The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external court decisions and legal statutes.
Ingestion points: The skill ingests untrusted data through the outputs of the get_decision and get_legal_text tools in SKILL.md.
Boundary markers: The instructions establish a strict protocol (Rule 2) that distinguishes between machine-generated summaries and verbatim court text, instructing the agent to ignore snippets and summaries in favor of the full source text.
Capability inventory: The skill provides tools for searching and reading legal data. It does not appear to have access to file-system write operations or arbitrary shell execution.
Sanitization: The skill enforces a rigorous verification process (Rule 1 and Rule 3) requiring the agent to read the complete text before making any authoritative claims, mitigating the risk of being misled by partial or manipulated data.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of a remote MCP server located at https://librejustice.fr/mcp.
This endpoint is the official service provider for the LibreJustice legal research tools.
The installation instructions in references/install-mcp.md use standard platform-specific commands for Claude Code and Codex, representing legitimate configuration of the environment to enable the skill's functionality.

recherche-jurisprudence