The Agent Skills Directory

[SAFE]: The skill employs a robust security architecture that implements a clear separation between untrusted 'audit' data and a sanitized 'safeChart' projection. This architecture ensures that the AI model only interacts with validated, structured data rather than free-form text provided by users.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests external data (transcribed charts), it effectively mitigates injection risks through several layers of defense:
Sanitization: All text inputs are checked for length limits and the presence of control characters.
Validation: Fields are strictly validated against whitelist enums (e.g., specific Heavenly Stems, Earthly Branches, and Qi Men stars).
Isolation: The 'safeChart' projection replaces all raw user text with existence flags and stable references (provenanceRef), ensuring that any malicious instructions hidden in the transcription are stripped before they reach the model's context.
[DATA_EXPOSURE]: There is no evidence of hardcoded credentials, access to sensitive local files (like .env or .ssh), or unauthorized data transmission. The skill operates entirely locally on the provided input.
[COMMAND_EXECUTION]: The skill does not use any dangerous functions such as eval() or exec(), nor does it spawn subprocesses or execute shell commands. It relies on standard JavaScript logic and local internal modules.
[EXTERNAL_DOWNLOADS]: The skill does not perform any network requests or external downloads. All dependencies are internal to the project environment.

qimen