The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted PR comments from external sources (humans and bots) and uses them to guide code-fixing subagents. This presents an indirect prompt injection surface.
Ingestion points: SKILL.md and references/github-api.md describe fetching comments via the GitHub API.
Boundary markers: SKILL.md uses a template to isolate findings for subagents, though content is derived from external strings.
Capability inventory: SKILL.md executes subagents with file-writing capabilities and uses the gh CLI for git operations and thread resolution.
Sanitization: No specific sanitization or filtering of comment text is noted prior to subagent interpolation.
Mitigation: A mandatory human approval gate is required in Phase 3 before any fixes are executed.
[COMMAND_EXECUTION]: Executes gh CLI commands to interact with GitHub PRs and git to commit and push changes.
[EXTERNAL_DOWNLOADS]: Fetches data from GitHub, a well-known service, to retrieve pull request comments and review threads.

pr-comments