The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted text from pull request reviews, conversation comments, and review threads, which creates an indirect prompt injection surface.
Ingestion points: Review bodies and comments are fetched via gh api as described in SKILL.md (Step 2) and the hunt recipes in REFERENCE.md.
Boundary markers: The skill does not implement explicit delimiters or instructions to ignore embedded commands within the ingested text.
Capability inventory: The agent has the ability to modify the codebase through git commit and git push (Step 5) and interact with PR participants by posting replies (Step 7).
Sanitization: The skill relies on triage verdicts and "unslop" formatting rules for replies, but lacks robust sanitization to prevent adversarial instructions in the input feedback from influencing agent behavior.
[COMMAND_EXECUTION]: The skill performs shell operations using git (fetch, checkout, commit, push) and gh (GitHub CLI) to manage the repository and pull request state. It also invokes JavaScript runtimes like bun, node, and tsx to execute auxiliary scripts.
[EXTERNAL_DOWNLOADS]: The skill depends on external script files located in sibling directories, such as ../gh/pr-threads.ts and ../gh/ci-failures.ts, which are treated as internal vendor dependencies.

fix-pr