fix: scope OpenSpec sentinel per-change to prevent stale task queue across runs

The .openspec-populated sentinel was a boolean marker (touch file) that only
tracked whether OpenSpec tasks had been loaded, not which change they came from.
This caused three bugs:

1. Switching between OpenSpec changes reused stale tasks from the previous change
2. Running without --openspec still served leftover OpenSpec tasks to agents
3. Task ID collisions (openspec-N.M format) prevented new tasks from loading
   when old IDs matched via deduplication

Fix:
- Sentinel now stores the full change path; populate_openspec_queue() compares
  it on read and purges stale tasks when the change switches
- CLI clears sentinel before adapter runs (--openspec provided) and removes all
  OpenSpec artifacts when --openspec is absent
- Task IDs now include the change name (openspec-{name}-N.M) to prevent
  cross-change collisions

Closes #150

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Author: 2 people

autonomy/loki

@@ -902,6 +902,9 @@ cmd_start() {
         # Export for run.sh to access
         export OPENSPEC_CHANGE_PATH="$openspec_change_path"
 
+        # Clear stale OpenSpec sentinel (will be re-created by run.sh for current change)
+        rm -f "$LOKI_DIR/queue/.openspec-populated"
+
         # Ensure .loki directory exists for adapter output
         mkdir -p "$LOKI_DIR"
 
@@ -934,6 +937,33 @@ cmd_start() {
         fi
     fi
 
+    # When --openspec is not provided, clean up leftover OpenSpec state
+    if [[ -z "$openspec_change_path" ]]; then
+        rm -f "$LOKI_DIR/queue/.openspec-populated"
+        rm -f "$LOKI_DIR/openspec-tasks.json"
+        rm -f "$LOKI_DIR/openspec-prd-normalized.md"
+        rm -rf "$LOKI_DIR/openspec/"
+        # Also purge any stale OpenSpec tasks from pending queue
+        # NOTE: Similar purge logic exists in autonomy/run.sh (change-switch cleanup)
+        if [[ -f "$LOKI_DIR/queue/pending.json" ]]; then
+            LOKI_QUEUE_DIR="$LOKI_DIR" python3 -c "
+import json, os
+try:
+    queue_dir = os.environ['LOKI_QUEUE_DIR']
+    pending_path = os.path.join(queue_dir, 'queue', 'pending.json')
+    with open(pending_path) as f:
+        tasks = json.load(f)
+    filtered = [t for t in tasks if not (isinstance(t, dict) and t.get('source') == 'openspec')]
+    if len(filtered) < len(tasks):
+        with open(pending_path, 'w') as f:
+            json.dump(filtered, f, indent=2)
+        print(f'Cleaned {len(tasks) - len(filtered)} stale OpenSpec tasks from queue')
+except Exception as e:
+    print(f'Warning: OpenSpec queue cleanup failed: {e}', file=__import__('sys').stderr)
+" 2>/dev/null
+        fi
+    fi
+
     # MiroFish market validation (optional, non-blocking)
     if [[ -z "$mirofish_url" ]] && [[ -n "${LOKI_MIROFISH_URL:-}" ]] && [[ "$mirofish_disabled" != "true" ]]; then
         mirofish_url="$LOKI_MIROFISH_URL"

autonomy/openspec-adapter.py

@@ -345,9 +345,13 @@ def _parse_scenario(name: str, body: str) -> Dict[str, Any]:
 
 # -- Tasks Parsing ------------------------------------------------------------
 
-def parse_tasks(tasks_path: Path) -> Tuple[List[Dict[str, Any]], Dict[str, Dict[str, Any]]]:
+def parse_tasks(tasks_path: Path, change_name: str = "") -> Tuple[List[Dict[str, Any]], Dict[str, Dict[str, Any]]]:
     """Parse tasks.md into structured task list and source map.
 
+    Args:
+        tasks_path: Path to tasks.md
+        change_name: Name of the OpenSpec change (used to scope task IDs)
+
     Returns:
         (tasks_list, source_map)
         tasks_list: list of task objects
@@ -373,7 +377,7 @@ def parse_tasks(tasks_path: Path) -> Tuple[List[Dict[str, Any]], Dict[str, Dict[
             checked = task_match.group(1).lower() == "x"
             task_id_num = task_match.group(2)
             description = task_match.group(3).strip()
-            task_id = f"openspec-{task_id_num}"
+            task_id = f"openspec-{change_name}-{task_id_num}" if change_name else f"openspec-{task_id_num}"
 
             task = {
                 "id": task_id,
@@ -696,7 +700,7 @@ def run(
     source_map: Dict[str, Dict[str, Any]] = {}
     tasks_path = change_dir / "tasks.md"
     if tasks_path.exists():
-        tasks_list, source_map = parse_tasks(tasks_path)
+        tasks_list, source_map = parse_tasks(tasks_path, change_name=change_name)
 
     # -- Parse design.md (optional) --
     design_data: Optional[Dict[str, str]] = None

autonomy/run.sh

@@ -8740,10 +8740,33 @@ populate_openspec_queue() {
         return 0
     fi
 
-    # Skip if already populated (marker file)
+    # Skip if already populated for the SAME change (sentinel stores change path)
     if [[ -f ".loki/queue/.openspec-populated" ]]; then
-        log_info "OpenSpec queue already populated, skipping"
-        return 0
+        local stored_change
+        stored_change="$(cat ".loki/queue/.openspec-populated" 2>/dev/null)"
+        if [[ "$stored_change" == "$OPENSPEC_CHANGE_PATH" ]]; then
+            log_info "OpenSpec queue already populated for this change, skipping"
+            return 0
+        else
+            log_info "OpenSpec change switched (was: $stored_change, now: $OPENSPEC_CHANGE_PATH) -- purging stale tasks"
+            # Purge stale OpenSpec tasks from pending queue before re-populating
+            # NOTE: Similar purge logic exists in autonomy/loki (no-openspec cleanup)
+            if [[ -f ".loki/queue/pending.json" ]]; then
+                python3 -c "
+import json
+try:
+    with open('.loki/queue/pending.json') as f:
+        tasks = json.load(f)
+    filtered = [t for t in tasks if not (isinstance(t, dict) and t.get('source') == 'openspec')]
+    if len(filtered) < len(tasks):
+        with open('.loki/queue/pending.json', 'w') as f:
+            json.dump(filtered, f, indent=2)
+        print(f'Purged {len(tasks) - len(filtered)} stale OpenSpec tasks from queue')
+except Exception as e:
+    print(f'Warning: OpenSpec queue purge failed: {e}', file=__import__('sys').stderr)
+" 2>/dev/null
+            fi
+        fi
     fi
 
     log_step "Populating task queue from OpenSpec tasks..."
@@ -8816,8 +8839,8 @@ OPENSPEC_QUEUE_EOF
         return 0
     fi
 
-    # Mark as populated so we don't re-add on restart
-    touch ".loki/queue/.openspec-populated"
+    # Mark as populated for this specific change so we don't re-add on restart
+    echo "$OPENSPEC_CHANGE_PATH" > ".loki/queue/.openspec-populated"
     log_info "OpenSpec queue population complete"
 }
 

skills/openspec-integration.md

@@ -76,7 +76,7 @@ Tasks are generated from OpenSpec `tasks.md` and loaded into `.loki/queue/pendin
 
 ```json
 {
-  "id": "openspec-1.3",
+  "id": "openspec-session-hardening-1.3",
   "title": "Implement session timeout change",
   "description": "[OpenSpec] Authentication: Implement session timeout change",
   "priority": "medium",
@@ -143,9 +143,9 @@ The verification map tracks each scenario with `"verified": false` initially. Af
 
 ```json
 {
-  "openspec-1.1": { "file": "tasks.md", "line": 3, "group": "Authentication" },
-  "openspec-1.2": { "file": "tasks.md", "line": 4, "group": "Authentication" },
-  "openspec-2.1": { "file": "tasks.md", "line": 7, "group": "Dashboard" }
+  "openspec-session-hardening-1.1": { "file": "tasks.md", "line": 3, "group": "Authentication" },
+  "openspec-session-hardening-1.2": { "file": "tasks.md", "line": 4, "group": "Authentication" },
+  "openspec-session-hardening-2.1": { "file": "tasks.md", "line": 7, "group": "Dashboard" }
 }
 ```
 

tests/test_openspec_adapter.py

@@ -163,16 +163,19 @@ def test_task_groups_extracted(self, output_dir):
         assert len(groups) >= 3, f"Expected at least 3 groups, got {groups}"
 
     def test_task_ids_hierarchical(self, output_dir):
-        """Task IDs should be in openspec-N.M format."""
+        """Task IDs should be in openspec-{change_name}-N.M format."""
         rc, stdout, _ = run_adapter(FIXTURES_DIR / "simple-feature", output_dir, "--json")
         assert rc == 0
         data = json.loads(stdout)
         for task in data["tasks"]:
             assert task["id"].startswith("openspec-"), f"ID {task['id']} should start with openspec-"
-            # Should match openspec-N.M pattern
-            suffix = task["id"].replace("openspec-", "")
-            parts = suffix.split(".")
-            assert len(parts) == 2, f"ID suffix {suffix} should be N.M"
+            # Should match openspec-{change_name}-N.M pattern
+            prefix = "openspec-simple-feature-"
+            assert task["id"].startswith(prefix), \
+                f"ID {task['id']} should start with {prefix}"
+            num_part = task["id"][len(prefix):]  # e.g., "1.1"
+            parts = num_part.split(".")
+            assert len(parts) == 2, f"Numeric suffix {num_part} should be N.M"
             assert parts[0].isdigit() and parts[1].isdigit()
 
     def test_malformed_tasks_no_checkboxes(self, output_dir):