The Agent Skills Directory

[COMMAND_EXECUTION]: Subprocess calls in validation/smoke.py and tests/test_liquidity_stress.py are used exclusively for internal validation and testing of the skill's core script. These calls are controlled, using sys.executable to run local files without shell interpolation of untrusted input.
[INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface via CSV files. The risk is addressed by the scripts/stress_liquidity.py file, which implements strict numeric validation, finite-number checks, and unique identifier requirements. The instructions explicitly direct the AI to treat inputs as unverified evidence, providing a logical boundary against data-driven attacks.
[SAFE]: The skill operates entirely offline and follows best practices by delegating data acquisition to specialized sibling skills rather than implementing its own authentication or network logic.

portfolio-liquidity-stress-test