The Agent Skills Directory

[SAFE]: Comprehensive review of the skill instructions and source code reveals no malicious patterns, obfuscation, or attempts to bypass agent security filters.
[EXTERNAL_DOWNLOADS]: The skill retrieves research data from official and well-known legal services, including the Free Law Project's CourtListener API, the Electronic Code of Federal Regulations (eCFR), and the Federal Register. These connections are used exclusively for data retrieval to support document drafting.
[COMMAND_EXECUTION]: Local Python scripts (verify-legal.py, courtlistener.py, and ecfr_lookup.py) are executed to perform document auditing and API interaction. These scripts are bundled with the skill and are transparent in their functionality.
[INDIRECT_PROMPT_INJECTION]: The skill maintains an attack surface for indirect prompt injection by processing external data from legal research APIs into the agent's context.
Ingestion points: External case law and regulatory text are downloaded and stored in research/cases.json and research/regulations.json within the document workspace.
Boundary markers: No specific delimiters or "ignore previous instructions" guards are implemented for the ingested research data.
Capability inventory: The skill possesses capabilities for local file system manipulation (writing and reading .docx and .json files) and network operations (performing HTTP GET requests to legal databases).
Sanitization: The ecfr_lookup.py script includes logic to strip XML tags from retrieved regulatory data using regular expressions, providing basic sanitization of external content.

legal-writer