The Agent Skills Directory

[COMMAND_EXECUTION]: The skill workflow involves executing arbitrary project CLI commands (e.g., uv run python -m smart_sales.xxx.cli) and using networking tools like curl. This allows the agent to execute any executable available in the project environment.
[REMOTE_CODE_EXECUTION]: The skill specifically instructs the agent to "write a dedicated running script" (run.py) at runtime when existing tools are unavailable, and subsequently execute it using uv run. This dynamic code generation and execution pattern is a high-risk vector if the generated logic is influenced by untrusted inputs.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it processes external test cases which may contain malicious instructions.
Ingestion points: Test data files located in .tests/{test-name}/wave-{N}/input/case-{N}.json.
Boundary markers: Absent. The instructions do not specify any delimiters or warnings to ignore instructions embedded within the test cases.
Capability inventory: File system write access (creating test directories and reports), subprocess execution (uv run), and network operations (httpx, curl).
Sanitization: Absent. There are no requirements to validate or sanitize the JSON input before it is used to invoke agents or processed by the test runner.

agent-testing