); got:\n{yaml}");
+    std::fs::write(ws.join("team.spec.yaml"), yaml).unwrap();
+    crate::state::persist::save_runtime_state(
+        &ws,
+        &json!({
+            "session_name": "team-laneateam",
+            "agents": {
+                "alpha": {"status": "stopped", "provider": "codex", "window": "alpha"},
+                "bravo": {"status": "stopped", "provider": "codex", "window": "bravo"}
+            }
+        }),
+    )
+    .unwrap();
+    ws
+}
+// ── bug-B [✦ DIVERGENCE — REVERSES the previously-DECLINED byte-parity decision (ticket:154)] ─────────
+// Removing a NORMAL routed worker must SUCCEED and surgically PRUNE that worker's dangling routing rule,
+// NOT refuse+rollback. Golden agents.py:88-94 removes the agent only from `agents` + `runtime.startup_order`
+// (routing UNTOUCHED), then validate_spec RAISES `/routing/rules/N/assign_to: unknown agent 'bravo'`
+// (spec.py:341-346) -> remove catches -> rollback (rollback_ok=true). Because compile_team auto-wires a
+// `route-<id>` rule (assign_to:<id>) for EVERY agent (compiler.py:57), NO compiled worker is removable in
+// golden. Probed live (/tmp/probe_removeB2.py): validate_spec of the bravo-removed spec raises exactly
+// "/routing/rules/1/assign_to: unknown agent 'bravo'". The advisor/leader requests the cascade-prune
+// parity-IMPROVEMENT (ticket:155,197): remove must drop routing rules whose assign_to == the removed
+// agent (and reset default_assignee if it pointed at it) so the spec stays valid. Current Rust faithfully
+// mirrors golden (restart.rs:897-899 validate_spec -> Err(Compile) -> rollback) -> RED.
+//
+// NOTE for the porter: this REVERSES the documented byte-parity intent at restart.rs:931-937 and the
+// lanea_team_ws/lanea_ws_agents fixtures' re-point workaround. Implementing the prune is correct ONLY
+// if the leader's ✦ divergence ruling stands — do not "fix" by reverting this test.
+//
+// Removing `bravo` (the SECOND agent) isolates the routing-rule prune cleanly: only `route-bravo` dangles;
+// default_assignee (alpha) and tasks[0].assignee (alpha) stay valid, so this test does NOT entangle the
+// default_assignee / task-reassignment policy.
+#[test]
+fn remove_routed_worker_prunes_dangling_routing_rule_divergence() {
+    let ws = bugb_routed_team_ws();
+    let tx = LaneTransport::new("team-laneateam", &[]); // bravo not running (no windows)
+    let outcome = remove_agent_with_transport(&ws, &aid("bravo"), true, true, None, &tx); // from_spec + force
+    assert!(
+        matches!(outcome, Ok(RemoveAgentOutcome::Removed { .. })),
+        "✦ divergence: removing the routed worker 'bravo' must SUCCEED (Removed), not refuse+rollback. \
+         Golden + current Rust raise /routing/rules/.../assign_to: unknown agent 'bravo' -> \
+         StatePersist rollback_ok=true; got {outcome:?}"
+    );
+    let yaml = std::fs::read_to_string(ws.join("team.spec.yaml")).expect("read spec after remove");
+    assert!(
+        !yaml.contains("assign_to: \"bravo\""),
+        "✦ divergence: remove must PRUNE the dangling route-bravo rule (assign_to: \"bravo\"); spec still has it:\n{yaml}"
+    );
+    assert!(
+        yaml.contains("assign_to: \"alpha\""),
+        "the prune must be SURGICAL — route-alpha (the STAYING agent) must remain; got:\n{yaml}"
+    );
+    assert!(
+        yaml.contains("default_assignee: \"alpha\""),
+        "default_assignee (alpha, still present) must be untouched by the prune; got:\n{yaml}"
+    );
+    let state = crate::state::persist::load_runtime_state(&ws).expect("load state");
+    assert!(
+        state.get("agents").and_then(serde_json::Value::as_object).is_some_and(|a| !a.contains_key("bravo")),
+        "bravo must also be removed from state.agents; got {state:?}"
+    );
+    let _ = std::fs::remove_dir_all(&ws);
+}

package/crates/team-agent/src/lifecycle/tests.rs

@@ -0,0 +1,27 @@
+#![allow(clippy::unwrap_used, clippy::expect_used, clippy::panic)]
+use super::*;
+use serde_json::json;
+use std::sync::atomic::{AtomicU32, Ordering};
+
+static SEQ: AtomicU32 = AtomicU32::new(0);
+
+/// 隔离的空 workspace。仿 sibling(event_log.rs)风格,不依赖 tempfile dev-dep。
+fn temp_ws() -> PathBuf {
+    let n = SEQ.fetch_add(1, Ordering::Relaxed);
+    let ws = std::env::temp_dir().join(format!("ta_rs_lc_{}_{}", std::process::id(), n));
+    std::fs::create_dir_all(&ws).unwrap();
+    ws
+}
+
+fn aid(s: &str) -> AgentId {
+    AgentId::new(s)
+}
+fn sess(s: &str) -> SessionName {
+    SessionName(s.to_string())
+}
+
+mod core;
+mod launch_spawn;
+mod agent_ops;
+mod lane_ops;
+mod main_preserved;