describe-pr

Pass

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on the GitHub CLI (gh) and git to manage pull requests, stage commits, and push changes to remote repositories as part of its core workflow. This is standard behavior for a PR management tool.- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external, attacker-controllable sources such as pull request diffs, PR comments, and ticket artifacts to generate descriptions.
  • Ingestion points: SKILL.md (Step 2) and references/pr_targeting_and_diffs.md use gh pr view and gh pr diff to pull external content into the agent's context.
  • Boundary markers: While the skill includes guardrails to prevent exfiltrating credentials or sensitive data, it lacks explicit structural delimiters to isolate external text from instructions.
  • Capability inventory: The skill can write to the local filesystem (.agent-work/tasks/), modify local files via script, and update remote GitHub pull request bodies using gh pr edit.
  • Sanitization: The provided scripts/inject-walkthrough-diffs.sh script sanitizes diff content by escaping </script> tags to prevent breaking the generated HTML structure.- [DYNAMIC_EXECUTION]: The skill generates interactive HTML walkthroughs by injecting processed diff data into a template using a local bash script (inject-walkthrough-diffs.sh). This dynamic artifact creation is a core feature but constitutes a runtime generation pattern.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 5, 2026, 07:16 AM