implement-specs

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from GitHub issues and pull requests, which can be authored by untrusted external parties.
  • Ingestion points: The scripts/fetch_github_context.py script fetches issue bodies, PR descriptions, and comments. The SKILL.md instructions also reference local context files such as issue_context.json, issue_comments.txt, and pr_diff.txt.
  • Boundary markers: The script implements a provenance system by wrapping fetched content in headers that include metadata such as the source, author, and author_association. It explicitly marks content from trusted roles (OWNER, MEMBER, COLLABORATOR) with a trust=TRUSTED label.
  • Capability inventory: The skill enables the agent to write implementation code and update repository specifications. The helper script uses subprocess.run to interact with the GitHub CLI (gh).
  • Sanitization: No content-level sanitization (e.g., stripping markdown or HTML comments) is performed on the fetched data. The skill relies on the agent following instructions to treat this content as data rather than instructions.
  • [COMMAND_EXECUTION]: The Python script scripts/fetch_github_context.py executes shell commands using the subprocess module to interact with the GitHub CLI.
  • Evidence: It uses subprocess.run(["gh", *args], ...) and subprocess.run(["gh", "pr", "diff", ...], ...) to fetch data.
  • Context: These calls use argument lists rather than shell strings, which is a safe practice that prevents command injection from the repository or number parameters.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 01:12 AM