# Repo Verify

> Run a repository verification loop with runner-first execution, deterministic stages, and explicit PASS/FAIL/SKIP output. Use for fast or full validation of a consumer repository. Not for: mutation, fixing failures, dependency remediation, or merge approval. Output: runner summary or a complete fallback stage table.

- Skill: `hmj1026/repo-verify` (Agent Skill, multi-file: 6 files)
- Install (CLI): `npx get-skill add hmj1026/repo-verify`
- Raw SKILL.md: https://api.skillmd.com/api/skills/hmj1026/repo-verify/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: hmj1026 (https://skillmd.com/u/hmj1026)
- Updated: 2026-11-16
- Page: https://skillmd.com/skills/hmj1026/repo-verify

---


# Repository Verify

## Workflow

Run read-only in the consumer repository. Normalize no mode or `full` to full
stages and `fast` to lint + unit; accept only `fast` or `full`. Attach
`--integration <path>` and `--e2e <path>` only to their matching stages.

Use a project-installed
`.claude/dhpk/skills/repo-verify/scripts/verify-runner.js` when present. If it
is absent, use the package-local `$SKILL_DIR/scripts/verify-runner.js` for a
Node project. `$SKILL_DIR` is the installed directory containing this
`SKILL.md`; its adjacent `scripts/lib/runner-utils.js` is part of the same
local closure, and the consumer cwd remains unchanged. A runner's crash,
missing overall verdict, execution error, or explicit semantic `FAIL` is
terminal and does not fall through. A Markdown `FAIL` or `summary.json`
`overallPass=false` remains failure even when the runner exits zero. When no
applicable runner is available, use the non-Node/manifest fallback and stage rules in
[`references/workflow.md`](references/workflow.md); unsupported manifests are
terminal failures.

For evidence applicability and reuse, follow the kernel's
[Evidence snapshots](https://github.com/hmj1026/dhpk/blob/main/rules/execution-policy-kernel.md#evidence-snapshots)
rule.

## When NOT to Use

- A check should edit, format, build, or fix the repository.
- Dependency security review is the goal (use `$change-verdict deps`).
- Merge approval or code-review verdict is the goal.

## Output

For a runner, prefix output with `Source: runner` and preserve its summary.
For fallback, report `Source: fallback`, every lint/typecheck/unit/
integration/e2e stage as `PASS`, `FAIL`, or `SKIP` with command/reason, and an
agreeing overall verdict. Invalid modes, unsupported manifests, runner
failures, and execution errors are terminal `FAIL`/`BLOCKED` results.

## Verification

- [ ] Mode and optional paths were normalized before execution.
- [ ] Runner precedence and terminal failure boundary were respected.
- [ ] Every applicable stage has a result and exact command or skip reason.
- [ ] Semantic overall verdict agrees with the stage results and source is
  named, regardless of CLI exit status.

## References

- [`references/workflow.md`](references/workflow.md) — runner precedence,
  fallback ecosystems, Node script rules, and report template.

