The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill included custom Python scripts that fetch legal metadata and verify reference accessibility through official government and academic APIs, including Légifrance, EUR-Lex, HUDOC (ECHR), and HAL (French Open Archive). These operations target well-known, authoritative legal services and scholarly repositories essential to the skill's academic purpose.\n- [DATA_EXFILTRATION]: No unauthorized data exfiltration patterns were detected. Network communications are strictly limited to fetching metadata and validating URLs for legal references. The scripts do not transmit user data to unknown third-party domains.\n- [PROMPT_INJECTION]: The instructions include a comprehensive anti-hallucination framework that requires the agent to prioritize official data retrieval over its internal training data. It enforces a strict sequence (Search -> Find -> Read -> Verify -> Cite) to prevent erroneous legal assertions.\n- [REMOTE_CODE_EXECUTION]: The skill provides local Python scripts for document parsing (DOCX XML extraction) and citation formatting. These scripts use standard libraries and do not execute remote code or implement unsafe dynamic execution patterns.

mandarinat