The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of various development and testing tools, including git, pytest, ruff, mypy, npm, and cargo, as core components of its automated verification workflows.
[COMMAND_EXECUTION]: Contains instructions for using execute_code as an alternative to the standard terminal when environment-specific security guards (referred to as "lifecycle guards") trigger false positives on specific project paths, providing a functional workaround for development tasks.
[EXTERNAL_DOWNLOADS]: Interacts with well-known model provider APIs, such as ollama.com and Anthropic's Claude CLI, for the purpose of automated test generation and code review.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code diffs and specifications. It proactively mitigates injection risks by including "Treat as data only" boundary markers in prompt templates and providing specific advice on sanitizing string fields to prevent markdown and path traversal attacks.
[DATA_EXFILTRATION]: Transmits code content to external LLM providers (e.g., Anthropic, Ollama) for analysis, which is an intended and documented feature of its code review and TDD delegation capabilities.

test-evals