The Agent Skills Directory

[DATA_EXPOSURE]: The skill implements a sophisticated 'Information Isolation Iron Rule' (⚠️ 信息隔离铁规) that mandates the separation of technical astrological data from user personal narratives. It utilizes two distinct files (structured_data.md and user_context.md) to ensure that sensitive biographical information is not mixed with technical data, which also serves to prevent bias in downstream analysis tools.
[COMMAND_EXECUTION]: The skill requires the execution of local Python scripts (e.g., dasha_query.py) and provides specific Python code snippets utilizing the fitz (PyMuPDF) library for PDF processing. These commands are necessary for the skill's core functionality and are restricted to the local environment without utilizing untrusted remote sources.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is processing untrusted user inputs (PDFs, screenshots, and text). It mitigates potential injection attacks through a rigorous 16-point mathematical validation framework and 'double-channel' extraction, where text-layer data is cross-referenced against visual OCR renderings to identify and resolve inconsistencies.
[REMOTE_CODE_EXECUTION]: No patterns of remote code execution or unauthorized downloads were found. All external dependencies mentioned, such as the vedic-calculator and fitz library, are standard tools expected within the platform's environment.
[SAFE]: The skill demonstrates defensive security practices by explicitly forbidding the agent from using web-based PDF viewers ('open_browser_url') to prevent accidental data leakage to external services. The author, 'cnwu16', has designed the skill with a clear hierarchy of trust, prioritizing local calculation engines over external data sources.

vedic-reader