ag-auditar-mcp

ag-auditar-mcp is a skill for Claude Code from andregusman-raiz/a-gusman-claude. It costs 53 tokens per session (2,054 once invoked), scanned A, original, MIT.

An offline security review for third-party MCP servers and skills. MCP is a way for an agent to call external tools; the review looks for hidden instructions in tool or skill descriptions, unsafe dependencies, exposed secrets, and related supply-chain risks.

In plain words
What is it for?
Auditing a cloned MCP or skill repository before installation, reviewing new MCP servers and third-party skills, and investigating unexpected agent actions.
Why use it?
It helps detect tool poisoning, where text presented as ordinary tool information tries to influence the agent's actions. It can also help catch later changes to an already approved server or skill.

Skill for Claude Code

Written for Claude Code: allowed-tools in frontmatter. Also seen: reads .claude/ paths.

Good fit Auditing a cloned MCP or skill repository before installation, reviewing new MCP servers and third-party skills, and investigating unexpected agent actions.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/andregusman-raiz/a-gusman-claude/ag-auditar-mcp
Install

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.

Any agent
npx skills add andregusman-raiz/a-gusman-claude --skill ag-auditar-mcp
Clone the repo
git clone --depth 1 https://github.com/andregusman-raiz/a-gusman-claude

Made for: Claude Code.

Wrote 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.

agentmods badge for ag-auditar-mcp

README.md
[![agentmods](https://agentmods.dev/badge/skills/andregusman-raiz/a-gusman-claude/ag-auditar-mcp/github.svg)](https://agentmods.dev/skills/andregusman-raiz/a-gusman-claude/ag-auditar-mcp)
Your own site
<a href="https://agentmods.dev/skills/andregusman-raiz/a-gusman-claude/ag-auditar-mcp"><img src="https://agentmods.dev/badge/skills/andregusman-raiz/a-gusman-claude/ag-auditar-mcp/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.

agentmods 80×15 button for ag-auditar-mcp

Your own site · 80×15
<a href="https://agentmods.dev/skills/andregusman-raiz/a-gusman-claude/ag-auditar-mcp"><img src="https://agentmods.dev/badge/skills/andregusman-raiz/a-gusman-claude/ag-auditar-mcp.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 53 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,054 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. ✓ AI security review Fable 5.1 · 7 Sept 2026 📄 Read the review
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00053 $0.02054
Opus 5 $0.00026 $0.01027
Sonnet 5 $0.00011 $0.00411
Haiku 4.5 $0.00005 $0.00205

Measured 12d ago against content hash aebced019786, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-12, from the pricing page.

Security

Grade A, and why

ag-auditar-mcp 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 12d 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.

skills/ag-auditar-mcp/SKILL.md · 135 lines

How it starts

The opening of the file, as written. The whole thing — 135 lines — stays where its author put it; the contents beside it link to each section on GitHub.

ag-auditar-mcp — auditoria da superficie MCP e de skills de terceiros

Por que existe (delimitacao vs ag-auditar-harness)

ag-auditar-harness / HCS.injection cobre codigo do harness: eval $(...) em hook, bash -c "$var", curl | sh. Este skill cobre um vetor diferente e nao coberto: texto que o modelo le como dado e obedece como instrucao.

Uma tool MCP anuncia um name e uma description em linguagem natural. O modelo le a description antes de decidir chamar a tool. Instrucao escondida ali e obedecida sem que nada apareca para o usuario — e injecao indireta entregue pela cadeia de suprimentos (OWASP MCP03:2025, MITRE ATLAS AML.T0010 / AML.T0051.001). O mesmo vale para o description: de qualquer SKILL.md de terceiro instalado no harness.

Quando invocar

  • Antes de adicionar um MCP server novo (proprio ou de terceiro)
  • Antes de instalar skills/plugins de terceiro — o --scan roda no repo clonado, antes do install
  • Periodicamente, para pegar rug pull: um servidor ja aprovado cuja config/descricao muda depois
  • Quando um agente tomou acao inesperada compativel com tool poisoning

Ferramenta local (offline, stdlib, nao executa MCP server)

S=~/Claude/.claude/scripts/mcp-baseline-audit.py

python3 $S                       # inventario + diff contra baseline
python3 $S --save-baseline       # (re)arma deteccao de rug pull
python3 $S --scan-skills <dir>   # varre SKILL.md de terceiros antes de instalar
python3 $S --json                # machine-readable
python3 $S --strict              # exit 2 se houver P0 (uso em gate/CI)

Baseline: ~/.claude/state/mcp-baseline.json (hash de command/args/url/chaves de env e headers por servidor). Mudanca silenciosa nesse conjunto = P0 config mudou.

O que ele detecta:

Sinal Sev Por que importa
Config de servidor mudou vs baseline P0 rug pull: aprovado num estado, rodando em outro
Instrucao ao assistente em descricao ("ignore previous", "nao conte ao usuario") P0 tool/skill poisoning
Caractere invisivel (ZWSP/ZWJ/BOM) ou bidi override P0 payload escondido do revisor humano
Path de segredo (~/.ssh, id_rsa, .aws/credentials) em descricao P0 alvo classico de exfiltracao
curl | sh no start do servidor P0 download+exec a cada boot
Transporte http:// remoto sem TLS P0 tool call em claro
npx -y pkg@latest sem pin de versao P1 codigo remoto novo a cada start, sem revisao
Segredo literal em env/headers e config legivel por outros P1 qualquer processo da maquina le — chmod 600
Segredo literal em config ja 0600 P2 aceito: mover para env var nao melhora (ver abaixo)
Servidor novo desde o baseline P1 entrou sem passar por revisao
Homoglifo CYRILLIC+LATIN no nome P1 typosquat de tool confiavel

Read the full file on GitHub · 135 lines

Changes

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.

  1. 12d ago First seen · 135 lines · 53 tokens per session scan F aebced019786

Subscribe to this mod's changes

ag-auditar-mcp is a skill published in the GitHub repository andregusman-raiz/a-gusman-claude (19 stars, last pushed 3d ago), licensed MIT. It adds 53 tokens to every session and 2,054 once invoked, about $0.0003 per session on Opus 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.

Related

Other skills, from other repositories

taiyi-ui-design

A design-planning guide for describing how an application's user interface should look and behave. It produces a UI-DESIGN.md document covering layouts, components, interactions, accessibility, and error states.

Dong90/oh-my-taiyiforge · 35 tokens

taiyi-evolve

A workflow skill that compares the implemented code with the frozen design after development and testing. It records architecture changes and proposes updates to DESIGN.md, the document describing the intended system structure.

Dong90/oh-my-taiyiforge · 37 tokens

taiyi-diagram-c4

A code-scanning tool that builds C4 architecture documents from a repository. It separates facts observed in the code from conclusions inferred about the design and uses Mermaid diagrams as the source format.

Dong90/oh-my-taiyiforge · 48 tokens

retrospective-audit

Stage B of /prflow:retrospective-weekly: given a most-recent-first subset of one recurring pattern's occurrence-PR context bundles (bounded by auditbundlecap), re-derive the root cause and return one JSON object carrying a ranked findings array (one to three sub-patterns) — no edits, no worktree. Invoked as a subagent…

The01Geek/prflow · 91 tokens

docs

Use when documentation generally needs to catch up with a branch before pushing or merging, covering internal developer docs, external customer-facing docs, and release notes together — "update the docs", "do a docs pass before I merge", "make sure everything's documented". Prefer this when no single documentation…

The01Geek/prflow · 97 tokens

tmux-lane-orchestrator

Manage one tmux agent lane from its matching ops pane, inspect live pane state and Codex logs on cold start, and produce concise manager summaries for OpenClaw and adjacent project work.

vincentkoc/dotskills · 46 tokens