test(issue-5): add layer-1 herdr and layer-2 watcher test fixtures

2026-09-06 13:27:45 +09:00
1 parent 720774c1e7
commit f85fa5b6cf
@@ -0,0 +1,38 @@
+{
+  "schema_version": "1.0.0",
+  "scenario_id": "blocked_permission",
+  "backend": "backend_agnostic",
+  "provenance": "simulated",
+  "description": "Native status blocked on interactive permission prompt. Completion forbidden.",
+  "dispatch_context": {
+    "job_id": "test-job-blocked-01",
+    "attempt_id": 1,
+    "agent_session_uuid": "1c53dd89-81fc-49a6-b15e-bef3dfb61697",
+    "pane_process_generation": "w2X:p3:5678:2026-09-06T00:00:00Z",
+    "dispatch_ack": true,
+    "execution_anchor_observed": true
+  },
+  "timeline": [
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 0,
+      "source_kind": "herdr",
+      "source_provenance": "simulated",
+      "payload": {
+        "agent_status": "blocked",
+        "interactive_ready": false,
+        "matched_rule": "permission_prompt"
+      }
+    }
+  ],
+  "filesystem": {
+    "report_path": ".mam/jobs/test-job-blocked-01/report-final.md",
+    "exists": false,
+    "manifest_committed": false
+  },
+  "expected_policy": {
+    "decision": "hold_blocked",
+    "reason": "agent terminal is blocked awaiting user permission",
+    "policy_version": "2.0.0"
+  }
+}
@@ -0,0 +1,61 @@
+{
+  "schema_version": "1.0.0",
+  "scenario_id": "finalized_candidate",
+  "backend": "backend_agnostic",
+  "provenance": "simulated",
+  "description": "Execution anchor verified, native status idle/done, atomic rename complete with valid SHA-256 commit manifest. Inferred candidate approved for CAS.",
+  "dispatch_context": {
+    "job_id": "test-job-finalized-01",
+    "attempt_id": 1,
+    "agent_session_uuid": "7d5700d0-bab1-4f65-9665-6b887396a9a6",
+    "pane_process_generation": "w2X:p1:1234:2026-09-06T00:00:00Z",
+    "dispatch_ack": true,
+    "execution_anchor_observed": true
+  },
+  "timeline": [
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 0,
+      "source_kind": "event",
+      "source_provenance": "simulated",
+      "payload": {
+        "event": "started",
+        "job_id": "test-job-finalized-01"
+      }
+    },
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 5000,
+      "source_kind": "filesystem",
+      "source_provenance": "simulated",
+      "payload": {
+        "report_manifest": {
+          "report_hash": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+          "attempt_id": 1,
+          "finalized": true
+        }
+      }
+    },
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 6000,
+      "source_kind": "herdr",
+      "source_provenance": "simulated",
+      "payload": {
+        "agent_status": "idle",
+        "interactive_ready": true
+      }
+    }
+  ],
+  "filesystem": {
+    "report_path": ".mam/jobs/test-job-finalized-01/report-final.md",
+    "exists": true,
+    "manifest_committed": true,
+    "hash_verified": true
+  },
+  "expected_policy": {
+    "decision": "candidate_approved",
+    "reason": "all 4-fold attribution, fresh idle status, and atomic finalization manifest verified",
+    "policy_version": "2.0.0"
+  }
+}
@@ -0,0 +1,47 @@
+{
+  "schema_version": "1.0.0",
+  "scenario_id": "initial_idle",
+  "backend": "backend_agnostic",
+  "provenance": "simulated",
+  "description": "Dispatch record exists and native status is idle, but no execution anchor observed since dispatch_ack. Inferred completion must be held.",
+  "dispatch_context": {
+    "job_id": "test-job-init-idle-01",
+    "attempt_id": 1,
+    "agent_session_uuid": "7d5700d0-bab1-4f65-9665-6b887396a9a6",
+    "pane_process_generation": "w2X:p1:1234:2026-09-06T00:00:00Z",
+    "dispatch_ack": true,
+    "execution_anchor_observed": false
+  },
+  "timeline": [
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 0,
+      "source_kind": "event",
+      "source_provenance": "simulated",
+      "payload": {
+        "event": "dispatched",
+        "job_id": "test-job-init-idle-01"
+      }
+    },
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 500,
+      "source_kind": "herdr",
+      "source_provenance": "simulated",
+      "payload": {
+        "agent_status": "idle",
+        "interactive_ready": true
+      }
+    }
+  ],
+  "filesystem": {
+    "report_path": ".mam/jobs/test-job-init-idle-01/report-final.md",
+    "exists": false,
+    "manifest_committed": false
+  },
+  "expected_policy": {
+    "decision": "hold",
+    "reason": "no execution anchor observed for this attempt",
+    "policy_version": "2.0.0"
+  }
+}
@@ -0,0 +1,48 @@
+{
+  "schema_version": "1.0.0",
+  "description": "Multi-channel Watcher composite timeline scenario manifest",
+  "scenarios": {
+    "initial_idle": {
+      "file": "initial_idle.json",
+      "provenance": "simulated",
+      "expected_decision": "hold",
+      "description": "Dispatch record exists and native status is idle, but no execution anchor observed since dispatch_ack. Inferred completion must be held."
+    },
+    "tool_running": {
+      "file": "tool_running.json",
+      "provenance": "simulated",
+      "expected_decision": "hold",
+      "description": "Native status working or interactive_ready with active background tool execution. Inferred completion must be held."
+    },
+    "report_pause_15s": {
+      "file": "report_pause_15s.json",
+      "provenance": "simulated",
+      "expected_decision": "hold",
+      "description": "Partial report written and stable for >= 15s debounce, but lacks atomic finalization manifest. Completion must be held."
+    },
+    "finalized_candidate": {
+      "file": "finalized_candidate.json",
+      "provenance": "simulated",
+      "expected_decision": "candidate_approved",
+      "description": "Execution anchor verified, native status idle/done, atomic rename complete with valid SHA-256 commit manifest. Inferred candidate approved for CAS."
+    },
+    "blocked_permission": {
+      "file": "blocked_permission.json",
+      "provenance": "simulated",
+      "expected_decision": "hold_blocked",
+      "description": "Native status blocked on interactive permission prompt. Completion forbidden."
+    },
+    "transport_disconnected": {
+      "file": "transport_disconnected.json",
+      "provenance": "simulated",
+      "expected_decision": "hold_reconnect",
+      "description": "Herdr socket connection failure or freshness gap. Cached evidence invalidated; must reconnect and resnapshot."
+    },
+    "restarted_generation": {
+      "file": "restarted_generation.json",
+      "provenance": "simulated",
+      "expected_decision": "reject_stale",
+      "description": "Pane process restarted with new PID/start time; stale attempt rejected."
+    }
+  }
+}
@@ -0,0 +1,60 @@
+{
+  "schema_version": "1.0.0",
+  "scenario_id": "report_pause_15s",
+  "backend": "backend_agnostic",
+  "provenance": "simulated",
+  "description": "Partial report written and stable for >= 15s debounce, but lacks atomic finalization manifest. Completion must be held.",
+  "dispatch_context": {
+    "job_id": "test-job-pause-01",
+    "attempt_id": 1,
+    "agent_session_uuid": "1c53dd89-81fc-49a6-b15e-bef3dfb61697",
+    "pane_process_generation": "w2X:p3:5678:2026-09-06T00:00:00Z",
+    "dispatch_ack": true,
+    "execution_anchor_observed": true
+  },
+  "timeline": [
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 0,
+      "source_kind": "filesystem",
+      "source_provenance": "simulated",
+      "payload": {
+        "file": "report-final.md.tmp",
+        "bytes": 250,
+        "phase": "writing"
+      }
+    },
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 16000,
+      "source_kind": "filesystem",
+      "source_provenance": "simulated",
+      "payload": {
+        "file": "report-final.md.tmp",
+        "bytes": 250,
+        "mtime_stale_sec": 16.0
+      }
+    },
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 16100,
+      "source_kind": "herdr",
+      "source_provenance": "simulated",
+      "payload": {
+        "agent_status": "working",
+        "interactive_ready": true
+      }
+    }
+  ],
+  "filesystem": {
+    "report_path": ".mam/jobs/test-job-pause-01/report-final.md",
+    "exists": false,
+    "tmp_file_exists": true,
+    "manifest_committed": false
+  },
+  "expected_policy": {
+    "decision": "hold",
+    "reason": "debounce elapsed on partial file, but report is unfinalized without manifest",
+    "policy_version": "2.0.0"
+  }
+}
@@ -0,0 +1,38 @@
+{
+  "schema_version": "1.0.0",
+  "scenario_id": "restarted_generation",
+  "backend": "backend_agnostic",
+  "provenance": "simulated",
+  "description": "Pane process restarted with new PID/start time; stale attempt rejected.",
+  "dispatch_context": {
+    "job_id": "test-job-restarted-01",
+    "attempt_id": 1,
+    "agent_session_uuid": "7d5700d0-bab1-4f65-9665-6b887396a9a6",
+    "pane_process_generation": "w2X:p1:1234:2026-09-06T00:00:00Z",
+    "dispatch_ack": true,
+    "execution_anchor_observed": false
+  },
+  "timeline": [
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 0,
+      "source_kind": "herdr",
+      "source_provenance": "simulated",
+      "payload": {
+        "pane_id": "w2X:p1",
+        "pane_process_generation": "w2X:p1:9999:2026-09-06T01:00:00Z",
+        "agent_status": "idle"
+      }
+    }
+  ],
+  "filesystem": {
+    "report_path": ".mam/jobs/test-job-restarted-01/report-final.md",
+    "exists": false,
+    "manifest_committed": false
+  },
+  "expected_policy": {
+    "decision": "reject_stale",
+    "reason": "observed process generation does not match active dispatch attempt",
+    "policy_version": "2.0.0"
+  }
+}
@@ -0,0 +1,48 @@
+{
+  "schema_version": "1.0.0",
+  "scenario_id": "tool_running",
+  "backend": "backend_agnostic",
+  "provenance": "simulated",
+  "description": "Native status working or interactive_ready with active background tool execution. Inferred completion must be held.",
+  "dispatch_context": {
+    "job_id": "test-job-tool-running-01",
+    "attempt_id": 1,
+    "agent_session_uuid": "1c53dd89-81fc-49a6-b15e-bef3dfb61697",
+    "pane_process_generation": "w2X:p3:5678:2026-09-06T00:00:00Z",
+    "dispatch_ack": true,
+    "execution_anchor_observed": true
+  },
+  "timeline": [
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 0,
+      "source_kind": "herdr",
+      "source_provenance": "simulated",
+      "payload": {
+        "agent_status": "working",
+        "interactive_ready": true
+      }
+    },
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 200,
+      "source_kind": "harness",
+      "source_provenance": "simulated",
+      "payload": {
+        "tool": "bash",
+        "active": true,
+        "command": "pytest"
+      }
+    }
+  ],
+  "filesystem": {
+    "report_path": ".mam/jobs/test-job-tool-running-01/report-final.md",
+    "exists": false,
+    "manifest_committed": false
+  },
+  "expected_policy": {
+    "decision": "hold",
+    "reason": "external tool execution currently active",
+    "policy_version": "2.0.0"
+  }
+}
@@ -0,0 +1,37 @@
+{
+  "schema_version": "1.0.0",
+  "scenario_id": "transport_disconnected",
+  "backend": "backend_agnostic",
+  "provenance": "simulated",
+  "description": "Herdr socket connection failure or freshness gap. Cached evidence invalidated; must reconnect and resnapshot.",
+  "dispatch_context": {
+    "job_id": "test-job-disconnect-01",
+    "attempt_id": 1,
+    "agent_session_uuid": "7d5700d0-bab1-4f65-9665-6b887396a9a6",
+    "pane_process_generation": "w2X:p1:1234:2026-09-06T00:00:00Z",
+    "dispatch_ack": true,
+    "execution_anchor_observed": true
+  },
+  "timeline": [
+    {
+      "collector_clock_id": "CLOCK_MONOTONIC",
+      "relative_ms": 0,
+      "source_kind": "herdr",
+      "source_provenance": "simulated",
+      "payload": {
+        "error": "connection_refused",
+        "transport_healthy": false
+      }
+    }
+  ],
+  "filesystem": {
+    "report_path": ".mam/jobs/test-job-disconnect-01/report-final.md",
+    "exists": false,
+    "manifest_committed": false
+  },
+  "expected_policy": {
+    "decision": "hold_reconnect",
+    "reason": "transport connection failure; previous idle observations cannot be trusted",
+    "policy_version": "2.0.0"
+  }
+}