agentsclimarketplace

Cairo audit skill

Skill alebeta06/cairo-audit-skill

AI-assisted security auditing for Cairo smart contracts. An open-source Agent Skill for Claude Code.

Install
npx -y skills add alebeta06/cairo-audit-skill

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Assist with security audits of Cairo smart contracts on Starknet. Use when reviewing Cairo/Starknet contracts for vulnerabilities, running audits, threat modeling, or writing snforge security tests. Triggers on "audit", "Cairo", "Starknet", "vulnerability", "security review".

SKILL.md

3.6 KB, as published. Nobody here has run it

<!-- 🇪🇸 NOTA: El bloque de arriba entre "---" es el FRONTMATTER (metadatos en formato YAML). DEBE ser lo PRIMERO del archivo: ningún texto ni comentario puede ir antes del primer "---", o Claude Code no detectará el skill. Solo 'name' y 'description' son obligatorios. --> <!-- 🇪🇸 NOTA: 'description' es lo ÚNICO que Claude lee SIEMPRE (Nivel 1 de "progressive disclosure"). Por eso lleva las palabras-gatillo (audit, Cairo, Starknet, vulnerability...) al frente: con ellas Claude decide si activar el skill. El cuerpo de abajo solo se carga cuando el skill se activa (Nivel 2). --> <!-- 🇪🇸 NOTA: 'allowed-tools' limita qué herramientas puede usar el skill. Aquí solo lo necesario para auditar: Read (leer código), Grep/Glob (buscar patrones) y Bash (correr scarb/snforge/ caracal). Más adelante podríamos añadir Agent/Task para escaneo en paralelo. -->

Cairo Audit Skill

Status: 🚧 Skeleton — the workflow and reference files below are placeholders to be filled in future sessions.

What This Skill Does

Guides AI-assisted security audits of Cairo smart contracts on Starknet. It structures the review process, surfaces known vulnerability patterns, and helps write snforge security tests. It does not replace a human auditor — it accelerates and organizes their work.

When to Use

  • Reviewing a Cairo/Starknet contract before merge or deployment.
  • Threat modeling a Starknet protocol (accounts, upgrades, cross-domain messaging).
  • Writing or reviewing snforge security / fuzz tests.

When NOT to Use

  • Writing new feature code (this skill is read-and-review oriented).
  • Deployment-only operations.
  • Non-Cairo ecosystems (use a Solidity- or Rust-specific skill instead).
<!-- 🇪🇸 NOTA: "Reference Files" es el corazón del progressive disclosure: el SKILL.md se queda corto y APUNTA a archivos externos que Claude carga SOLO cuando los necesita (Nivel 3). Hoy están vacíos (placeholders); en próximas sesiones llenaremos patterns/, examples/ y docs/. -->

Reference Files

ContextPath
Vulnerability patterns (by category)patterns/ (TODO)
Vulnerable vs. secure contract examplesexamples/ (TODO)
snforge security test templatestests-templates/ (TODO)
Extended documentationdocs/ (TODO)

Audit Workflow

<!-- 🇪🇸 NOTA: Estos pasos son un PLACEHOLDER. La metodología real (qué buscar en cada paso y qué comandos correr) se documentará en próximas sesiones. -->
  1. Initial read — understand the architecture before hunting for bugs. (TODO)
  2. Threat model — who attacks, what they want, their capabilities. (TODO)
  3. Static analysis — run Caracal. (TODO)
  4. Manual line-by-line review — where the real bugs live. (TODO)
  5. Property testing & fuzzingsnforge. (TODO)
  6. Report — severity + recommendations. (TODO)

Principles

  • Assume every external call is hostile.
  • Assume any balance, price, or cross-domain message can be manipulated in the current block.
  • Hunt for implicit invariants the developer assumed but never validated.
  • Never declare code "secure" — declare "no issues found under this threat model".

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.