Skip to content

Historical field note

This record preserves prior investigation or planning evidence. It is not a current operating contract; use the numbered book and current runbooks for instructions.

Resolution: task_97627e43b1034100831e726f8981e5e2

Task: Act on the triage verdict from the prerequisite task (task_8bf37845abf445149d99fb4a1e3a41d5) to satisfy the remaining acceptance criteria of the parent review (task_b56b24a5d319476d88d6ff8eec05944b, "Investigate low-confidence dream finding: skill"). Finding under review: dream-cycle repair finding dreamrepair:bf68720937810742ef53e6bb00fcffd5 (kind failure_pattern, scope project, confidence low). Affected label: Skills:'skill' only (no tools, providers, or other repo areas). Skill surface in scope: src/mac/skill_auto_repair.py, skill docs under src/mac/_hermes/skills/ and skills/, and the skill contract tests tests/test_skill_auto_repair.py, tests/test_fleet_skills.py, tests/test_hermes_skills_dedup.py. Resolution date: 2026-07-16 Resolved by: fleet worker (documented closure; no skill or tool edits, per the triage verdict and task scope).

Chosen resolution: DOCUMENTED CLOSURE — case (b), NOT ACTIONABLE

The prerequisite triage (docs/assessment-task-8bf378.md) determined the finding is not actionable as a skill defect: it is an environment-prerequisite artifact backed by a single low-confidence evidence record with a real evidence gap. Per the task branch for that outcome, the correct deliverable is a written closure with the reason and the specific evidence gap, and no change to skill code, skill docs, or tools. This document records that closure and its verification.

Why no skill/tool edit is warranted

  • The named skill surface is healthy. The skill module imports and runs, and every skill-scoped contract test passes (see verification below). There is no reproducer and no failing skill test to repair, so a "smallest repair" would be a speculative change with nothing to fix — explicitly disallowed by the task.
  • The finding's confidence/kind are consolidation artifacts. The finding is keyed by repair_fingerprint() over a dream candidate, not a pointer into a skill module. nap_consolidator._dream_kind() buckets any record whose text contains failure/failed/error as failure_pattern, and _confidence_for_records() returns low for a single supporting record. The one evidence record is a prior repair failure, so both the failure_pattern kind and the low confidence follow mechanically from the heuristics acting on one self-referential memory.
  • The origin failure is an environment/prerequisite failure. The supporting evidence record's failure_class is environment; its summary frames the work as repairing environment prerequisites. The skill label is an artifact of the origin task's title ("audit skill modules and tests"), not of any observed skill fault. An origin task that failed at the execution-environment/prerequisite stage never exercised the skill code.

Specific evidence gap

The finding rests on a single, low-confidence, self-referential record: the sole evidence is a prior failure to repair a low-confidence skill finding. There is - no independent corroboration (support count is 1, below the >=2 threshold that would raise confidence), - no failing skill test and no reproducer, - no failure_class other than environment in the supporting record.

Closing action for the finding: it should be aged out or superseded as low-confidence/unsubstantiated once the executor/preflight conditions on the task line are resolved. Any genuine follow-up belongs to the executor/preflight layer (coding-agent sandbox preflight), not to the skill modules.

Verification (no code changed; checks confirm the surface is green)

Run in the task-owned worktree with the bootstrapped interpreter (python3 3.12.13; pytest/coverage present).

Skill-scoped contract tests:

scripts/run-contract-tests.sh \
  tests/test_skill_auto_repair.py \
  tests/test_fleet_skills.py \
  tests/test_hermes_skills_dedup.py
# => 46 passed

Docs identity guard (this closure doc reads generic for any fleet owner):

scripts/run-contract-tests.sh tests/test_docs_no_operator_identity.py
# => passed

Full repository contract suite (baseline and after adding this doc):

scripts/run-contract-tests.sh
# => passed, coverage floors met

Determination

  • Do not open a skill-behavior repair from this finding.
  • Treat the finding as low-confidence/unsubstantiated; close it out with the evidence gap recorded above.
  • Keep the skill code, skill docs, and skill tools unchanged.