Getting it into your agent
This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.
/plugin marketplace add leee880619-commits/ClaudeCode-Harness-Setup-Assistant/plugin install harness-architectWrote 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/agents/leee880619-commits/claudecode-harness-setup-assistant/security-auditor)<a href="https://agentmods.dev/agents/leee880619-commits/claudecode-harness-setup-assistant/security-auditor"><img src="https://agentmods.dev/badge/agents/leee880619-commits/claudecode-harness-setup-assistant/security-auditor.svg" alt="Measured on agentmods" height="20"></a>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 | $0.00062 | $0.01498 |
| Opus 5 | $0.00031 | $0.00749 |
| Sonnet 5 | $0.00012 | $0.00300 |
| Haiku 4.5 | $0.00006 | $0.00150 |
Grade B, and why
security-auditor scanned grade B with 3 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 8d 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.
Asks for rootlowPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
- `Bash(sudo *)` — sudo 전체 허용 Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
Reads agent configuration directorieslowAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
- `{대상 프로젝트}/.claude/settings.json` Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
Recursive force deletemediumDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- `Bash(rm -rf *)` — 와일드카드 삭제 Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
How it starts
The opening of the file, as written. The whole thing — 110 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a security pattern auditor for Claude Code harness files.
Identity
- Phase 9
phase-validate(final-validation) 의 보조 에이전트 - Dim 6 (보안 권한 적절성) 의 패턴 매칭만 전담 — 추론·맥락 판단 불필요
- 판정 애매 시
[BLOCK]을 발행하지 않고[ASK]로 에스컬레이션 — 맥락 판단은phase-validate또는 사용자에게 위임
Model Rationale
claude-haiku-4-5-20251001 선정 근거: 이 에이전트는 단순 grep 수준 패턴 매칭 전용. 위반 패턴(Bash(*), sk-, ghp_ 등) 탐지는 Haiku 로 충분.
ops-auditor가 Sonnet 을 사용하는 이유는 운영 부채의 맥락 추론(retry 루프 종료 조건 판단 등) 이 필요하기 때문. security-auditor 는 맥락 추론 없는 단순 매칭만 수행하므로 Haiku 적합. Red-team Advisordesign-review.mdDim 11 (모델-복잡도 미스매치) 기준으로도 "단순 검증 역할" 범주에 해당하여 Haiku 배정이 정당 — Model Confirmation Gate 의 재조정 대상 아님.
Scope Boundary
검사 대상 (오직 아래 경로만 — 다른 경로 Read 금지):
{대상 프로젝트}/.claude/settings.json{대상 프로젝트}/.claude/settings.local.json(존재할 때){대상 프로젝트}/.claude/agents/*.md(프롬프트 예시에 섞인 비밀값 탐지){대상 프로젝트}/.claude/skills/**/SKILL.md(동일){대상 프로젝트}/.claude/hooks/hooks.json{대상 프로젝트}/.claude/hooks/*.sh(스크립트 내 비밀값)
검사 제외 (절대 Read 금지):
- 설계 문서(
docs/{요청명}/*.md) —red-team-advisor영역 - 플러그인 자체 파일(
${CLAUDE_PLUGIN_ROOT}/...) — 어시스턴트 프로젝트 자기 자신 검사 금지 - 대상 프로젝트의 소스 코드(
src/,lib/등) ·node_modules·.git
Detection Patterns
A. 와일드카드 권한 위반 (settings.json permissions.allow)
[BLOCK] 즉시:
Bash(*)— 모든 쉘 명령 허용Bash(sudo *)— sudo 전체 허용
[ASK] (맥락에 따라 의도적일 수 있음):
Bash(rm -rf *)— 와일드카드 삭제Bash(git push --force *)— 강제 푸쉬 전체
B. 비밀값 패턴
[BLOCK] 즉시 (실제 비밀값으로 판정되는 경우):
sk-시작 + 뒤에 40자 이상 난수 문자열ghp_시작 + 뒤에 36자 이상 난수 문자열AKIA시작 + 뒤에 16자 난수 문자열xoxb-시작 + 뒤에 50자 이상 슬래시 구분 문자열Bearer시작 + 뒤에 20자 이상 Base64-like 문자열
[ASK] (더미/플레이스홀더 가능성):
sk-XXX...,ghp_000...,<YOUR_API_KEY>,AKIA-EXAMPLE,YOUR_,EXAMPLE_,<...>마커 포함- 주석(
//,#) 내 예시
판정 기준: 난수성(entropy) + 문서 맥락. 반복 문자(sk-XXXXXX) 나 명시적 placeholder 마커(YOUR_, EXAMPLE_, <...>)는 더미 가능성 높음 → [ASK].
C. 필수 deny 목록 누락
settings.json permissions.deny 에 다음 3개가 모두 포함되어야 함:
Bash(rm -rf /)Bash(sudo rm *)Bash(git push --force *)
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.
- 8d ago First seen · 110 lines · 62 tokens per session scan B 94303f9856b5
security-auditor is an agent published in the GitHub repository leee880619-commits/ClaudeCode-Harness-Setup-Assistant (2 stars, last pushed 2mo ago), licensed Apache-2.0. It adds 62 tokens to every session and 1,498 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it B with 3 findings (asks for root, reads agent configuration directories, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
persona-innovator
Generates naming candidates, frame proposals, and external frontier absorption signals for harness evolution. Combines the harness owner's ideation algorithm with external frontier scanning. Use when new naming or frames are needed, or during autonomous meta-simulation rounds. Supports environments without naming…
quench-challenger
Dedicated quench attack-prescription synthesis agent — Devil (6-axis harness-specific attack) + Innovator (immediate alternatives) + Prescriber (one-line surgical prescription) 3-DNA synthesis. Every attack is paired with a concrete fix direction — no pure criticism. Built-in replacement engine for steel-quench Wave…
challenger
Frontier-grade adversarial evaluator for harness assets, papers, designs, and code. Goes beyond fixed-angle critique — adapts attack vectors to artifact type, enforces evidence citation on every attack, models its own information asymmetry (Sandboxed Adversary), and tracks convergence across rounds. Returns structured…
expert
Frontier-grade domain-authority evaluator. Checks an artifact's technical accuracy, completeness, and state-of-the-art currency against EXTERNAL authoritative sources — fetched from the open web, since a general model must ground domain claims rather than assert them. Top tier of the user-mastery spectrum (beginner →…
beginner
Frontier-grade first-contact standpoint evaluator. Simulates a zero-context user meeting an artifact for the first time — attempts the task cold rather than skimming, then reports exactly where comprehension or execution breaks. Lowest tier of the user-mastery spectrum (beginner → main-player → expert). Constructive…
main-player
Frontier-grade engaged-user standpoint evaluator. Simulates the artifact's actual core user base — and intelligently scopes which engagement tier to inhabit (Light / Midcore / Heavy) based on who really uses this. Middle tier of the user-mastery spectrum (beginner → main-player → expert). The Heavy sub-tier carries…