From harnish
Inspection skill. Triggers: "점검해", "확인해", "검증해", "ralphi", "셀프점검", "커버리지 확인", "테스트 갭", "고쳐", "수정해", "점검하고 고쳐", "자동으로 처리해", "inspect", "check", "verify", "self-check", "coverage check", "test gap", "fix", "repair", "inspect and fix", "handle automatically"
How this skill is triggered — by the user, by Claude, or both
Slash command
/harnish:ralphiThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Utterance contains "고쳐", "수정해", "fix", "처리해" → **Autonomous** (proceeds to fix)
SKILL.ko.mdreferences/criteria-code.ko.mdreferences/criteria-code.mdreferences/criteria-prd.ko.mdreferences/criteria-prd.mdreferences/criteria-project.ko.mdreferences/criteria-project.mdreferences/criteria-script.ko.mdreferences/criteria-script.mdreferences/criteria-skill.ko.mdreferences/criteria-skill.mdUtterance contains "고쳐", "수정해", "fix", "처리해" → Autonomous (proceeds to fix) Otherwise → HITL (report only and wait)
everything / all are not valid scopes — pick a concrete target.)" No guessing. Do not run git diff on your own.docs/prd-*.md (check §section structure) → criteria-prd.md*/SKILL.md (check frontmatter name:) → criteria-skill.md*.sh (check shebang #!/) → criteria-script.md.py .ts .js .go etc. source code → criteria-code.md{path} — what type? (prd / skill / script / code / other?)" No guessing.references/. Do not read other criteria.pytest, npm test, go test ./..., cargo test, custom?)" No guessing. Tool not installed (command not found) → SKIP that check + warn. Do not attempt installation.Used by Step 3A.5 (per component) and Step 3B.6 (per diff hunk).
Ask the four questions:
Verdict: Q1 and Q4 are the only gates. Q2 and Q3 are scaffolding to reach honest answers.
Items that fail Q1 or Q4 → [warning] unjustified existence (kept only by convention / inheritance / authority).
Report in Step 5 format → wait for user judgment.
Clear instructions (execute immediately):
Ambiguous responses (must re-ask):
Fix immediately in critical→warning→coverage order → test after each fix → report results in Step 5 format when all done.
Fix prefers subtraction. Default action for [unjustified existence] is removal. For [critical/warning] violations, minimal additive patches are allowed (null check, missing import, type fix, error guard) when needed to satisfy the criteria. Do not introduce new abstractions, modules, or features. If a fix would require new abstraction/module/feature, mark as [unfixed] structural change required.
If there are unfixed items, hand off to user judgment.
Severity: critical (behavioral error) / warning (potential issue) / coverage (test gap)
HITL:
## ralphi Inspection Results
Target: {path | "user-specified scope"}
### Findings ({N} items)
1. [{severity}] {file:line} — {one-line cause}
Which issues should I fix first?
Autonomous:
## ralphi Inspection + Fix Results
Target: {path | "user-specified scope"}
### Fixed ({M}/{N} items)
1. [fixed] {file:line} — {one-line fix description}
### Unfixed ({K} items) ← only when present
1. [{severity}] {file:line} — {cause} → {reason unfixed}
Tests: {PASS | FAIL (details)}
No issues found → ralphi inspection complete, no issues found. Single line. No enumeration.
| When | What is read |
|---|---|
| Step 3A (file inspection) | Exactly 1 criteria file from references/, matched to detected type. No others. |
| Step 3B (project inspection) | git diff only. No full file reads. Read only the function relevant to a scenario walkthrough. |
| Necessity check | Operates on already-loaded content. No new file reads. |
| Step 5 (report) | No file reads. Output only. |
npx claudepluginhub jhu900202/plugins --plugin harnishValidates code changes against original plan using git diffs and full file reads, then launches parallel subagents for quality, security, and test coverage checks. Use post-implementation or /recheck.
Dispatches 5 specialized agents for multi-perspective code review on correctness, architecture, security, production readiness, and test quality. Merges findings, auto-fixes Critical/Important issues up to 3 rounds.
Reviews diffs, PRs, and agent output for bugs, security issues, mocks, and code quality. Automates codebase audits with domain-specific checks and deep scanning.