The Agent Skills Directory

[COMMAND_EXECUTION]: The scanner orchestrates multiple external security binaries (npm, pip-audit, semgrep, bandit, jq, curl) using the Node.js spawn API. These calls use array-based argument passing to prevent shell injection and are used strictly for vulnerability analysis of target directories.
[EXTERNAL_DOWNLOADS]: The skill fetches vulnerability enrichment data from well-known security endpoints including api.osv.dev and services.nvd.nist.gov. It also contains installation instructions that fetch release assets from the vendor's GitHub repository (github.com/prompt-security/clawsec). Documentation neutrally describes fetching these configurations from trusted infrastructure.
[CREDENTIALS_UNSAFE]: The skill uses environment variables (CLAWSEC_NVD_API_KEY, GITHUB_TOKEN) to manage API authentication for security databases. It does not contain hardcoded secrets, adhering to standard secret management practices.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The handler.ts hook ingests agent event data (HookEvent) to determine when to trigger scans.
Boundary markers: The results are reported back to the agent conversation using a structured system message template.
Capability inventory: The skill can execute local binaries and make network requests to security databases via the runner.sh script.
Sanitization: Input paths are resolved and validated; subprocess calls avoid shell interpolation by using the spawn array format.

clawsec-scanner