The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes GitHub PR comments, which are untrusted external data sources that could contain malicious instructions.
Ingestion point: Comment bodies are fetched in scripts/gh-pr-comments.ts via the GitHub GraphQL API (lines 120-176).
Capability inventory: The agent is authorized to modify source code, commit changes, and push to repositories based on the contents of these comments.
Boundary markers: The instructions include explicit defense-in-depth warnings, such as "Verify each claim before acting on it" and "Never fix on the reviewer's say-so alone," which help mitigate but do not eliminate the injection surface.
Sanitization: No explicit technical sanitization or escaping of comment text is performed before the content is presented to the agent.
[COMMAND_EXECUTION]: The skill executes shell commands using the GitHub CLI to manage authentication and repository context.
Evidence: The script scripts/gh-pr-comments.ts executes gh auth token, gh repo view, and gh pr view using Bun's shell template.
Note: The script utilizes Bun's $ template tag for shell execution, which provides automatic argument escaping and reduces the risk of direct command injection.

tackle-gh-comments