The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute local development commands, including git for version control and gh for interacting with GitHub's API.
[COMMAND_EXECUTION]: The primary logic is executed through a local script scripts/reviewer.py, which is called as a subprocess to perform the automated reviews.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content from external Pull Requests. Malicious instructions embedded in code changes or PR descriptions could theoretically influence the output of the model being tested.
[CREDENTIALS_UNSAFE]: The skill requires and uses sensitive environment variables (ANTHROPIC_API_KEY and GITHUB_TOKEN) to authenticate with model providers and GitHub. These are passed to the execution environment of the internal reviewer script as intended by the tool's design.

prompt-test