The Agent Skills Directory

[COMMAND_EXECUTION]: The helper scripts inspect_vitest.py and run_vitest.py execute shell commands to run Node.js and Vitest. These operations are protected by a custom security model that sanitizes the environment and validates commands.
[DATA_EXPOSURE]: The skill reads project configuration files (e.g., package.json, .nvmrc). This is standard behavior for a test runner and no sensitive data is exfiltrated or exposed beyond the intended scope.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data but mitigates injection risks by normalizing inspection reports into enums and counts, and by strictly filtering and length-limiting any project text rendered to the console.
Ingestion points: Project files are read in scripts/inspect_vitest.py and scripts/run_vitest.py.
Boundary markers: inspect_vitest.py normalizes output into enums to prevent the agent from interpreting repository text as instructions.
Capability inventory: Subprocess calls are strictly limited to node and vitest after validation.
Sanitization: All project-controlled text rendered in output is length-limited and filtered for control characters.
[SAFE]: The presence of dedicated security unit tests (scripts/test_inspect_vitest.py, scripts/test_run_vitest.py) and a clearly documented security model indicates a high standard of safety and defensive programming.

vitest