The Agent Skills Directory

[PROMPT_INJECTION]: The skill analyzes untrusted code from pull requests, creating a surface for indirect prompt injection.
Ingestion points: The agent reads file content using commands like grep, head, and tail as instructed in SKILL.md.
Boundary markers: No delimiters or instructions are provided to help the agent distinguish between code data and potentially malicious instructions within the PR content.
Capability inventory: The agent has the ability to execute shell commands (gh, curl) and perform network operations to the GitHub API using the GITHUB_TOKEN.
Sanitization: No sanitization or escaping of the file content is mandated before it is used in the review body or API payloads.
[COMMAND_EXECUTION]: The skill relies on shell commands to perform its core functionality.
Evidence: Templates for gh api, curl, grep, head, and tail are provided in both README.md and SKILL.md to facilitate finding code lines and posting review comments.

github-pr-review