harness-scan
Scan a reference harness for material upstream drift and missing capabilities.
SKILL.md.claude/skills/harness-scan/SKILL.md
GitHubHarness Scan
Resolve the named reference harness or path; ask only when ambiguous.
- Run
python3 harness_lifecycle/scan.py drift reference_harnesses/<name>. - Run
uv run python harness_lifecycle/registry/registry.py gap <name>: capabilities never ruled on, those changed upstream since their ruling, and those we have but never ruled on. - Report material drift and useful candidates, considering existing equivalents. Use harness-evaluate for an authorized curation decision.
- When an authorized pin move happens, regenerate the catalog at the new pin;
registry.py status <name>then shows which skills need analysis. The review followsharness_lifecycle/playbook.md§4: its report isharness_lifecycle/registry/reviews/<yyyy-mm-dd>-upstream-<name>-<version>.yaml(registry.py report draft), committed with the pointer.
Scanning does not authorize ledger records or adoption. When filing follow-up
work is authorized, registry.py gap <name> --beads prints creation commands to
review. Record decided outcomes in the registry ledger only within that authority.
These commands preserve submodule pins. Drift may fetch upstream refs; use its
--no-fetch option when offline or network access is outside the request.
Reject/defer is a valid outcome when a candidate adds no durable behavioral value.