pr-comment-responder

Pass

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted PR comments from GitHub and passes them to an orchestrator agent with code-modification capabilities.
  • Ingestion points: The skill fetches external comments using get_pr_review_comments.py in Phase 1 and references/workflow.md.
  • Boundary markers: PR comments are interpolated into orchestrator prompts without explicit delimiters or warnings to ignore embedded instructions.
  • Capability inventory: The skill can post replies via post_pr_comment_reply.py, resolve threads, and delegate code implementations and commits to a subagent.
  • Sanitization: No sanitization or escaping of PR comment content is performed before processing.
  • [COMMAND_EXECUTION]: The workflow relies on shell commands (e.g., sed, jq, echo) and Python scripts that incorporate variables derived from user prompts and PR data. For example, extract_github_context.py takes the raw user prompt as an argument via the --text flag, and sed -i is used to update local artifacts based on comment metadata, which could be problematic if the input contains unexpected shell metacharacters or delimiters.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 7, 2026, 02:42 AM