The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources that could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: SKILL.md (Step 2) and REFERENCE.md (GitHub hunt recipes) ingest pull request review threads, conversation comments, and CI failure logs using the GitHub API.
Boundary markers: The skill does not explicitly define delimiters for external content, though it mandates a normalization step to split and deduplicate findings.
Capability inventory: The skill possesses significant capabilities including modifying project source code (Step 4), committing changes, and pushing to remote repositories (Step 5).
Sanitization: While the skill includes a triage step (Step 3) where the agent must verify and reproduce findings before acting, it lacks automated sanitization or escaping of the ingested GitHub comment content.
[EXTERNAL_DOWNLOADS]: references/unslop-reply-drafts.md references a provenance URL pointing to the official Cursor plugins repository on GitHub for its reply-cleaning logic. This targets a well-known service and repository associated with the Cursor AI editor.

fix-pr