The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the GitHub CLI using Python's subprocess.run with argument lists. This approach is secure against shell injection. It allows specifying the path to the gh binary via a command-line argument, which is standard for such tools.\n- [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted pull request comments from GitHub. This risk is managed through a documented evidence chain: (1) Ingestion points: Review comments are fetched via scripts/build_thread_plan.py; (2) Boundary markers: External content is isolated in a structured thread-plan.json file; (3) Capability inventory: Modifications include local file edits and GitHub mutations via scripts/apply_review_threads.py; (4) Sanitization: The workflow requires user approval of the plan before execution.\n- [SAFE]: Analysis found no evidence of credential exfiltration, obfuscation, or malicious persistence. All operations are traceable and rely on the existing GitHub authentication of the environment.

krt-review-herald