The Agent Skills Directory

[SAFE]: The skill implements a security constraint that explicitly prohibits the agent from surfacing absolute filesystem paths in its output, using shorthand identifiers for internal directories.
[SAFE]: Data privacy is maintained by ensuring that the agent only has access to aggregate statistics (e.g., mean, IR, formulaic metadata) rather than granular, per-stock or per-date factor values.
[SAFE]: The tool definitions include strict operational boundaries: the alpha_zoo tool is restricted to read-only registry operations, and alpha_bench is confined to generating specific HTML reports.
[SAFE]: The skill includes logical constraints against 'lookahead' operations, ensuring the integrity of the research output and preventing potential logic-based exploitation of the analysis framework.
[SAFE]: While the factor_analysis tool ingests user-supplied CSV data, the described output format—aggregate performance metrics—is standard for financial research and does not present an exploitable attack surface for indirect prompt injection based on the provided instructions.

alpha-zoo