The Agent Skills Directory

[SAFE]: The skill implements a redaction mechanism via the _redact function that effectively prevents the FMP API key from being exposed in error messages, exception strings, or logs.
[COMMAND_EXECUTION]: The use of subprocess.run within scripts/tests/test_analyze_news_reaction.py is identified as safe. It is employed solely for executing CLI tests of the skill's own scripts during the testing process and does not represent a runtime risk for the AI agent.
[EXTERNAL_DOWNLOADS]: The skill retrieves price series data from Financial Modeling Prep (financialmodelingprep.com), which is a well-known financial information service. This activity is restricted to the skill's primary function and follows standard industry practices for data ingestion.
[INDIRECT_PROMPT_INJECTION]: The skill processes news headlines from external sources, creating a potential surface for indirect prompt injection.
Ingestion points: External data enters through the events-json file, specifically the event and source_url fields in scripts/analyze_news_reaction.py.
Boundary markers: The reports generated by the skill do not use explicit delimiters to isolate news content from instructions.
Capability inventory: The skill possesses no capabilities for executing code or modifying system settings based on the news data, as its logic is strictly statistical.
Sanitization: Although news headlines are not sanitized for potential injection patterns, the drift-significance math used to generate the final verdict acts as a natural buffer against non-financial instructions.

news-reaction-failure-analyzer