aegis-audit
oleh sanguineseal
DevOps & CloudRingkasan
Deep behavioral security audit for AI agent skills and MCP tools.
Deskripsi
Apa yang Dilakukan Skill Ini
Aegis runs deterministic static analysis (AST parsing, Semgrep, and 15 specialized scanners) on AI agent skills and MCP tools to produce CWE-mapped, OWASP-tagged security reports with 0-100 risk scores. It generates Ed25519-signed lockfiles for tamper detection and supports optional LLM-powered intent analysis via Gemini, OpenAI, Claude, or local models. Runs fully offline by default.
Deterministic AST and Semgrep analysis with CWE/OWASP mapping produces reproducible, cryptographically signed reports that manual review or basic safety summaries cannot match.
Kapan Menggunakannya
- Scanning a new MCP server before adding it to Cursor
- Generating a signed lockfile before publishing a skill
- Running security gates in CI pipelines for skill deployments
- Auditing a third-party plugin for secret leakage or shell injection
- Verifying a skill's code hasn't changed since its last audit
Contoh Alur Kerja
User asks: Is this MCP skill safe to install?
- 1
Runs
aegis scan ./skill-directory --no-llmon the target skill - 2
Parses AST and applies 15 specialized scanners including secret detection and shell analysis
- 3
Maps findings to CWE and OWASP categories with severity tiers
- 4
Assigns a 0-100 composite risk score and a persona classification
- 5
Outputs a full security report to the console or as JSON for CI
Security report listing scoped capabilities, flagged patterns, risk score, and a persona (e.g. 'Permission Goblin' or 'LGTM')
Persyaratan
- GEMINI_API_KEY environment variable (optional, for Google Gemini LLM analysis)
- OPENAI_API_KEY environment variable (optional, for OpenAI LLM analysis)
- ANTHROPIC_API_KEY environment variable (optional, for Anthropic Claude LLM analysis)
Instalasi
clawhub install sanguineseal/aegis-auditJalankan perintah CLI-nya, atau tempel prompt ke Claude Code, Cursor, Codex, atau coding agent apa pun yang Anda pakai.