The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill is susceptible to instructions embedded in GitHub issues and PRs. * Ingestion points: Issue and PR data are fetched via GraphQL in scripts/triage/github.ts. * Boundary markers: None detected in the prompts. * Capability inventory: Commands git and gh are executed via execFileSync, and reports are written to the filesystem. * Sanitization: Basic phrase sanitization exists in derive.ts for configuration overrides.
Command Execution (SAFE): CLI tools are invoked using execFileSync with argument arrays, which is a secure pattern for preventing shell injection.
Data Exposure & Exfiltration (SAFE): The skill's access is restricted to repository maintenance tasks, with no evidence of sensitive data exfiltration or unauthorized network calls.

open-source-maintainer