The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted pull request data including code diffs, titles, and descriptions. This creates a surface for indirect prompt injection where malicious instructions embedded in a PR could influence the agent to bypass review standards or perform unintended actions.\n
Ingestion points: PR metadata and code changes retrieved via gh commands and file system access as described in SKILL.md and references/github-pr-review-actions.md.\n
Boundary markers: None identified in the provided instructions to separate untrusted content from the agent's instructions.\n
Capability inventory: File read/write access and gh CLI execution for interaction with the GitHub API.\n
Sanitization: No explicit sanitization or escaping of the untrusted PR content is defined before processing.\n- [COMMAND_EXECUTION]: The skill instructions in references/github-pr-review-actions.md and references/ci-optimized-workflow.md rely on the gh (GitHub CLI) tool to retrieve PR data and post review results. These commands are executed in the shell to interact with the official GitHub API.\n- [EXTERNAL_DOWNLOADS]: The skill recommends installing additional skills from vercel-labs (a well-known service) and the author xtone. These references are documented as part of the intended workflow for extending code review capabilities.

code-review