The Agent Skills Directory

Structured Local File Access: The skill is designed to interact with a specific local configuration and document storage structure (~/.claude/plugins/config/claude-for-legal/...).
While this involves reading potentially sensitive corporate data (board compositions, prior resolutions), the skill uses this data exclusively to inform its drafting process and writes outputs to designated matter folders.
Indirect Prompt Injection Surface: The skill processes external data in the form of "precedents" or "seed documents" found in the user's repository.
Ingestion points: Data enters the context via CLAUDE.md, the consents repository, and user-pasted precedents.
Boundary markers: The skill implements "hard stops" that require explicit user selection of precedents and confirmation of action types before proceeding.
Capability inventory: The skill is capable of writing drafted documents and signatory checklists to the local filesystem using agent tools.
Sanitization: The instructions focus on extracting specific legal language (resolutions, recitals) to match house formatting, though it relies on the underlying agent's safety filters for content validation.
Risk Mitigation Gates: The skill includes sophisticated internal logic to identify "major one-off actions" (e.g., M&A, financing) and implements a hard stop if a user requests same-day signature for such actions, requiring a specific choice of path to proceed. This is a strong safety feature that prevents the agent from being pressured into generating high-risk documents without proper review cycles.

written-consent