---
name: repo-harness-check
description: Plan scoped harness work, review a plan, or assess recorded checks and risk. Do not execute implementation or publish by default.
when_to_use: "repo-harness-check"
---

# repo-harness-check

Bot entrypoint for planning and verification decisions.
Confirm the repo, scope, subject revision and worker ownership.

## Mode Selection

- Create a plan only when a design decision needs it: `references/create.md`.
- Review an existing plan: `references/review.md`.
- Assess verification: consume commands, results, revision and environment.
  Ask the worker for missing checks from the target repo's agent instructions.
  Select independent review by the diff's Review by risk tier in the
  development flow; a low tier gets none.
- Deployment readiness: `references/deploy-readiness.md`.
- PR classification, and pre-merge verification of an external PR:
  `references/pr-verification.md`.
- Review-comment triage for PR follow-up: `references/review-comment-triage.md`.
- Architecture or refactor evidence for a scheduling decision:
  `references/architecture-evidence.md`. repo-harness supplies the evidence;
  the Bot decides what to schedule.

## Boundaries

Ordinary tasks do not require an Approved plan or a contract file.
Reuse passing evidence only for its recorded subject. Run delta checks after edits.
Report failed, stale, timed-out or omitted coverage. Missing optional tooling is
a diagnostic. Use independent review for large changes, security/permissions,
or unresolved uncertainty. Review does not authorize implementation or publication.

For skill-effectiveness claims, require full_test_count > 0, dry_run_ratio <= 30%,
and graders reported. Mark non-authoritative: dry-run-heavy or all-dry-run evidence.
Mark unavailable: no current eval evidence.
Does not claim skill-effectiveness authority from dry-run benchmark output.