From 0dec3756316213b3e50355607600f78685b1bdbb Mon Sep 17 00:00:00 2001
From: Alexandre Teixeira <111787685+alteixeira20@users.noreply.github.com>
Date: Thu, 1 Oct 2026 22:26:36 +0100
Subject: [PATCH] fix(runtime): release and report failed background supervisor
 setup

---
 src/containment_worker.py            | 11 ++++++++--
 tests/test_background_containment.py | 31 ++++++++++++++++++++++++++++
 2 files changed, 40 insertions(+), 2 deletions(-)

diff --git a/src/containment_worker.py b/src/containment_worker.py
index 196efe8aa..84edee8fd 100644
--- a/src/containment_worker.py
+++ b/src/containment_worker.py
@@ -57,6 +57,9 @@ async def supervise(payload: dict) -> None:
             output = "\n…[output truncated by containment capture limit]…\n"
     except BaseException as exc:
         record = containment._load_records().get(grant.id, {})
+        if not record.get("pid") and not record.get("release"):
+            containment.release(grant, grace_s=0)
+            record = containment._load_records().get(grant.id, {})
         output, code = f"background execution failed: {type(exc).__name__}: {exc}\n", 1
         report = {"containment": grant.to_dict(), "teardown": record.get("release") or {"dead": False},
                   "output_truncated": False}
@@ -66,8 +69,12 @@ async def supervise(payload: dict) -> None:
         if isinstance(exc, containment.ContainmentUnavailable):
             report.update(containment.unavailable_tool_result(exc, tool="bash"))
     if output:
-        with open(payload["log_path"], "a", encoding="utf-8") as log:
-            log.write(output)
+        try:
+            with open(payload["log_path"], "a", encoding="utf-8") as log:
+                log.write(output)
+        except OSError:
+            # A failed log initialization must not hide completion metadata.
+            sys.stderr.write(output)
     atomic_write_json(payload["result_path"], report)
     # Publish completion last: refresh must never see an exit without metadata.
     atomic_write_text(payload["exit_path"], str(code if code is not None else 1))
diff --git a/tests/test_background_containment.py b/tests/test_background_containment.py
index 52d01f483..664788173 100644
--- a/tests/test_background_containment.py
+++ b/tests/test_background_containment.py
@@ -52,6 +52,37 @@ def test_detached_execution_owns_boundary_and_reports_death(jobs):
     assert result["teardown"]["dead"] is True
 
 
+def test_supervisor_setup_failure_closes_unstarted_grant(jobs):
+    import json
+    import subprocess
+    import sys
+    from pathlib import Path
+    path, _ = jobs
+    spec = containment.agent_spec(str(path), dict(os.environ), 5)
+    grant = containment.acquire(spec, owner="failed-supervisor")
+    payload = {
+        "store_path": str(containment._store_path()),
+        "grant": {**grant.to_dict(), "owner": grant.owner},
+        "spec": {"workspace": str(path), "env": dict(spec.env), "wall_clock_s": 5,
+                 "required": sorted(spec.required)},
+        "command": "printf effect > must-not-exist",
+        "log_path": str(path / "missing-directory" / "job.log"),
+        "result_path": str(path / "result.json"), "exit_path": str(path / "exit"),
+    }
+    worker = Path(containment.__file__).with_name("containment_worker.py")
+    result = subprocess.run([sys.executable, str(worker)], input=json.dumps(payload),
+                            capture_output=True, text=True, timeout=10)
+    assert result.returncode == 0  # Supervisor publishes the failed job result.
+    assert "FileNotFoundError" in result.stderr
+    assert not (path / "must-not-exist").exists()
+    assert containment.active_grants() == []
+    assert (path / "exit").read_text() == "1"
+    report = json.loads((path / "result.json").read_text())
+    assert report["containment"]["executed"] is False
+    assert report["containment"]["contained"] is False
+    assert report["teardown"]["dead"] is True
+
+
 async def test_bg_marker_refuses_without_spawning_and_authority_still_gates(jobs, monkeypatch):
     path, _ = jobs
     monkeypatch.setattr(containment, "CONTAINMENT_MODE", containment.MODE_ENFORCING)