code-coordinator 0.5.229__tar.gz → 0.5.231__tar.gz

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.229
+Version: 0.5.231
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/code_coordinator.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.229
+Version: 0.5.231
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/code_coordinator.egg-info/SOURCES.txt

@@ -541,6 +541,7 @@ tests/test_human_attended_437.py
 tests/test_init.py
 tests/test_integration.py
 tests/test_interactive_briefing_echo.py
+tests/test_interactive_pane_log_2541.py
 tests/test_interactive_prefill_state.py
 tests/test_interactive_session_relaunch.py
 tests/test_interactive_test_verdict_backstop.py

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/code_coordinator.egg-info/scm_file_list.json

@@ -611,6 +611,7 @@
     "tests/test_init.py",
     "tests/test_integration.py",
     "tests/test_interactive_briefing_echo.py",
+    "tests/test_interactive_pane_log_2541.py",
     "tests/test_interactive_prefill_state.py",
     "tests/test_interactive_session_relaunch.py",
     "tests/test_interactive_test_verdict_backstop.py",

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/code_coordinator.egg-info/scm_version.json

@@ -1,7 +1,7 @@
 {
-  "tag": "0.5.229",
+  "tag": "0.5.231",
   "distance": 0,
-  "node": "gc425ec07e071d15ed23c72606aaa805685c13e0e",
+  "node": "g9bcc6d2ee8084e339463aeb3572a47a58b9b1657",
   "dirty": false,
   "branch": "HEAD",
   "node_date": "2026-08-22"

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/coord/agent.py

@@ -5762,7 +5762,8 @@ class AgentServer:
 
     @staticmethod
     def _tmux_session_alive(assignment_id: str) -> bool:
-        """Return True when a ``coord-<assignment_id>`` tmux session is up.
+        """Return True when a ``coord-<assignment_id>`` tmux session is
+        genuinely still in use — alive AND its pane is not dead.
 
         Load-bearing guard for :meth:`clean_worktrees` (#1295): an
         interactive Test/Review/Merge/Work pane keeps its tmux session
@@ -5772,11 +5773,16 @@ class AgentServer:
         is the only reliable "someone is still using this worktree"
         signal.
 
-        Uses ``tmux has-session -t coord-<assignment_id>``; exit 0 means
-        the session exists.  Delegates to
-        :func:`coord.interactive.tmux_session_alive` so a single
-        implementation handles the subprocess and error-swallowing
-        semantics.  When ``tmux`` is not installed / not running / the
+        Delegates to :func:`coord.interactive.tmux_session_running` (a bare
+        ``tmux has-session`` alone is NOT enough — #2541 set
+        ``remain-on-exit on`` on every freshly-created coord tmux session so
+        a crashed pane's screen stays inspectable, which means
+        ``has-session`` now stays ``True`` after ANY pane exit, clean
+        success or crash, until a reaper notices and kills it. Without the
+        pane-dead check this guard would keep protecting — and thus never
+        cleaning up — a worktree whose interactive session has already
+        finished, for as long as the now-longer-lived dead session
+        lingers). When ``tmux`` is not installed / not running / the
         subprocess errors, we return ``False`` — "no live session, keep
         sweeping" — rather than raising, so one broken tmux install
         never aborts a fleet sweep across the rest of an agent's
@@ -5788,15 +5794,15 @@ class AgentServer:
         try:
             from coord.interactive import (  # noqa: PLC0415
                 tmux_available,
-                tmux_session_alive,
                 tmux_session_name,
+                tmux_session_running,
             )
         except Exception:  # noqa: BLE001 — defensive; module missing → no guard
             return False
         try:
             if not tmux_available():
                 return False
-            return tmux_session_alive(tmux_session_name(assignment_id))
+            return tmux_session_running(tmux_session_name(assignment_id))
         except Exception:  # noqa: BLE001 — never propagate out of the sweep
             return False
 

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/coord/agent_app.py

@@ -1231,11 +1231,41 @@ def build_app(
     async def logs(request: Request) -> Response:
         assignment_id = request.path_params["id"]
         assignment = server.get(assignment_id)
-        if assignment is None or assignment.log_path is None:
-            return JSONResponse(
-                {"error": f"unknown assignment {assignment_id}"}, status_code=404
-            )
-        log_path = Path(assignment.log_path)
+        status = assignment.status if assignment is not None else "unknown"
+        if assignment is not None and assignment.log_path is not None:
+            log_path = Path(assignment.log_path)
+        else:
+            # #2541: the human-attended ssh+tmux interactive launcher
+            # (`coord assign --interactive ...`) never goes through this
+            # agent's `/assign` — it SSHes in directly from the
+            # coordinator and drives `claude` inside a tmux session, so
+            # `server.get()` has no record of it. Those sessions still
+            # persist a best-effort pane-capture snapshot at the SAME
+            # conventional per-assignment path
+            # (`coord.interactive._persist_interactive_pane_log`) so a
+            # launch failure is diagnosable via `coord log <id>` without a
+            # manual SSH repro. Fall back to that path directly rather
+            # than 404ing just because this agent never tracked the
+            # assignment in memory.
+            # assignment_id is attacker-controllable (raw path param) and,
+            # unlike `assignment.log_path` above, is interpolated directly
+            # into a filesystem path here — reject anything that isn't the
+            # plain `[a-zA-Z0-9_-]+` shape every assignment_id generator in
+            # this codebase actually produces, so a `../` can't escape
+            # `logs/`.
+            if not re.fullmatch(r"[A-Za-z0-9_-]+", assignment_id):
+                return JSONResponse(
+                    {"error": f"invalid assignment id {assignment_id!r}"},
+                    status_code=400,
+                )
+
+            # `server.log_dir` (not the module-level `DEFAULT_STATE_DIR`
+            # default) so this honours a non-default `state_dir` exactly
+            # like every other lookup in this handler would if the
+            # assignment WERE tracked — production agents never override
+            # it, but this keeps the fallback truthful to what this
+            # specific server instance is actually configured with.
+            log_path = server.log_dir / f"{assignment_id}.log"
         if not log_path.exists():
             return JSONResponse(
                 {"error": f"no log file for assignment {assignment_id}"}, status_code=404
@@ -1255,7 +1285,7 @@ def build_app(
         total_size = log_path.stat().st_size
         headers = {
             "X-Coord-Log-Total": str(total_size),
-            "X-Coord-Log-Status": assignment.status,
+            "X-Coord-Log-Status": status,
         }
         return PlainTextResponse(body.decode("utf-8", errors="replace"), headers=headers)
 

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/coord/commands/dispatch_workers.py

@@ -141,8 +141,8 @@ def _dispatch_review_of(
         finalize_interactive_exit,
         launch_human_attended_interactive,
         tmux_available as _tmux_avail,
-        tmux_session_alive as _tmux_alive,
         tmux_session_name as _tmux_name,
+        tmux_session_running as _tmux_alive,
     )
 
     from coord.review import (  # noqa: PLC0415
@@ -648,8 +648,8 @@ def _dispatch_smoke_of(
         launch_human_attended_interactive,
         snapshot_live_checkout_for_smoke,
         tmux_available as _tmux_avail,
-        tmux_session_alive as _tmux_alive,
         tmux_session_name as _tmux_name,
+        tmux_session_running as _tmux_alive,
     )
 
     from coord.models import Assignment as _AssignmentSm  # noqa: PLC0415
@@ -1142,8 +1142,8 @@ def _dispatch_audit_of(
         finalize_interactive_exit,
         launch_human_attended_interactive,
         tmux_available as _tmux_avail,
-        tmux_session_alive as _tmux_alive,
         tmux_session_name as _tmux_name,
+        tmux_session_running as _tmux_alive,
     )
     from coord.models import Assignment as _AssignmentAu  # noqa: PLC0415
     from coord.state import record_dispatched_assignment as _record_au  # noqa: PLC0415
@@ -1486,8 +1486,8 @@ def _dispatch_milestone_chat_of(
         finalize_interactive_exit,
         launch_human_attended_interactive,
         tmux_available as _tmux_avail,
-        tmux_session_alive as _tmux_alive,
         tmux_session_name as _tmux_name,
+        tmux_session_running as _tmux_alive,
     )
     from coord.milestone_chat import resolve_milestone_chat_briefing  # noqa: PLC0415
     from coord.board_service import read_board as _read_board_mc  # noqa: PLC0415
@@ -1682,8 +1682,8 @@ def _run_troubleshoot_or_chat(
         finalize_interactive_exit,
         launch_human_attended_interactive,
         tmux_available as _tmux_avail,
-        tmux_session_alive as _tmux_alive,
         tmux_session_name as _tmux_name,
+        tmux_session_running as _tmux_alive,
     )
 
     from coord.board_service import read_board as _read_board_ts  # noqa: PLC0415
@@ -2044,8 +2044,8 @@ def _dispatch_fix_of(
         launch_human_attended_interactive,
         remote_worktree_exists as _remote_wt_exists,
         tmux_available as _tmux_avail,
-        tmux_session_alive as _tmux_alive,
         tmux_session_name as _tmux_name,
+        tmux_session_running as _tmux_alive,
     )
 
     from coord.auto_loop import (  # noqa: PLC0415
@@ -2731,8 +2731,8 @@ def _dispatch_rework_of(
         launch_human_attended_interactive,
         remote_worktree_exists as _remote_wt_exists,
         tmux_available as _tmux_avail,
-        tmux_session_alive as _tmux_alive,
         tmux_session_name as _tmux_name,
+        tmux_session_running as _tmux_alive,
     )
 
     from coord.agent import (  # noqa: PLC0415
@@ -3294,8 +3294,8 @@ def _dispatch_merge_of(
         launch_human_attended_interactive,
         remote_worktree_exists as _remote_wt_exists,
         tmux_available as _tmux_avail,
-        tmux_session_alive as _tmux_alive,
         tmux_session_name as _tmux_name,
+        tmux_session_running as _tmux_alive,
     )
 
     from coord.agent import (  # noqa: PLC0415
@@ -3932,8 +3932,8 @@ def _dispatch_interactive_work(
         launch_human_attended_interactive,
         remote_worktree_exists as _remote_wt_exists,
         tmux_available as _tmux_avail,
-        tmux_session_alive as _tmux_alive,
         tmux_session_name as _tmux_name,
+        tmux_session_running as _tmux_alive,
     )
 
     if _is_local:

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/coord/commands/milestone.py

@@ -48,6 +48,7 @@ from coord.milestone_order import (
     replace_sub_issues_section,
     replace_work_order_section,
     validate_milestone_membership,
+    validate_no_shared_oracle_group,
 )
 
 
@@ -1088,7 +1089,15 @@ def milestone_dispatch_cmd(
     # issue as its pre-reqs land; that also makes `register_milestone_drain`
     # redundant here (the queue IS the durable drain, and registering would
     # have the daemon's direct-dispatch drain racing it).
-    queue_plan = plan_queue(ctx.work_order, ctx.terminal_issues, repo_entry.name)
+    # #2542: under oracle-loop control (Gate A already checked above), chain
+    # the whole milestone's drive-queue entries so the tick can never launch
+    # two of them at once — see plan_queue's own docstring for why.
+    queue_plan = plan_queue(
+        ctx.work_order,
+        ctx.terminal_issues,
+        repo_entry.name,
+        oracle_loop=cfg.acceptance.has_driver(repo_entry.name),
+    )
 
     if not queue_plan:
         click.echo("Nothing to queue — every work-order node is already closed.")
@@ -1277,18 +1286,25 @@ def milestone_drive_cmd(
 
     # The frontier half of the dry-run report. Computed for both modes so the
     # non-dry-run confirmation shows what the first `work` tick will do.
+    # #2542: pass the same `oracle_loop` the gate tick itself will use, so
+    # this preview never shows more than the one entry the actual `work`
+    # gate would dispatch for an oracle-loop milestone.
     if ctx.work_order.nodes:
         plan = plan_dispatch(
-            ctx.work_order, board, cfg, repo_entry, ctx.terminal_issues
+            ctx.work_order, board, cfg, repo_entry, ctx.terminal_issues,
+            oracle_loop=cfg.acceptance.has_driver(repo_entry.name),
+        )
+        to_dispatch, skipped, waiting, deferred = (
+            plan.to_dispatch, plan.skipped, plan.waiting, plan.deferred,
         )
-        to_dispatch, skipped, waiting = plan.to_dispatch, plan.skipped, plan.waiting
     else:
-        to_dispatch, skipped, waiting = [], [], []
+        to_dispatch, skipped, waiting, deferred = [], [], [], []
 
     if dry_run:
         for line in mg.format_plan(
             record, steps,
             to_dispatch=to_dispatch, skipped=skipped, waiting=waiting,
+            deferred=deferred,
         ):
             click.echo(line)
         return
@@ -1311,7 +1327,7 @@ def milestone_drive_cmd(
     state.save_milestone_gate(record.to_dict())
     for line in mg.format_plan(
         record, steps, to_dispatch=to_dispatch, skipped=skipped, waiting=waiting,
-        footer=False,
+        deferred=deferred, footer=False,
     ):
         click.echo(line)
     click.echo()
@@ -1448,6 +1464,22 @@ def milestone_write_order_cmd(
         click.echo(f"error: {e}", err=True)
         sys.exit(1)
 
+    # #2542: refuse a proposed work order that puts two issues in the same
+    # `{group: ...}` cohort when this milestone is under oracle-loop control
+    # — every group member's JIT acceptance slice would additively append to
+    # the SAME tests/acceptance/ms-N/manifest.yml, a manifest race that
+    # exists regardless of whether the issues' actual feature work overlaps
+    # (coord-portal#122). Cheap/early check for the explicit-group case only
+    # — see validate_no_shared_oracle_group's docstring for what it can't
+    # catch.
+    try:
+        validate_no_shared_oracle_group(
+            work_order, oracle_loop=cfg.acceptance.has_driver(repo_entry.name)
+        )
+    except WorkOrderError as e:
+        click.echo(f"error: {e}", err=True)
+        sys.exit(1)
+
     # #1057: a valid `## Work order` block alone doesn't make this issue
     # findable as the milestone's tracking epic — `coord plans`/the TUI
     # Plans panel key off the `epic` label (TRACKING_ISSUE_LABEL). Ensure it

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/coord/commands/sessions.py

@@ -1084,6 +1084,7 @@ def reattach(assignment_id: str, config_path: Path) -> None:
         tmux_available,
         tmux_session_alive,
         tmux_session_name,
+        tmux_session_running,
     )
 
     if not tmux_available():
@@ -1609,7 +1610,12 @@ def reattach(assignment_id: str, config_path: Path) -> None:
         exit_code = 1
 
     # After attach returns: check if session ended or user detached.
-    if tmux_session_alive(sname, host=_tmux_host):
+    # #2541: use tmux_session_running (alive AND pane not dead), not the
+    # bare has-session check — remain-on-exit means a session whose claude
+    # already exited (crash OR a clean, successful completion) stays
+    # has-session-alive until a reaper notices, so a bare check here would
+    # misreport a finished session as "still running" and skip finalize.
+    if tmux_session_running(sname, host=_tmux_host):
         click.echo(
             f"\n  Session is still running.  "
             f"Reattach later with: coord reattach {assignment_id}"

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/coord/commands/status.py

@@ -1951,7 +1951,7 @@ def _diagnose_orphan_worktrees(config_path: Path, *, dry_run: bool) -> None:
     from coord.interactive import (  # noqa: PLC0415
         tmux_available,
         tmux_session_name,
-        tmux_session_alive,
+        tmux_session_running,
     )
     from coord.agent import check_worktree_writable, find_blocking_deny_rule  # noqa: PLC0415
     from coord.board_service import read_board  # noqa: PLC0415
@@ -2002,6 +2002,11 @@ def _diagnose_orphan_worktrees(config_path: Path, *, dry_run: bool) -> None:
         return
 
     # Collect all assignment_ids with live tmux sessions.
+    # #2541: tmux_session_running (alive AND pane not dead) — remain-on-exit
+    # keeps a session's has-session bit True after ANY pane exit (clean
+    # success or crash) until a reaper notices, so the bare alive check
+    # would count a finished session's worktree as still in use and skip it
+    # in this sweep.
     tmux_ok = tmux_available()
     live_tmux: set[str] = set()
     if tmux_ok:
@@ -2009,7 +2014,7 @@ def _diagnose_orphan_worktrees(config_path: Path, *, dry_run: bool) -> None:
             if not entry.is_dir():
                 continue
             aid = entry.name
-            if tmux_session_alive(tmux_session_name(aid)):
+            if tmux_session_running(tmux_session_name(aid)):
                 live_tmux.add(aid)
 
     # All running/pending assignment_ids from the board (includes live tmux ones

{code_coordinator-0.5.229 → code_coordinator-0.5.231}/coord/diagnose.py

@@ -254,8 +254,8 @@ def _session_state(assignment: "Assignment", config: "Config") -> str:
 
     from coord.interactive import (  # noqa: PLC0415
         TmuxHost,
-        tmux_session_alive,
         tmux_session_name,
+        tmux_session_running,
     )
 
     if not assignment.assignment_id:
@@ -276,7 +276,14 @@ def _session_state(assignment: "Assignment", config: "Config") -> str:
     host = TmuxHost(ssh_target=ssh_target)
     sname = tmux_session_name(assignment.assignment_id)
     try:
-        if tmux_session_alive(sname, host=host):
+        # #2541: tmux_session_running (alive AND pane not dead), not the
+        # bare has-session check this used to make — remain-on-exit keeps
+        # has-session True after ANY pane exit (clean success or crash)
+        # until a reaper notices, so a bare check here would report a
+        # crashed/finished --merge-of session as "live" and never fall
+        # through to the agent cross-check below, undermining exactly the
+        # diagnosability this probe exists for (#1658).
+        if tmux_session_running(sname, host=host):
             return "live"
     except Exception:  # noqa: BLE001 — never let a probe error finalize a session
         return "unknown"