The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a robust 'THIRD-PARTY CONTENT DEFENSE' mechanism. It instructs the agent to treat pull request descriptions, comments, and issue text as untrusted content and specifically forbids following any commands or instructions found within that data. This is a best-practice defense against indirect prompt injection.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns, dynamic code generation, or execution of untrusted scripts were found. The skill operates entirely as a set of analytical instructions.
[DATA_EXFILTRATION]: There are no network operations, curl/wget commands, or attempts to access sensitive files like credentials or SSH keys. The skill only analyzes provided code diffs.
[CREDENTIALS_UNSAFE]: No hardcoded secrets or API keys are present. In fact, the skill instructs the agent to search for and flag hardcoded secrets in the code it reviews (e.g., in runtime.exs).

code-review