Fix: find_active_run treats skill artifacts as loop-state candidates (GH-36)

CHANGELOG.md

@@ -9,6 +9,10 @@ Versioning follows [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+### Fixed
+
+- **(GH-36)** `find_active_run` no longer mistakes a skill's own JSON artifact (e.g. `validate` writing `.dmx/jobs/{job_id}/validation-report.json`) for a second, permanently-non-terminal loop run. A file now has to actually look like loop state — carry `loop_name`, `task_id`, and `status` — before it's considered a candidate at all; anything else in the job directory is skipped outright rather than defaulting to "non-terminal" when `status` is absent. Previously, a correct `validate` run left two JSON files in the same directory and `loop_advance`/`loop_continue` raised `AmbiguousActiveRun` on every attempt to proceed, with no fix short of hand-editing the artifact to fake a `status: complete` it doesn't have.
+
 ## [0.4.0] — 2026-09-09
 
 ### Added

src/dmx/loop_state.py

@@ -277,6 +277,14 @@ def find_active_run(workspace_root: Path, job_id: str) -> tuple[str, str] | None
             next starts — so this should never happen in normal operation.
             Rather than guess which one is "active", this fails loudly so
             it can be investigated.
+
+    Note:
+        The job directory can hold non-state JSON artifacts a skill was
+        told to write there (e.g. ``validate`` writing
+        ``validation-report.json`` — see ``dmx-validate.md`` Step 9 and
+        ``validators/spec_adherence.py``). Those files have no
+        ``loop_name``/``task_id``/``status`` and are skipped rather than
+        misread as a second, permanently-non-terminal run (see GH-36).
     """
     job_dir = _job_dir(workspace_root, job_id)
     if not job_dir.exists():
@@ -288,6 +296,10 @@ def find_active_run(workspace_root: Path, job_id: str) -> tuple[str, str] | None
             data = json.loads(path.read_text(encoding="utf-8"))
         except (json.JSONDecodeError, OSError):
             continue
+        if not isinstance(data, dict) or not {"loop_name", "task_id", "status"} <= data.keys():
+            # Not a loop-state file (e.g. a skill artifact like
+            # validation-report.json) — not a run candidate at all.
+            continue
         if data.get("status") in _TERMINAL_STATUSES:
             continue
         candidates.append((data.get("loop_name", ""), data.get("task_id", "")))

tests/test_loop_state.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import json
 from typing import TYPE_CHECKING
 
 import pytest
@@ -234,6 +235,48 @@ def test_ignores_malformed_state_file(self, tmp_path: Path) -> None:
         (job_dir / "dev-broken.json").write_text("not json", encoding="utf-8")
         assert find_active_run(root, "PAY-1") is None
 
+    def test_ignores_skill_artifact_alongside_paused_run(self, tmp_path: Path) -> None:
+        """GH-36: a skill artifact like validate's validation-report.json
+        lives in the same job directory but has no loop_name/task_id/status
+        — it must not be mistaken for a second non-terminal run."""
+        root = self._workspace(tmp_path)
+        write_initial_state(root, "validate", "PAY-1", "task-1", ["s"])
+        write_state(root, "PAY-1", "validate", "task-1", {"status": LoopStatus.paused.value})
+        job_dir = root / ".dmx" / "jobs" / "PAY-1"
+        (job_dir / "validation-report.json").write_text(
+            json.dumps(
+                {
+                    "commit": "deadbeef",
+                    "scope_items": [],
+                    "scope_creep": [],
+                    "regressions": [],
+                    "edge_cases": [],
+                }
+            ),
+            encoding="utf-8",
+        )
+        assert find_active_run(root, "PAY-1") == ("validate", "task-1")
+
+    def test_ignores_skill_artifact_with_no_real_run(self, tmp_path: Path) -> None:
+        """The artifact alone (no loop-state file at all) must not surface
+        as an active run."""
+        root = self._workspace(tmp_path)
+        job_dir = root / ".dmx" / "jobs" / "PAY-1"
+        job_dir.mkdir(parents=True)
+        (job_dir / "validation-report.json").write_text(
+            json.dumps({"commit": "deadbeef", "scope_items": []}), encoding="utf-8"
+        )
+        assert find_active_run(root, "PAY-1") is None
+
+    def test_ignores_non_dict_json_file(self, tmp_path: Path) -> None:
+        """A JSON file that parses but isn't an object (e.g. a bare list)
+        must not crash the scan or be mistaken for a run."""
+        root = self._workspace(tmp_path)
+        job_dir = root / ".dmx" / "jobs" / "PAY-1"
+        job_dir.mkdir(parents=True)
+        (job_dir / "weird.json").write_text("[1, 2, 3]", encoding="utf-8")
+        assert find_active_run(root, "PAY-1") is None
+
 
 # ---------------------------------------------------------------------------
 # find_pending_run — resuming a loop before its real job id is resolvable