The Agent Skills Directory

[SAFE]: The skill securely handles Financial Modeling Prep API credentials by using environment variables (FMP_API_KEY) and implements multi-layered redaction logic (_redact and _cot_redact) to ensure keys are stripped from error logs and generated reports.
[INDIRECT_PROMPT_INJECTION]: The skill fetches external data from the FMP API. The risk is minimized as the ingestion focuses on numeric financial fields which are sanitized through integer/float conversion for calculation, and the output is presented in structured Markdown tables.
Ingestion points: FMP Commitment of Traders API endpoints (stable/commitment-of-traders-report) in scripts/fmp_client.py.
Boundary markers: Implicitly handled by Markdown table formatting in the generated reports produced by scripts/screen_cot_crowding.py.
Capability inventory: Writing JSON and Markdown reports to a local directory; guiding the agent through manual news analysis steps using standard instructions.
Sanitization: Strict numeric conversion of position and open interest data before processing in scripts/cot_index.py.
[DYNAMIC_EXECUTION]: The test suite (scripts/tests/test_fmp_client_contract.py) contains a dynamic import of the standard datetime module. This is used solely for monkeypatching the current date during testing and is a benign development practice.

cot-contrarian-detector