The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted source code, creating a surface for indirect prompt injection. Ingestion points: reads files found via Glob and git diff patterns defined in Step 2. Boundary markers: The skill does not define explicit delimiters for untrusted code content during the analysis phase. Capability inventory: utilizes file system exploration (Glob), git version control commands, and markdown report generation. Sanitization: includes strict instructions to exclude sensitive files (secrets, keys, credentials) from exploration and prevents the agent from reproducing detected secret values in its output report.
[COMMAND_EXECUTION]: The skill directs the agent to execute file system operations such as directory listing and file reading, as well as version control commands like git diff. These are restricted to the local environment and the specific scope defined by the user in the audit request.
[DATA_EXFILTRATION]: While the skill accesses codebase data, it implements robust defensive measures to prevent sensitive data exposure. It defines a blacklist for sensitive files (e.g., .env, .pem, .key, credentials) and mandates that no secret values found during analysis should be included in the generated report.

code-review