The Agent Skills Directory

[SAFE]: The skill provides legitimate utility for software testing audits. It includes scripts for static analysis of test code and a pytest plugin for observing test behavior.
[COMMAND_EXECUTION]: The skill instructs the agent to run bundled Python scripts and external test runners (e.g., pytest, cargo nextest). This behavior is consistent with the skill's purpose as a test audit tool. The instructions emphasize performing audits in isolated scratch copies for safety.
[DYNAMIC_EXECUTION]: The pytest_reach_probe.py script monkey-patches subprocess.Popen.communicate to record error output during test execution. This is a common instrumentation technique and is explicitly documented within the skill's references.
[INDIRECT_PROMPT_INJECTION]: The skill operates on user-provided source and test files. While this presents an ingestion surface for untrusted data, the bundled analysis scripts use AST and token-based parsing rather than execution of untrusted strings. The reach probe, which does execute code, is designed to run the existing test suite, which is an expected behavior for this class of tool.

test-audit