devflow-engine 1.1.26__tar.gz → 1.1.28__tar.gz

{devflow_engine-1.1.26 → devflow_engine-1.1.28}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: devflow-engine
-Version: 1.1.26
+Version: 1.1.28
 Summary: DevFlow v2 CLI-first engine (projects, stories, execution store, review packets)
 Author: DevFlow
 License: Proprietary

{devflow_engine-1.1.26 → devflow_engine-1.1.28}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "devflow-engine"
-version = "1.1.26"
+version = "1.1.28"
 description = "DevFlow v2 CLI-first engine (projects, stories, execution store, review packets)"
 readme = "README.md"
 requires-python = ">=3.11"

{devflow_engine-1.1.26 → devflow_engine-1.1.28}/src/devflow_engine/__init__.py

@@ -1,3 +1,3 @@
 __all__ = ["__version__"]
 
-__version__ = "1.1.24"
+__version__ = "1.1.28"

{devflow_engine-1.1.26 → devflow_engine-1.1.28}/src/devflow_engine/errors/runtime_observability.py

@@ -111,6 +111,45 @@ _FINAL_REQUIRED_KEYS = (
 )
 
 
+def _coerce_repro_artifact_to_observability_sandbox(
+    *,
+    state: dict[str, Any],
+    sandbox_rel: str,
+) -> None:
+    """Annotate repo-mutating repro artifacts without invalidating evidence.
+
+    The observe phase may create evidence artifacts, but they must live under
+    `.devflow/observability/<error_task_id>/`. If an agent writes a repro test
+    into the product test suite, that is useful draft work and must be cleaned
+    up by the workspace-preservation path. It is not, by itself, proof that the
+    reported symptom was unobserved.
+    """
+    artifact = str(state.get("repro_artifact_path") or "").strip()
+    command = str(state.get("repro_run_command") or "").strip()
+    if not artifact and not command:
+        return
+
+    sandbox_prefix = sandbox_rel.rstrip("/") + "/"
+    artifact_ok = artifact == sandbox_rel or artifact.startswith(sandbox_prefix)
+    command_ok = (not command) or (sandbox_prefix in command or sandbox_rel in command)
+    if artifact_ok and command_ok:
+        return
+
+    note = (
+        "Repro artifact was outside the observability sandbox; observation "
+        "requires artifacts under "
+        f"{sandbox_prefix} so observe does not mutate product tests."
+    )
+    if state.get("observed") is not True:
+        state["confidence"] = "low"
+    state["repro_output"] = "\n".join(
+        part for part in [str(state.get("repro_output") or "").strip(), note] if part
+    )[-3000:]
+    deferred = state.setdefault("deferred_questions", [])
+    if note not in deferred:
+        deferred.append(note)
+
+
 def _looks_like_file_hit(item: Any) -> bool:
     return (
         isinstance(item, dict)
@@ -440,6 +479,10 @@ def _loop_observe_and_resolve(
 
     state["_loop_iterations_used"] = len(iteration_history)
     state["_loop_history"] = iteration_history
+    _coerce_repro_artifact_to_observability_sandbox(
+        state=state,
+        sandbox_rel=sandbox_rel,
+    )
     # Empty out open_questions on exit if the loop ran to budget — any
     # still-unresolved questions become deferred (the loop is over;
     # they're now runtime-or-later questions by default).

{devflow_engine-1.1.26 → devflow_engine-1.1.28}/tests/test_observability_evidence_contract.py

@@ -76,3 +76,44 @@ def test_required_keys_encode_the_evidence_contract():
         assert k in ro._PHASE1_REQUIRED_KEYS
     for k in ("reported_symptom", "candidate_causes", "evidence"):
         assert k in ro._FINAL_REQUIRED_KEYS
+
+
+def test_repro_artifact_outside_observability_sandbox_is_hygiene_warning_not_evidence_failure() -> None:
+    state = {
+        "observed": True,
+        "confidence": "high",
+        "repro_artifact_path": "tests/playwright/repro.spec.js",
+        "repro_run_command": "npx playwright test tests/playwright/repro.spec.js",
+        "repro_output": "passed: reproduced visible panel",
+        "deferred_questions": [],
+    }
+
+    ro._coerce_repro_artifact_to_observability_sandbox(
+        state=state,
+        sandbox_rel=".devflow/observability/bug-1",
+    )
+
+    assert state["observed"] is True
+    assert state["confidence"] == "high"
+    assert "outside the observability sandbox" in state["repro_output"]
+    assert state["deferred_questions"]
+
+
+def test_repro_artifact_inside_observability_sandbox_remains_evidence() -> None:
+    state = {
+        "observed": True,
+        "confidence": "high",
+        "repro_artifact_path": ".devflow/observability/bug-1/repro.spec.js",
+        "repro_run_command": "npx playwright test .devflow/observability/bug-1/repro.spec.js",
+        "repro_output": "failed as expected",
+        "deferred_questions": [],
+    }
+
+    ro._coerce_repro_artifact_to_observability_sandbox(
+        state=state,
+        sandbox_rel=".devflow/observability/bug-1",
+    )
+
+    assert state["observed"] is True
+    assert state["confidence"] == "high"
+    assert state["deferred_questions"] == []

{devflow_engine-1.1.26 → devflow_engine-1.1.28}/uv.lock

@@ -172,7 +172,7 @@ wheels = [
 
 [[package]]
 name = "devflow-engine"
-version = "1.1.26"
+version = "1.1.27"
 source = { editable = "." }
 dependencies = [
     { name = "cryptography" },