Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add konamgil/mandu --skill mandu-mcp-verifygit clone --depth 1 https://github.com/konamgil/manduWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/konamgil/mandu/mandu-mcp-verify)<a href="https://agentmods.dev/skills/konamgil/mandu/mandu-mcp-verify"><img src="https://agentmods.dev/badge/skills/konamgil/mandu/mandu-mcp-verify/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/konamgil/mandu/mandu-mcp-verify"><img src="https://agentmods.dev/badge/skills/konamgil/mandu/mandu-mcp-verify.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00066 | $0.02068 |
| Opus 5.5 | $0.00026 | $0.00827 |
| Sonnet 5.5 | $0.00013 | $0.00414 |
| Haiku 4.5 | $0.00007 | $0.00207 |
Grade A, and why
mandu-mcp-verify scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 183 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Mandu MCP Verify
편집 직후 / 생성 직후 / pre-commit 에서 도는 표준 검증 루프. 핵심은 "하나의 ate 파이프라인 + guard + 구조 진단" 을 병렬 fast path 로 먼저 돌리고, 실패했을 때만 세분 도구로 drill-down 하는 것.
Trigger
- 파일 편집 직후 (특히
app/,spec/,src/shared/contracts/) mandu_generate완료 직후 (create-flow 가 자동 전이)- "check 해줘", "제대로 된 거야?", "괜찮아?"
- stop-hook / pre-commit 자동 실행
- 사용자가 명시적으로 "verify", "validate", "검증"
Fast Path (병렬, 기본값)
네 도구는 서로 독립. 병렬 호출.
┌─────────────────────────────┐
│ mandu.ate.auto_pipeline │ ← Tier-0: extract/generate/run/heal 전부
├─────────────────────────────┤
│ mandu_guard_check │ ← Tier-1: 아키텍처 규칙 위반 스캔
├─────────────────────────────┤
│ mandu.lint │ ← 가드레일-lint: 코드 품질 스캔 (oxlint)
├─────────────────────────────┤
│ mandu_doctor │ ← Tier-0: 구조/import/location 진단
└─────────────────────────────┘
네 개를 parallel 호출
인자 예:
mandu.ate.auto_pipeline({
repoRoot: ".",
oracleLevel: "standard",
useImpactAnalysis: true, // 변경된 라우트만 대상
autoHeal: false // heal 은 별도 단계
})
mandu_guard_check({ repoRoot: "." })
mandu.lint({ typeAware: false }) // oxlint 미설치 시 조용히 skip
mandu_doctor({ repoRoot: "." })
네 응답을 수집하고 pass/fail 매트릭스를 만든다:
| 도구 | pass | fail 의 의미 |
|---|---|---|
ate.auto_pipeline |
테스트 통과 + 계약 일치 | 테스트 실패 또는 impact 드리프트 |
guard_check |
레이어/금지 import 위반 없음 | 아키텍처 규칙 위반 |
lint |
oxlint error 0 | error 1+ (warning 은 informational) |
doctor |
파일 위치 / import 정상 | 구조 결함 |
네 개 모두 green → verify 종료. 사용자에게 압축 요약 제시.
Drill-Down (실패 시에만)
fast path 가 빨간 항목을 돌려주면 그 카테고리별로만 drill-down. 세 종류가 다 실패했다고 모든 drill-down 을 돌리지 말고, 응답에 들어온 카테고리만.
Guard 실패
mandu_guard_explain({ violationId }) ← 먼저: 왜 위반인지 설명
→ (자동 수정 가능하면) mandu_guard_heal({ autofix: true, violationIds: [...] })
→ (수동 수정이면) 사용자에게 수정 지시 후 fast path 재실행
ATE 실패 (테스트 실패 / 힐 후보 있음)
mandu.ate.feedback({ runId }) ← 힐 후보의 카테고리/우선순위
→ mandu.ate.apply_heal({ runId, healIndex, createBackup: true })
→ fast path 의 ate.auto_pipeline 재실행
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2mo ago First seen · 183 lines · 66 tokens per session scan A 3a77a2ef1e77
mandu-mcp-verify is a skill published in the GitHub repository konamgil/mandu (47 stars, last pushed 2mo ago), licensed MPL-2.0. It adds 66 tokens to every session and 2,068 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
confirm-failures-are-causally-linked-to-the-task-before-reportin
When delegating a task affected by this skill, include.
isolate-the-specific-failure-mode-before-proposing-a-fix
When delegating a task affected by this skill, include.
qa-sweep
Codex adapter for independent QA review of opencode-swarm changes. Use after implementation or debugging when the user asks for a QA sweep, regression check, review pass, quality gate, or confidence check before publishing.
bounded-code-repair
Repair one focused, inspectable code behavior with a declared file boundary, actual validation, and a bounded retry count. Do not use for broad feature delivery, architecture decisions, security review, or documentation-only work.
fix-bug
Investigates and fixes a reported bug. Reads relevant code, identifies the root cause, proposes a fix, and adds or updates tests to prevent regression. Invoked when the user describes unexpected behavior, an error, a crash, or a failing test.
investigate
A structured root-cause investigation method for software bugs. It moves from history and reproduction through tested hypotheses, a focused fix, verification, and a report.