draft-public-rules

Pass

Audited by Gen Agent Trust Hub on Apr 28, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: Utilizes git log and filesystem traversal to analyze the project history and structure for the purpose of generating contextually relevant documentation.
  • [DATA_EXFILTRATION]: Scans local project paths and global agent directories (e.g., ~/.claude/, ~/.cursor/) to identify installed tools and their usage hints. No external network requests or exfiltration of sensitive credentials were identified.
  • [PROMPT_INJECTION]: Ingests potentially untrusted content from the project environment (README files, commit messages) during the analysis phase.
  • Ingestion points: Project metadata files, README.md, and git commit history.
  • Boundary markers: The skill workflow terminates in a user-reviewable draft, preventing automated execution of injected instructions.
  • Capability inventory: Filesystem write operations limited to documentation (AGENTS.md, CLAUDE.md) and bridge settings.
  • Sanitization: Content is paraphrased and curated before being included in the generated output.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 28, 2026, 12:21 PM
Security Audit — agent-trust-hub — draft-public-rules