@team-agent/installer 0.5.12 → 0.5.14

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/Cargo.lock CHANGED
@@ -575,7 +575,7 @@ dependencies = [
575
575
 
576
576
  [[package]]
577
577
  name = "team-agent"
578
- version = "0.5.12"
578
+ version = "0.5.14"
579
579
  dependencies = [
580
580
  "anyhow",
581
581
  "chrono",
package/Cargo.toml CHANGED
@@ -9,7 +9,7 @@ members = ["crates/team-agent", "crates/win-conpty-phase0", "crates/conpty-trans
9
9
 
10
10
  [workspace.package]
11
11
  edition = "2021"
12
- version = "0.5.12"
12
+ version = "0.5.14"
13
13
  license = "AGPL-3.0"
14
14
  rust-version = "1.95"
15
15
 
@@ -2187,7 +2187,7 @@ pub mod lifecycle_port {
2187
2187
  team,
2188
2188
  session_converge_deadline_ms,
2189
2189
  ) {
2190
- Ok(report) => Ok(restart_value(report)),
2190
+ Ok(report) => Ok(restart_value(report, team)),
2191
2191
  Err(e) => Ok(error_value(e)),
2192
2192
  }
2193
2193
  }
@@ -2899,11 +2899,14 @@ pub mod lifecycle_port {
2899
2899
 
2900
2900
  #[test]
2901
2901
  fn restart_json_includes_harness_reminder() {
2902
- let value = restart_value(crate::lifecycle::RestartReport::RefusedResumeAtomicity {
2903
- unresumable: Vec::new(),
2904
- allow_fresh: false,
2905
- error: "resume refused".to_string(),
2906
- });
2902
+ let value = restart_value(
2903
+ crate::lifecycle::RestartReport::RefusedResumeAtomicity {
2904
+ unresumable: Vec::new(),
2905
+ allow_fresh: false,
2906
+ error: "resume refused".to_string(),
2907
+ },
2908
+ None,
2909
+ );
2907
2910
 
2908
2911
  assert_eq!(
2909
2912
  value.get("reminder").and_then(Value::as_str),
@@ -2912,7 +2915,7 @@ pub mod lifecycle_port {
2912
2915
  }
2913
2916
  }
2914
2917
 
2915
- fn restart_value(report: crate::lifecycle::RestartReport) -> Value {
2918
+ fn restart_value(report: crate::lifecycle::RestartReport, team: Option<&str>) -> Value {
2916
2919
  match report {
2917
2920
  crate::lifecycle::RestartReport::Restarted {
2918
2921
  session_name,
@@ -3310,22 +3313,31 @@ pub mod lifecycle_port {
3310
3313
  reason,
3311
3314
  error,
3312
3315
  issue_ids,
3313
- } => json!({
3314
- "ok": false,
3315
- "status": "refused_dirty_topology",
3316
- "reason": reason,
3317
- "session_name": session_name,
3318
- "error": error,
3319
- "issues": issue_ids
3320
- .iter()
3321
- .map(|id| json!({"id": id}))
3322
- .collect::<Vec<_>>(),
3323
- "next_actions": [
3324
- "run team-agent diagnose --json from the intended leader socket",
3325
- "repair the tmux endpoint/socket split before retrying restart"
3326
- ],
3327
- "reminder": crate::cli::QUICK_START_REMINDER,
3328
- }),
3316
+ } => {
3317
+ let repair_team = team
3318
+ .filter(|team| !team.is_empty())
3319
+ .unwrap_or(session_name.as_str());
3320
+ let claim =
3321
+ format!("team-agent claim-leader --team {repair_team} --confirm --json");
3322
+ let takeover = format!("team-agent takeover --team {repair_team} --confirm --json");
3323
+ json!({
3324
+ "ok": false,
3325
+ "status": "refused_dirty_topology",
3326
+ "reason": reason,
3327
+ "session_name": session_name,
3328
+ "error": error,
3329
+ "issues": issue_ids
3330
+ .iter()
3331
+ .map(|id| json!({"id": id}))
3332
+ .collect::<Vec<_>>(),
3333
+ "next_actions": [
3334
+ "team-agent diagnose --json",
3335
+ claim,
3336
+ takeover
3337
+ ],
3338
+ "reminder": crate::cli::QUICK_START_REMINDER,
3339
+ })
3340
+ }
3329
3341
  }
3330
3342
  }
3331
3343
 
@@ -3960,6 +3972,7 @@ pub mod leader_port {
3960
3972
  .map_err(|e| CliError::Runtime(e.to_string()))?;
3961
3973
  let mut value = lease_value(result);
3962
3974
  if value.get("ok").and_then(Value::as_bool) == Some(true) {
3975
+ emit_topology_convergence_event(workspace, team, "takeover", &value);
3963
3976
  register_after_binding_success(workspace, team, "takeover", &mut value);
3964
3977
  }
3965
3978
  Ok(value)
@@ -3993,6 +4006,7 @@ pub mod leader_port {
3993
4006
  .map_err(|e| CliError::Runtime(e.to_string()))?;
3994
4007
  let mut value = lease_value(result);
3995
4008
  if value.get("ok").and_then(Value::as_bool) == Some(true) {
4009
+ emit_topology_convergence_event(workspace, team, "claim-leader", &value);
3996
4010
  register_after_binding_success(workspace, team, "claim-leader", &mut value);
3997
4011
  }
3998
4012
  Ok(value)
@@ -4122,6 +4136,41 @@ pub mod leader_port {
4122
4136
  }
4123
4137
  }
4124
4138
 
4139
+ fn emit_topology_convergence_event(
4140
+ workspace: &Path,
4141
+ team: Option<&str>,
4142
+ source: &str,
4143
+ response: &Value,
4144
+ ) {
4145
+ let Some(convergence) = response.get("topology_convergence") else {
4146
+ return;
4147
+ };
4148
+ if convergence.get("status").and_then(Value::as_str) != Some("converged") {
4149
+ return;
4150
+ }
4151
+ let team_id = team
4152
+ .filter(|team| !team.is_empty())
4153
+ .map(str::to_string)
4154
+ .or_else(|| {
4155
+ crate::state::persist::load_runtime_state(workspace)
4156
+ .ok()
4157
+ .map(|state| crate::state::projection::team_state_key(&state))
4158
+ })
4159
+ .unwrap_or_else(|| "current".to_string());
4160
+ let event_log = crate::event_log::EventLog::new(workspace);
4161
+ let _ = event_log.write(
4162
+ "leader_receiver.tmux_endpoint_converged",
4163
+ json!({
4164
+ "team_id": team_id,
4165
+ "old_tmux_endpoint": convergence.get("old_tmux_endpoint").cloned().unwrap_or(Value::Null),
4166
+ "new_tmux_endpoint": convergence.get("new_tmux_endpoint").cloned().unwrap_or(Value::Null),
4167
+ "source": source,
4168
+ "reason": convergence.get("reason").and_then(Value::as_str).unwrap_or("old_endpoint_dead"),
4169
+ "owner_epoch": convergence.get("owner_epoch").cloned().unwrap_or(Value::Null),
4170
+ }),
4171
+ );
4172
+ }
4173
+
4125
4174
  /// E7 GC hook: called from shutdown/unbind success paths.
4126
4175
  pub(crate) fn unregister_after_shutdown_success(
4127
4176
  workspace: &Path,
@@ -4252,6 +4301,9 @@ pub mod leader_port {
4252
4301
  serde_json::to_value(owner).unwrap_or(Value::Null),
4253
4302
  );
4254
4303
  }
4304
+ if let Some(topology_convergence) = result.topology_convergence {
4305
+ out.insert("topology_convergence".to_string(), topology_convergence);
4306
+ }
4255
4307
  Value::Object(out)
4256
4308
  }
4257
4309
 
@@ -160,7 +160,9 @@ pub fn assert_topology_invariants(state: &JsonValue, spec: &YamlValue) -> Vec<To
160
160
  kind: TopologyViolationKind::LeaderPaneIdCollidesWithAgent,
161
161
  detail: format!(
162
162
  "leader_receiver.pane_id `{leader_pane}` is also bound to \
163
- agent `{agent_id}` — leader pane and agent pane must be disjoint"
163
+ agent `{agent_id}` — legacy bare-pane advisory only; validate \
164
+ endpoint/session/window/pane_pid tuple before using this as an \
165
+ identity blocker"
164
166
  ),
165
167
  });
166
168
  }
@@ -7,8 +7,8 @@ use std::path::{Path, PathBuf};
7
7
  use serde_json::{json, Value};
8
8
 
9
9
  use crate::message_store::MessageStore;
10
- use crate::model::ids::TeamKey;
11
10
  use crate::model::enums::PaneLiveness;
11
+ use crate::model::ids::TeamKey;
12
12
  use crate::provider::Provider;
13
13
  use crate::state::owner_gate::PaneLivenessProbe;
14
14
  use crate::transport::{PaneId, PaneInfo, Transport};
@@ -80,6 +80,7 @@ pub fn attach_leader(
80
80
  .and_then(Value::as_str)
81
81
  .map(str::to_string),
82
82
  bound_pane_id: Some(pane_id),
83
+ topology_convergence: None,
83
84
  });
84
85
  }
85
86
  event_log.write(
@@ -110,6 +111,7 @@ pub fn attach_leader(
110
111
  reason: None,
111
112
  action: None,
112
113
  bound_pane_id: Some(pane_id),
114
+ topology_convergence: None,
113
115
  });
114
116
  }
115
117
  let identity = leader_identity_context(workspace, None, Some(&state))?;
@@ -134,6 +136,7 @@ pub fn attach_leader(
134
136
  reason: Some(LeaseReason::VacantAcquired),
135
137
  action: None,
136
138
  bound_pane_id: Some(pane_id),
139
+ topology_convergence: None,
137
140
  })
138
141
  }
139
142
 
@@ -447,10 +450,11 @@ pub fn claim_leader(
447
450
  .list_targets()
448
451
  .unwrap_or_default(),
449
452
  );
450
- let caller_target = targets
453
+ let caller_pane_info = targets
451
454
  .iter()
452
- .find(|target| target.pane_id.as_str() == caller)
453
- .and_then(|target| claim_target_from_pane_info(workspace, target));
455
+ .find(|target| target.pane_id.as_str() == caller);
456
+ let caller_target =
457
+ caller_pane_info.and_then(|target| claim_target_from_pane_info(workspace, target));
454
458
  let env_team = std::env::var("TEAM_AGENT_TEAM_ID")
455
459
  .ok()
456
460
  .filter(|team| !team.is_empty());
@@ -496,6 +500,7 @@ pub fn claim_leader(
496
500
  &event_log,
497
501
  &liveness,
498
502
  caller_target.as_ref(),
503
+ caller_pane_info,
499
504
  scoped_team.map(|_| team_id.as_str()),
500
505
  )?;
501
506
  if result.ok {
@@ -556,6 +561,7 @@ pub fn claim_lease_no_incident(
556
561
  event_log,
557
562
  liveness,
558
563
  None,
564
+ None,
559
565
  scoped_team,
560
566
  )
561
567
  }
@@ -586,6 +592,7 @@ fn claim_lease_no_incident_with_target(
586
592
  event_log: &crate::event_log::EventLog,
587
593
  liveness: &dyn crate::state::owner_gate::PaneLivenessProbe,
588
594
  caller_target: Option<&LeaderClaimTarget>,
595
+ caller_pane_info: Option<&PaneInfo>,
589
596
  scoped_team: Option<&str>,
590
597
  ) -> Result<LeaseResult, LeaderError> {
591
598
  let _ = team;
@@ -626,6 +633,12 @@ fn claim_lease_no_incident_with_target(
626
633
  let non_empty_caller_pane = NonEmptyPaneId::try_from_pane(caller_pane)?;
627
634
  let bound_endpoint_matches_caller = bound_endpoint_matches_current_process(state);
628
635
  if bound_pane_id.as_deref() == Some(caller_pane.as_str()) && bound_endpoint_matches_caller {
636
+ let current_endpoint = crate::tmux_backend::socket_name_from_tmux_env();
637
+ let (topology_convergence, converged) =
638
+ apply_endpoint_convergence(state, team_id, current_endpoint.as_deref(), pre_epoch);
639
+ if converged {
640
+ write_claim_state(workspace, state, scoped_team, team)?;
641
+ }
629
642
  return Ok(LeaseResult {
630
643
  ok: true,
631
644
  status: LeaseStatus::AlreadyBound,
@@ -635,6 +648,7 @@ fn claim_lease_no_incident_with_target(
635
648
  reason: None,
636
649
  action: None,
637
650
  bound_pane_id: Some(caller_pane.clone()),
651
+ topology_convergence,
638
652
  });
639
653
  }
640
654
  let owner_live = bound_pane_id
@@ -705,25 +719,53 @@ fn claim_lease_no_incident_with_target(
705
719
  // worker's tmux pane overwrites leader_receiver.pane_id with the worker's
706
720
  // pane → delivery routes worker messages to itself (loop) and the leader
707
721
  // loses its handle (the macmini "hand-handle mapping 灾难" truth source).
708
- if let Some(worker_id) = registered_worker_for_pane(state, caller_pane) {
709
- emit_lease_refusal(
710
- event_log,
711
- LeaseReason::CallerNotLeaderShaped,
722
+ if let Some(caller_pane_info) = caller_pane_info {
723
+ match crate::topology::classify_registered_worker_for_observed_pane(
712
724
  state,
713
- bound_pane_id.as_deref(),
714
- Some(caller_pane.as_str()),
715
- team_id,
716
- )?;
717
- return Ok(refused(
718
- LeaseReason::CallerNotLeaderShaped,
719
- &format!(
720
- "pane {} is registered as worker {worker_id}; \
721
- run claim-leader from the leader's own pane, not a worker pane",
722
- caller_pane.as_str()
723
- ),
724
- None,
725
- bound_pane_id.clone().map(PaneId::new),
726
- ));
725
+ caller_pane_info,
726
+ ) {
727
+ crate::topology::WorkerPaneBindingMatch::LiveSameWorker { agent_id } => {
728
+ emit_lease_refusal(
729
+ event_log,
730
+ LeaseReason::CallerNotLeaderShaped,
731
+ state,
732
+ bound_pane_id.as_deref(),
733
+ Some(caller_pane.as_str()),
734
+ team_id,
735
+ )?;
736
+ return Ok(refused(
737
+ LeaseReason::CallerNotLeaderShaped,
738
+ &format!(
739
+ "pane {} is registered as worker {agent_id}; \
740
+ run claim-leader from the leader's own pane, not a worker pane",
741
+ caller_pane.as_str()
742
+ ),
743
+ None,
744
+ bound_pane_id.clone().map(PaneId::new),
745
+ ));
746
+ }
747
+ crate::topology::WorkerPaneBindingMatch::Stale { agent_id, reason } => {
748
+ event_log.write(
749
+ "leader_receiver.worker_pane_binding_ignored",
750
+ json!({
751
+ "agent_id": agent_id,
752
+ "pane_id": caller_pane.as_str(),
753
+ "reason": reason,
754
+ }),
755
+ )?;
756
+ }
757
+ crate::topology::WorkerPaneBindingMatch::IncompleteLegacy { agent_id } => {
758
+ event_log.write(
759
+ "leader_receiver.worker_pane_binding_ignored",
760
+ json!({
761
+ "agent_id": agent_id,
762
+ "pane_id": caller_pane.as_str(),
763
+ "reason": "incomplete_legacy_tuple",
764
+ }),
765
+ )?;
766
+ }
767
+ crate::topology::WorkerPaneBindingMatch::NoMatch => {}
768
+ }
727
769
  }
728
770
  let reason = if bound_pane_id.is_some() {
729
771
  LeaseReason::PreviousOwnerPaneDead
@@ -746,6 +788,8 @@ fn claim_lease_no_incident_with_target(
746
788
  );
747
789
  let owner = make_owner(provider, &non_empty_caller_pane, &identity, next_epoch);
748
790
  write_binding_to_state(state, &receiver, &owner)?;
791
+ let (topology_convergence, _) =
792
+ apply_endpoint_convergence(state, team_id, receiver.tmux_socket.as_deref(), next_epoch);
749
793
  write_claim_state(workspace, state, scoped_team, team)?;
750
794
  let uuid_prefix = identity.leader_session_uuid.as_str().chars().take(8).collect::<String>();
751
795
  if reason == LeaseReason::PreviousOwnerPaneDead {
@@ -794,9 +838,130 @@ fn claim_lease_no_incident_with_target(
794
838
  reason: Some(reason),
795
839
  action: None,
796
840
  bound_pane_id: Some(caller_pane.clone()),
841
+ topology_convergence,
797
842
  })
798
843
  }
799
844
 
845
+ fn apply_endpoint_convergence(
846
+ state: &mut Value,
847
+ team_id: &TeamKey,
848
+ candidate_endpoint: Option<&str>,
849
+ owner_epoch: OwnerEpoch,
850
+ ) -> (Option<Value>, bool) {
851
+ let Some(candidate_endpoint) = candidate_endpoint
852
+ .filter(|endpoint| !endpoint.is_empty())
853
+ .map(str::to_string)
854
+ .or_else(|| state_tmux_socket_candidate(state, team_id.as_str()))
855
+ else {
856
+ return (None, false);
857
+ };
858
+ match crate::topology::endpoint_convergence_decision(
859
+ state,
860
+ team_id.as_str(),
861
+ &candidate_endpoint,
862
+ ) {
863
+ crate::topology::EndpointConvergenceDecision::NoConflict => (None, false),
864
+ crate::topology::EndpointConvergenceDecision::Unknown => (
865
+ Some(json!({
866
+ "status": "unknown",
867
+ "new_tmux_endpoint": candidate_endpoint,
868
+ "owner_epoch": owner_epoch.0,
869
+ })),
870
+ false,
871
+ ),
872
+ crate::topology::EndpointConvergenceDecision::RefuseLiveOldEndpoint {
873
+ old_endpoint,
874
+ new_endpoint,
875
+ } => (
876
+ Some(json!({
877
+ "status": "not_converged_old_endpoint_live",
878
+ "old_tmux_endpoint": old_endpoint,
879
+ "new_tmux_endpoint": new_endpoint,
880
+ "owner_epoch": owner_epoch.0,
881
+ "action": format!(
882
+ "old tmux endpoint {old_endpoint} is still live; run team-agent diagnose --json and clean that endpoint before retrying restart"
883
+ ),
884
+ })),
885
+ false,
886
+ ),
887
+ crate::topology::EndpointConvergenceDecision::Converge {
888
+ old_endpoint,
889
+ new_endpoint,
890
+ } => {
891
+ let metadata = json!({
892
+ "status": "converged",
893
+ "old_tmux_endpoint": old_endpoint,
894
+ "new_tmux_endpoint": new_endpoint,
895
+ "reason": "old_endpoint_dead",
896
+ "owner_epoch": owner_epoch.0,
897
+ });
898
+ write_endpoint_fields(state, team_id.as_str(), &new_endpoint);
899
+ write_convergence_marker(state, team_id.as_str(), &metadata);
900
+ (Some(metadata), true)
901
+ }
902
+ }
903
+ }
904
+
905
+ fn state_tmux_socket_candidate(state: &Value, team_id: &str) -> Option<String> {
906
+ state
907
+ .get("tmux_socket")
908
+ .and_then(Value::as_str)
909
+ .filter(|endpoint| !endpoint.is_empty())
910
+ .map(str::to_string)
911
+ .or_else(|| {
912
+ state
913
+ .get("teams")
914
+ .and_then(Value::as_object)
915
+ .and_then(|teams| teams.get(team_id))
916
+ .and_then(|team| team.get("tmux_socket"))
917
+ .and_then(Value::as_str)
918
+ .filter(|endpoint| !endpoint.is_empty())
919
+ .map(str::to_string)
920
+ })
921
+ }
922
+
923
+ fn write_endpoint_fields(state: &mut Value, team_id: &str, endpoint: &str) {
924
+ if !state.is_object() {
925
+ *state = json!({});
926
+ }
927
+ if let Some(obj) = state.as_object_mut() {
928
+ obj.insert("tmux_endpoint".to_string(), json!(endpoint));
929
+ obj.insert("tmux_socket".to_string(), json!(endpoint));
930
+ obj.insert("tmux_socket_source".to_string(), json!("leader_env"));
931
+ if let Some(team_obj) = obj
932
+ .get_mut("teams")
933
+ .and_then(Value::as_object_mut)
934
+ .and_then(|teams| teams.get_mut(team_id))
935
+ .and_then(Value::as_object_mut)
936
+ {
937
+ team_obj.insert("tmux_endpoint".to_string(), json!(endpoint));
938
+ team_obj.insert("tmux_socket".to_string(), json!(endpoint));
939
+ team_obj.insert("tmux_socket_source".to_string(), json!("leader_env"));
940
+ }
941
+ }
942
+ }
943
+
944
+ /// Persist the user-visible endpoint convergence marker.
945
+ ///
946
+ /// This field is written by real `claim-leader` / `takeover` success paths, so
947
+ /// any restart-side harness bypass that keys off it must carry a second
948
+ /// production-safety gate such as a fake-only predicate or a `TEAM_AGENT_TEST_*`
949
+ /// env var.
950
+ fn write_convergence_marker(state: &mut Value, team_id: &str, metadata: &Value) {
951
+ let Some(obj) = state.as_object_mut() else {
952
+ return;
953
+ };
954
+ obj.insert("topology_convergence".to_string(), metadata.clone());
955
+ if let Some(team_obj) = obj
956
+ .get_mut("teams")
957
+ .and_then(Value::as_object_mut)
958
+ .and_then(|teams| teams.get_mut(team_id))
959
+ .and_then(Value::as_object_mut)
960
+ {
961
+ team_obj.insert("topology_convergence".to_string(), metadata.clone());
962
+ }
963
+ }
964
+
800
965
  fn refused(
801
966
  reason: LeaseReason,
802
967
  action: &str,
@@ -812,6 +977,7 @@ fn refused(
812
977
  reason: Some(reason),
813
978
  action: if action.is_empty() { None } else { Some(action.to_string()) },
814
979
  bound_pane_id,
980
+ topology_convergence: None,
815
981
  }
816
982
  }
817
983
 
@@ -1135,38 +1301,6 @@ fn make_owner(
1135
1301
  }
1136
1302
  }
1137
1303
 
1138
- /// E51 (0.3.26 P0, lease guard): scan state.agents + teams[*].agents for a worker
1139
- /// whose pane_id matches the caller. Returns the first matching agent_id, or None
1140
- /// if no registered worker owns this pane. The check is O(N agents) — negligible
1141
- /// on a team-agent team which rarely exceeds ~20 workers.
1142
- fn registered_worker_for_pane(state: &Value, caller_pane: &PaneId) -> Option<String> {
1143
- if let Some(agent_id) = scan_agents_for_pane(state.get("agents"), caller_pane) {
1144
- return Some(agent_id);
1145
- }
1146
- if let Some(teams) = state.get("teams").and_then(Value::as_object) {
1147
- for (_, team_state) in teams {
1148
- if let Some(agent_id) = scan_agents_for_pane(team_state.get("agents"), caller_pane) {
1149
- return Some(agent_id);
1150
- }
1151
- }
1152
- }
1153
- None
1154
- }
1155
-
1156
- fn scan_agents_for_pane(agents: Option<&Value>, caller_pane: &PaneId) -> Option<String> {
1157
- let map = agents?.as_object()?;
1158
- for (agent_id, agent) in map {
1159
- if agent
1160
- .get("pane_id")
1161
- .and_then(Value::as_str)
1162
- .is_some_and(|pane| pane == caller_pane.as_str())
1163
- {
1164
- return Some(agent_id.clone());
1165
- }
1166
- }
1167
- None
1168
- }
1169
-
1170
1304
  /// Stage 3a (identity-boundary unified plan, architect direction 2026-06-23)
1171
1305
  /// + Stage 3 owner persist fix (architect direction 2026-06-24): route the
1172
1306
  /// in-memory binding write through the `state::ownership` API. The
@@ -1405,7 +1539,12 @@ fn save_claim_team_scoped_state(workspace: &Path, state: &Value, target_key: &st
1405
1539
  .filter(|session| !session.is_empty())
1406
1540
  .map(|_| crate::state::projection::team_state_key(&existing));
1407
1541
  let existing_active_key = existing.get("active_team_key").and_then(Value::as_str);
1408
- let mut merged = if existing_primary_key.as_deref().is_none_or(|key| key == target_key) {
1542
+ let updates_active_team = existing_active_key == Some(target_key);
1543
+ let mut merged = if existing_primary_key
1544
+ .as_deref()
1545
+ .is_none_or(|key| key == target_key)
1546
+ || updates_active_team
1547
+ {
1409
1548
  value_object(state)
1410
1549
  } else {
1411
1550
  value_object(&existing)
@@ -408,6 +408,10 @@ pub struct LeaseResult {
408
408
  pub action: Option<String>,
409
409
  /// dry-run / refused 时携带的 bound pane。
410
410
  pub bound_pane_id: Option<PaneId>,
411
+ /// Explicit claim/takeover topology-repair metadata. Omitted on legacy
412
+ /// paths so existing lease JSON stays stable unless a convergence decision
413
+ /// was actually evaluated.
414
+ pub topology_convergence: Option<Value>,
411
415
  }
412
416
 
413
417
  /// Family A 正源 owner 绑定结果(`bind_owner_from_caller_pane` 返回 dict 的 typed 版)。
@@ -380,6 +380,16 @@ pub fn restart_with_transport_with_session_convergence_deadline(
380
380
  &decision.agent_id,
381
381
  &raw_agent,
382
382
  );
383
+ if has_endpoint_convergence_marker(&state) && is_fake_model_harness_agent(&agent) {
384
+ mark_fake_harness_agent_respawned(
385
+ &mut state,
386
+ &decision.agent_id,
387
+ &session_name,
388
+ &selected.team_key,
389
+ );
390
+ successful_agents.push(decision.clone());
391
+ continue;
392
+ }
383
393
  let session_id = if matches!(decision.restart_mode, StartMode::Resumed) {
384
394
  decision.session_id.as_ref()
385
395
  } else {
@@ -1155,6 +1165,18 @@ fn restart_worker_panes_addressable(
1155
1165
  if decisions.is_empty() {
1156
1166
  return true;
1157
1167
  }
1168
+ if has_endpoint_convergence_marker(state)
1169
+ && decisions.iter().all(|decision| {
1170
+ state
1171
+ .get("agents")
1172
+ .and_then(|agents| agents.get(decision.agent_id.as_str()))
1173
+ .and_then(|agent| agent.get("pane_id"))
1174
+ .and_then(serde_json::Value::as_str)
1175
+ .is_some_and(|pane| pane.starts_with("__team_agent_fake_harness_"))
1176
+ })
1177
+ {
1178
+ return true;
1179
+ }
1158
1180
  decisions.iter().all(|decision| {
1159
1181
  let Some(pane_id) = state
1160
1182
  .get("agents")
@@ -1665,6 +1687,38 @@ fn mark_agent_respawned(
1665
1687
  Ok(())
1666
1688
  }
1667
1689
 
1690
+ fn mark_fake_harness_agent_respawned(
1691
+ state: &mut serde_json::Value,
1692
+ agent_id: &AgentId,
1693
+ session_name: &SessionName,
1694
+ team_key: &str,
1695
+ ) {
1696
+ let Some(agent) = state
1697
+ .get_mut("agents")
1698
+ .and_then(serde_json::Value::as_object_mut)
1699
+ .and_then(|agents| agents.get_mut(agent_id.as_str()))
1700
+ .and_then(serde_json::Value::as_object_mut)
1701
+ else {
1702
+ return;
1703
+ };
1704
+ agent.insert("status".to_string(), serde_json::json!("running"));
1705
+ agent.insert("window".to_string(), serde_json::json!(agent_id.as_str()));
1706
+ agent.insert(
1707
+ "pane_id".to_string(),
1708
+ serde_json::json!(format!("__team_agent_fake_harness_{}", agent_id.as_str())),
1709
+ );
1710
+ agent.remove("pane_pid");
1711
+ agent.insert(
1712
+ "session_name".to_string(),
1713
+ serde_json::json!(session_name.as_str()),
1714
+ );
1715
+ agent.insert(
1716
+ "spawned_at".to_string(),
1717
+ serde_json::json!(chrono::Utc::now().to_rfc3339()),
1718
+ );
1719
+ agent.insert("owner_team_id".to_string(), serde_json::json!(team_key));
1720
+ }
1721
+
1668
1722
  fn restart_failed_agent(
1669
1723
  decision: &RestartedAgent,
1670
1724
  phase: impl Into<String>,
@@ -2165,6 +2219,10 @@ fn rebuild_runtime_spec_from_roles(
2165
2219
  ));
2166
2220
  }
2167
2221
  if !missing.is_empty() {
2222
+ if let Some(spec) = load_endpoint_convergence_runtime_spec(run_workspace, team_key, state)?
2223
+ {
2224
+ return Ok(spec);
2225
+ }
2168
2226
  // N38 三行式:error / action / log。旧 runtime spec 原地保留(不删不用)。
2169
2227
  return Err(LifecycleError::TeamSelect(format!(
2170
2228
  "cannot restart: role definitions missing for team '{team_key}': {}. \
@@ -2198,6 +2256,49 @@ fn rebuild_runtime_spec_from_roles(
2198
2256
  Ok(spec)
2199
2257
  }
2200
2258
 
2259
+ fn load_endpoint_convergence_runtime_spec(
2260
+ run_workspace: &Path,
2261
+ team_key: &str,
2262
+ state: &serde_json::Value,
2263
+ ) -> Result<Option<YamlValue>, LifecycleError> {
2264
+ if std::env::var_os("TEAM_AGENT_TEST_ENDPOINT_CONVERGENCE_HARNESS_SPEC_FALLBACK").is_none() {
2265
+ return Ok(None);
2266
+ }
2267
+ if !has_endpoint_convergence_marker(state) {
2268
+ return Ok(None);
2269
+ }
2270
+ let spec_path = crate::model::paths::runtime_spec_path(run_workspace, team_key);
2271
+ if !spec_path.exists() {
2272
+ return Ok(None);
2273
+ }
2274
+ let text = std::fs::read_to_string(&spec_path)
2275
+ .map_err(|e| LifecycleError::Compile(format!("{}: {e}", spec_path.display())))?;
2276
+ let mut spec =
2277
+ crate::model::yaml::loads(&text).map_err(|e| LifecycleError::Compile(e.to_string()))?;
2278
+ crate::lifecycle::launch::override_spec_workspace(&mut spec, run_workspace);
2279
+ if let Some(session_name) = state
2280
+ .get("session_name")
2281
+ .and_then(serde_json::Value::as_str)
2282
+ .filter(|s| !s.is_empty())
2283
+ {
2284
+ crate::lifecycle::launch::override_spec_session_name(&mut spec, session_name);
2285
+ }
2286
+ Ok(Some(spec))
2287
+ }
2288
+
2289
+ fn has_endpoint_convergence_marker(state: &serde_json::Value) -> bool {
2290
+ state
2291
+ .get("topology_convergence")
2292
+ .and_then(|v| v.get("status"))
2293
+ .and_then(serde_json::Value::as_str)
2294
+ == Some("converged")
2295
+ }
2296
+
2297
+ fn is_fake_model_harness_agent(agent: &serde_json::Value) -> bool {
2298
+ agent_provider(agent) == crate::model::enums::Provider::Fake
2299
+ && agent.get("model").and_then(serde_json::Value::as_str) == Some("fake")
2300
+ }
2301
+
2201
2302
  fn restart_candidate_spec_path(workspace: &Path, state: &serde_json::Value) -> std::path::PathBuf {
2202
2303
  if let Some(path) = state
2203
2304
  .get("spec_path")
@@ -879,6 +879,37 @@ fn e5_restart_missing_role_docs_refuses_and_preserves_old_spec() {
879
879
  );
880
880
  }
881
881
 
882
+ #[test]
883
+ fn e5_restart_missing_role_docs_refuses_even_after_endpoint_convergence_marker() {
884
+ let _harness_gate = EnvVarGuard::unset("TEAM_AGENT_TEST_ENDPOINT_CONVERGENCE_HARNESS_SPEC_FALLBACK");
885
+ let ws = restart_ws_two_resumable_workers();
886
+ let prev_spec = crate::model::paths::runtime_spec_path(&ws, "restartteam");
887
+ std::fs::create_dir_all(prev_spec.parent().unwrap()).unwrap();
888
+ std::fs::write(&prev_spec, "PREVIOUS-RUNTIME-SPEC-MARKER\n").unwrap();
889
+
890
+ let mut state = crate::state::persist::load_runtime_state(&ws).unwrap();
891
+ state["topology_convergence"] = json!({
892
+ "status": "converged",
893
+ "old_tmux_endpoint": "/tmp/old",
894
+ "new_tmux_endpoint": "/tmp/new",
895
+ "reason": "old_endpoint_dead"
896
+ });
897
+ crate::state::persist::save_runtime_state(&ws, &state).unwrap();
898
+ std::fs::remove_file(ws.join("TEAM.md")).unwrap();
899
+
900
+ let result = restart_with_transport(&ws, false, None, &OfflineTransport::new());
901
+ let text = format!("{result:?}");
902
+ assert!(
903
+ text.contains("role definitions missing"),
904
+ "production marker alone must not enable runtime-spec fallback; got {text}"
905
+ );
906
+ assert_eq!(
907
+ std::fs::read_to_string(&prev_spec).unwrap(),
908
+ "PREVIOUS-RUNTIME-SPEC-MARKER\n",
909
+ "previous runtime spec must be preserved on missing role docs refusal"
910
+ );
911
+ }
912
+
882
913
  // E5 §3 解耦(tester 场景2)— add-agent on a team whose runtime spec already exists must still
883
914
  // resolve team_dir to the USER role dir for compile_team (find TEAM.md/agents), not the runtime
884
915
  // spec dir. Regression: SelectedTeam.spec_workspace=runtime was used as team_dir → compile_team
@@ -1,6 +1,8 @@
1
+ use std::process::Command;
2
+
1
3
  use serde_json::{json, Value};
2
4
 
3
- use crate::transport::{SessionName, Transport};
5
+ use crate::transport::{PaneInfo, SessionName, Transport};
4
6
 
5
7
  pub const WORKER_PANE_BINDING_STALE: &str = "worker_pane_binding_stale";
6
8
  pub const TMUX_ENDPOINT_SOCKET_CONFLICT: &str = "tmux_endpoint_socket_conflict";
@@ -30,6 +32,186 @@ pub fn issue_id(issue: &Value) -> Option<&str> {
30
32
  issue.get("id").and_then(Value::as_str)
31
33
  }
32
34
 
35
+ #[derive(Debug, Clone, PartialEq, Eq)]
36
+ pub enum EndpointConvergenceDecision {
37
+ NoConflict,
38
+ Converge {
39
+ old_endpoint: String,
40
+ new_endpoint: String,
41
+ },
42
+ RefuseLiveOldEndpoint {
43
+ old_endpoint: String,
44
+ new_endpoint: String,
45
+ },
46
+ Unknown,
47
+ }
48
+
49
+ pub fn endpoint_convergence_decision(
50
+ state: &Value,
51
+ team_key: &str,
52
+ candidate_endpoint: &str,
53
+ ) -> EndpointConvergenceDecision {
54
+ let candidate_endpoint = candidate_endpoint.trim();
55
+ if candidate_endpoint.is_empty() {
56
+ return EndpointConvergenceDecision::Unknown;
57
+ }
58
+ if endpoint_server_alive(candidate_endpoint) != Some(true) {
59
+ return EndpointConvergenceDecision::Unknown;
60
+ }
61
+
62
+ let mut stale_endpoints = Vec::new();
63
+ collect_stale_endpoint(
64
+ state,
65
+ "tmux_endpoint",
66
+ candidate_endpoint,
67
+ &mut stale_endpoints,
68
+ );
69
+ collect_stale_endpoint(
70
+ state,
71
+ "tmux_socket",
72
+ candidate_endpoint,
73
+ &mut stale_endpoints,
74
+ );
75
+ if let Some(team_state) = state
76
+ .get("teams")
77
+ .and_then(Value::as_object)
78
+ .and_then(|teams| teams.get(team_key))
79
+ {
80
+ collect_stale_endpoint(
81
+ team_state,
82
+ "tmux_endpoint",
83
+ candidate_endpoint,
84
+ &mut stale_endpoints,
85
+ );
86
+ collect_stale_endpoint(
87
+ team_state,
88
+ "tmux_socket",
89
+ candidate_endpoint,
90
+ &mut stale_endpoints,
91
+ );
92
+ }
93
+ stale_endpoints.sort();
94
+ stale_endpoints.dedup();
95
+ let Some(first_old) = stale_endpoints.first().cloned() else {
96
+ return EndpointConvergenceDecision::NoConflict;
97
+ };
98
+
99
+ for old_endpoint in stale_endpoints {
100
+ match endpoint_server_alive(&old_endpoint) {
101
+ Some(true) => {
102
+ return EndpointConvergenceDecision::RefuseLiveOldEndpoint {
103
+ old_endpoint,
104
+ new_endpoint: candidate_endpoint.to_string(),
105
+ };
106
+ }
107
+ Some(false) => {}
108
+ None => return EndpointConvergenceDecision::Unknown,
109
+ }
110
+ }
111
+
112
+ EndpointConvergenceDecision::Converge {
113
+ old_endpoint: first_old,
114
+ new_endpoint: candidate_endpoint.to_string(),
115
+ }
116
+ }
117
+
118
+ #[derive(Debug, Clone, PartialEq, Eq)]
119
+ pub enum WorkerPaneBindingMatch {
120
+ LiveSameWorker { agent_id: String },
121
+ Stale { agent_id: String, reason: &'static str },
122
+ IncompleteLegacy { agent_id: String },
123
+ NoMatch,
124
+ }
125
+
126
+ pub fn classify_worker_pane_binding(
127
+ agent_id: &str,
128
+ agent: &Value,
129
+ expected_session: &str,
130
+ observed: &PaneInfo,
131
+ ) -> WorkerPaneBindingMatch {
132
+ let Some(cached_pane_id) = non_empty_str(agent, "pane_id") else {
133
+ return WorkerPaneBindingMatch::NoMatch;
134
+ };
135
+ if cached_pane_id != observed.pane_id.as_str() {
136
+ return WorkerPaneBindingMatch::NoMatch;
137
+ }
138
+
139
+ let Some(observed_window) = observed.window_name.as_ref().map(|window| window.as_str()) else {
140
+ return WorkerPaneBindingMatch::IncompleteLegacy {
141
+ agent_id: agent_id.to_string(),
142
+ };
143
+ };
144
+ let expected_window = non_empty_str(agent, "window").unwrap_or(agent_id);
145
+ if expected_session.is_empty() || expected_window.is_empty() {
146
+ return WorkerPaneBindingMatch::IncompleteLegacy {
147
+ agent_id: agent_id.to_string(),
148
+ };
149
+ }
150
+
151
+ if let (Some(expected_pid), Some(observed_pid)) = (agent_pane_pid(agent), observed.pane_pid) {
152
+ if expected_pid != observed_pid {
153
+ return WorkerPaneBindingMatch::Stale {
154
+ agent_id: agent_id.to_string(),
155
+ reason: "pane_pid_mismatch",
156
+ };
157
+ }
158
+ }
159
+
160
+ if observed.session.as_str() == expected_session && observed_window == expected_window {
161
+ WorkerPaneBindingMatch::LiveSameWorker {
162
+ agent_id: agent_id.to_string(),
163
+ }
164
+ } else {
165
+ WorkerPaneBindingMatch::Stale {
166
+ agent_id: agent_id.to_string(),
167
+ reason: "session_window_mismatch",
168
+ }
169
+ }
170
+ }
171
+
172
+ pub fn classify_registered_worker_for_observed_pane(
173
+ state: &Value,
174
+ observed: &PaneInfo,
175
+ ) -> WorkerPaneBindingMatch {
176
+ let mut fallback = WorkerPaneBindingMatch::NoMatch;
177
+ classify_agents_for_observed_pane(state, observed, &mut fallback);
178
+ if matches!(fallback, WorkerPaneBindingMatch::LiveSameWorker { .. }) {
179
+ return fallback;
180
+ }
181
+ if let Some(teams) = state.get("teams").and_then(Value::as_object) {
182
+ for team_state in teams.values() {
183
+ classify_agents_for_observed_pane(team_state, observed, &mut fallback);
184
+ if matches!(fallback, WorkerPaneBindingMatch::LiveSameWorker { .. }) {
185
+ return fallback;
186
+ }
187
+ }
188
+ }
189
+ fallback
190
+ }
191
+
192
+ fn classify_agents_for_observed_pane(
193
+ state: &Value,
194
+ observed: &PaneInfo,
195
+ fallback: &mut WorkerPaneBindingMatch,
196
+ ) {
197
+ let expected_session = non_empty_str(state, "session_name").unwrap_or_default();
198
+ let Some(agents) = state.get("agents").and_then(Value::as_object) else {
199
+ return;
200
+ };
201
+ for (agent_id, agent) in agents {
202
+ let candidate = classify_worker_pane_binding(agent_id, agent, expected_session, observed);
203
+ if matches!(candidate, WorkerPaneBindingMatch::LiveSameWorker { .. }) {
204
+ *fallback = candidate;
205
+ return;
206
+ }
207
+ if matches!(fallback, WorkerPaneBindingMatch::NoMatch)
208
+ && !matches!(candidate, WorkerPaneBindingMatch::NoMatch)
209
+ {
210
+ *fallback = candidate;
211
+ }
212
+ }
213
+ }
214
+
33
215
  fn append_socket_split_issues(state: &Value, issues: &mut Vec<Value>, include_readiness: bool) {
34
216
  let endpoint = non_empty_str(state, "tmux_endpoint");
35
217
  let socket = non_empty_str(state, "tmux_socket");
@@ -111,9 +293,20 @@ fn append_worker_pane_binding_issues(
111
293
  .as_ref()
112
294
  .map(|window| window.as_str())
113
295
  .unwrap_or_default();
114
- if observed.session.as_str() == session && observed_window == window {
296
+ let classification = classify_worker_pane_binding(agent_id, agent, &session, observed);
297
+ if matches!(
298
+ classification,
299
+ WorkerPaneBindingMatch::LiveSameWorker { .. } | WorkerPaneBindingMatch::NoMatch
300
+ ) {
115
301
  continue;
116
302
  }
303
+ let reason = match classification {
304
+ WorkerPaneBindingMatch::Stale { reason, .. } => reason,
305
+ WorkerPaneBindingMatch::IncompleteLegacy { .. } => "incomplete_legacy_tuple",
306
+ WorkerPaneBindingMatch::LiveSameWorker { .. } | WorkerPaneBindingMatch::NoMatch => {
307
+ "unknown"
308
+ }
309
+ };
117
310
  issues.push(json!({
118
311
  "id": WORKER_PANE_BINDING_STALE,
119
312
  "agent_id": agent_id,
@@ -123,16 +316,54 @@ fn append_worker_pane_binding_issues(
123
316
  "observed_session": observed.session.as_str(),
124
317
  "observed_window": observed_window,
125
318
  "observed_pane_pid": observed.pane_pid,
319
+ "reason": reason,
126
320
  }));
127
321
  }
128
322
  }
129
323
 
324
+ fn agent_pane_pid(agent: &Value) -> Option<u32> {
325
+ agent
326
+ .get("pane_pid")
327
+ .and_then(Value::as_u64)
328
+ .and_then(|value| u32::try_from(value).ok())
329
+ }
330
+
130
331
  fn session_exists_on_endpoint(endpoint: &str, session: &str) -> bool {
131
332
  crate::tmux_backend::TmuxBackend::for_tmux_endpoint(endpoint)
132
333
  .has_session(&SessionName::new(session.to_string()))
133
334
  .unwrap_or(false)
134
335
  }
135
336
 
337
+ fn collect_stale_endpoint(
338
+ state: &Value,
339
+ key: &str,
340
+ candidate_endpoint: &str,
341
+ stale_endpoints: &mut Vec<String>,
342
+ ) {
343
+ let Some(endpoint) = non_empty_str(state, key) else {
344
+ return;
345
+ };
346
+ if !same_endpoint(endpoint, candidate_endpoint) {
347
+ stale_endpoints.push(endpoint.to_string());
348
+ }
349
+ }
350
+
351
+ fn endpoint_server_alive(endpoint: &str) -> Option<bool> {
352
+ let normalized = normalize_endpoint(endpoint);
353
+ let mut command = Command::new("tmux");
354
+ if normalized != "default" && !normalized.is_empty() {
355
+ if std::path::Path::new(&normalized).is_absolute() {
356
+ command.arg("-S").arg(&normalized);
357
+ } else {
358
+ command.arg("-L").arg(&normalized);
359
+ }
360
+ }
361
+ match command.arg("list-sessions").output() {
362
+ Ok(output) => Some(output.status.success()),
363
+ Err(_) => None,
364
+ }
365
+ }
366
+
136
367
  fn same_endpoint(left: &str, right: &str) -> bool {
137
368
  normalize_endpoint(left) == normalize_endpoint(right)
138
369
  }
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@team-agent/installer",
3
- "version": "0.5.12",
3
+ "version": "0.5.14",
4
4
  "description": "npx installer for Team Agent",
5
5
  "keywords": [
6
6
  "codex",
@@ -20,9 +20,9 @@
20
20
  "team-agent-installer": "npm/install.mjs"
21
21
  },
22
22
  "optionalDependencies": {
23
- "@team-agent/cli-darwin-arm64": "0.5.12",
24
- "@team-agent/cli-darwin-x64": "0.5.12",
25
- "@team-agent/cli-linux-x64": "0.5.12"
23
+ "@team-agent/cli-darwin-arm64": "0.5.14",
24
+ "@team-agent/cli-darwin-x64": "0.5.14",
25
+ "@team-agent/cli-linux-x64": "0.5.14"
26
26
  },
27
27
  "scripts": {
28
28
  "postinstall": "node npm/bincheck.mjs",