pr-human-guide

Pass

Audited by Gen Agent Trust Hub on May 9, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: PR numbers provided as arguments are strictly validated against a numeric regex (^[1-9][0-9]{0,5}$) before being used in shell commands, preventing command injection.
  • [PROMPT_INJECTION]: The skill is designed to process untrusted pull request data (titles, bodies, and diffs), which presents an indirect prompt injection surface.
  • Ingestion points: PR metadata and diff content are retrieved via GitHub CLI commands (gh pr view, gh pr diff).
  • Boundary markers: All PR-derived content is wrapped in <untrusted_pr_content> tags with explicit instructions for the AI to ignore any embedded instructions.
  • Capability inventory: The skill possesses the capability to modify the pull request's description using the gh pr edit command.
  • Sanitization: In addition to argument validation, the skill uses temporary files and the --body-file flag to update the PR description, ensuring that markers are not corrupted by shell processing.
  • [SAFE]: The helper script marker-helper.py uses character codes for markers to avoid accidental interaction with shell history expansion and implements logic to strip stray markers, preventing replacement bound confusion.
Audit Metadata
Risk Level
SAFE
Analyzed
May 9, 2026, 09:33 PM