The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data including ../pr-review/references/review-intent.md, AGENTS.md, and repository code diffs via the coderabbit CLI. This creates a surface for indirect prompt injection if the files contain malicious instructions. The skill includes instructions to treat this content as untrusted evidence rather than authority. * Ingestion points: ../pr-review/references/review-intent.md, AGENTS.md, repository code, and CLI output. * Boundary markers: Explicit instructions to "separate defect validity from proposed-fix suitability" and treat output as "evidence, not authority". * Capability inventory: CLI execution (coderabbit), file reading, and authorized code fixing. * Sanitization: Warning against executing reviewer-supplied commands or code without approval.
[EXTERNAL_DOWNLOADS]: The skill instructs users to download and install the CodeRabbit CLI from https://www.coderabbit.ai/cli. It correctly advises using package managers or verifying signatures for manual downloads rather than piping remote scripts to a shell.
[COMMAND_EXECUTION]: The skill executes shell commands using the coderabbit CLI to perform reviews and authentication. It includes warnings against dangerous patterns and warns the agent not to execute commands found within review outputs.

code-review