write-feature-docs

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests content from external spec files (PRODUCT.md, TECH.md) which may contain instructions designed to influence agent behavior.
  • Ingestion points: Reads local spec files at specs/<id>/ as specified in Step 1.
  • Boundary markers: Includes an interactive outline confirmation step (Step 3) where the engineer must approve the plan before drafting.
  • Capability inventory: Shell execution (git, gh), PR creation, and optional GUI interaction (computer_use).
  • Sanitization: Content derived from TECH.md is flagged for review or replaced with placeholders in ambient mode to prevent accidental exposure of internal implementation details.
  • [COMMAND_EXECUTION]: The skill executes multiple shell commands to interact with GitHub and the local file system.
  • Evidence: Uses gh search code, gh pr list, gh api, and git log to discover information and automate workflows.
  • Mitigation: Implements regex-based input validation (^[A-Za-z0-9][A-Za-z0-9_-]*$) for user-provided spec IDs and feature tokens to mitigate shell injection risks.
  • Mitigation: Mandates the use of gh pr create --body-file to prevent shell interpretation and corruption of PR description content.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 11:35 PM
Security Audit — agent-trust-hub — write-feature-docs