The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Node.js scripts (gather-local-context.mjs and gather-pr-context.mjs) that execute local git and gh (GitHub CLI) commands to collect repository and pull request data. These commands use validated inputs, such as integer PR numbers, to minimize the risk of command injection.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests and processes untrusted data from external pull requests that could contain adversarial instructions designed to influence the agent's behavior.
Ingestion points: The agent receives PR metadata (title, body, author) and full unified code diffs retrieved via GitHub CLI or local git commands.
Boundary markers: The skill's instructions do not define specific delimiters or instructions to ignore or treat as data any potential commands embedded within the PR content.
Capability inventory: The agent is granted repository access with the ability to use Read, Glob, and Grep tools to explore the codebase based on the PR content.
Sanitization: There is no evidence of sanitization or filtering applied to the ingested PR metadata or diff content before it is processed by the agent.

pr-test-checker