The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard development commands like grep, jq, and ls to search the local codebase for tool definitions in Step 1. These operations are scoped to identifying configuration files and source code patterns without executing the found code.
[EXTERNAL_DOWNLOADS]: In Step 2, the skill fetches updated documentation from the vendor's official sources (ratel-ai.com, docs.ratel.sh, and github.com/ratel-ai). These are well-known vendor resources used to ensure the audit adheres to the latest platform standards.
[DATA_EXFILTRATION]: No sensitive data access or external transmission of user code was detected. The skill reads local files to generate a report and writes that report back to the local .ratel/ directory.
[PROMPT_INJECTION]: The instructions do not contain attempts to override agent safety guidelines. It uses instructional language like 'IMPORTANT' and 'CRITICAL' in a benign, operational context related to the audit process.
[CREDENTIALS_UNSAFE]: No hardcoded credentials or instructions to handle secrets unsafely were found. It follows safe practices by analyzing public tool definitions.

ratel-tune-definitions