EPS
← All tasks·#2019Completed

workflow-fix: scratch-eligibility for the workflow-lint scan node in step9c compare

kind: infra#wf-fix#wf-fix-fp:a85a7ae2235e
track:

Highlight text on any card (body, plan, or an event) to anchor a comment, or leave a whole-task comment here. Mention @claude to summon a reply.

No comments yet.

Overview / Motivation

Auto-filed by the workflow-fix-on-bug protocol from a workflow-fix candidate raised on task #1932 (emitting agent: orchestrator, /issue 1932 Step 9c).

Goal

Make the tests/test_workflow_lint.py no-flags scan node scratch-resolvable in scripts/step9c_baseline.py (add to FILE_ANCHORED_SCAN_TESTS, or an equivalent scratch-eligibility registry entry with source-verified anchoring notes) so compare --run-pristine can pristine-resolve a red on it from the default scratch-worktree oracle instead of going indeterminate on a dirty shared root.

Workflow gap

  • Bug observed: step9c_baseline.py compare returns indeterminate (exit 2, MF-4c dirty-root refusal) for a red on tests/test_workflow_lint.py::test_workflow_lint_default_exits_zero whenever the shared root carries visible code dirt, because the node is scratch-ineligible (absent from FILE_ANCHORED_SCAN_TESTS) — despite the test resolving its scan root from its own __file__ (_REPO_ROOT = Path(__file__).resolve().parents[1], tests/test_workflow_lint.py:30; subprocess cwd=_REPO_ROOT, :110), i.e. scratch-local by the same criterion the registry's existing members satisfy.
  • Why it is a workflow gap: the shared repo root is dirty essentially always (concurrent sessions' untracked drafts + in-flight agent-memory writes), and test_workflow_lint_default_exits_zero is the single most red-prone invariant node (near-cap size-ratchet files: a concurrent uncommitted MEMORY.md write pushed it over on #1932's gate). Every such gate run goes indeterminate and forces a manual orchestrator classification — #1932 (2026-08-02) hand-built the exact scratch run (1 passed in 381.58s in a detached sparse scratch worktree at HEAD) that the compare would have run itself were the node registry-eligible.
  • Confidence (emitter): medium — the eligibility bar requires reading the WHOLE test file's live-tree anchoring (all nodes, conftest channels), not just the one node; the planner makes that deliberate call with the source open (the registry comment mandates source-verified anchoring notes + the test_file_anchored_scan_tests_live_tree_pin drift pin must be updated).
  • verified-at-filing: grep -n 'FILE_ANCHORED_SCAN_TESTS' scripts/step9c_baseline.py → 5 hits (definition at :206; members: test_shared_vm_thread_caps.py, test_subprocess_env_explicit.py, test_select_step9c_tests.py per the pinned literal); grep -c 'test_workflow_lint' scripts/step9c_baseline.py → 2 (neither a registry membership — absence confirmed) (2026-08-02, this session). Empirical demonstration: /tmp/issue-1932-scratch-node.log (sparse scratch at HEAD, node PASSes rc=0).

Proposed change (candidate diff sketch — refine in planning)

 FILE_ANCHORED_SCAN_TESTS: frozenset[str] = frozenset(
     {
         ...
+        # _REPO_ROOT = Path(__file__).resolve().parents[1] (:30); _run() subprocess-runs
+        # scripts/workflow_lint.py from the test file's own tree with cwd=_REPO_ROOT (:110);
+        # <planner: verify remaining nodes/conftest channels are scratch-local before adding>
+        "tests/test_workflow_lint.py",
     }
 )

Plus the matching update to tests/test_step9c_baseline.py::test_file_anchored_scan_tests_live_tree_pin and, if the whole-file bar fails, an alternative node-scoped eligibility mechanism.

Scope / surfaces

  • Primary target: scripts/step9c_baseline.py
  • Secondary: tests/test_step9c_baseline.py (drift pin).
  • Grep the workflow surface for the pattern before editing (grep -rln 'FILE_ANCHORED_SCAN_TESTS' .claude/ CLAUDE.md scripts/ tests/) and update every hit; list them in the plan.

Constraints / invariants

  • Workflow-surface only — never experiment code, configs/, or tasks/.
  • FAIL-CLOSED doctrine preserved: a scan test absent from the registry keeps the refusal; eligibility is added only with source-verified anchoring (the registry's own curation rule).
  • scripts/workflow_lint.py --check-asks passes; ruff on touched files passes.
  • This session runs under EPM_WORKFLOW_FIX_SESSION=1 and carries a workflow_fix_target: Provenance line — it MUST NOT auto-route any of its own subagents' workflow-fix candidates (recursion guard).

Provenance

  • workflow_fix_target: scripts/step9c_baseline.py
  • fingerprint: a85a7ae2235e

target_file: scripts/step9c_baseline.py bug_observed: step9c compare returns indeterminate (exit 2) for a red on tests/test_workflow_lint.py::test_workflow_lint_default_exits_zero whenever the shared root carries visible code dirt, because the node is scratch-ineligible despite resolving its scan root from its own file why_workflow_gap: the always-dirty shared root makes every such gate run indeterminate, forcing manual orchestrator classification of the most red-prone invariant node (#1932, 2026-08-02: hand-built scratch run PASSed at HEAD) proposed_change: make the tests/test_workflow_lint.py no-flags scan node scratch-resolvable in step9c_baseline.py (FILE_ANCHORED_SCAN_TESTS or an equivalent scratch-eligibility registry) so compare can pristine-resolve it on a dirty shared root diff_sketch: |

  • "tests/test_workflow_lint.py", # in FILE_ANCHORED_SCAN_TESTS, with source-verified
  •                              # anchoring note (:30 __file__ root; :110 cwd=_REPO_ROOT)

(+ update tests/test_step9c_baseline.py::test_file_anchored_scan_tests_live_tree_pin) confidence: medium related_task: #1932

Activity