The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes regulatory rule text and policy content provided by users or external sources, which could potentially contain hidden instructions designed to influence the agent's behavior.
Ingestion points: Regulatory rule text (from user or research tools) and current approved policy text (from user or files) as described in Step 1 of SKILL.md.
Boundary markers (present): The skill instructs the agent to use [verify] tags and clear reviewer notes to delineate model-generated or unverified information.
Capability inventory: The skill has the ability to read and write files within the ~/.claude/plugins/config/ directory structure and can perform network-based research via web search or dedicated legal connectors.
Sanitization (absent): The skill emphasizes minimal edits and source tagging but does not specify logic for filtering or sanitizing instructions that might be embedded within the legal text it processes.
File System Access: The skill interacts with the local file system to manage policy libraries and gap trackers.
Context: Access is restricted to specific configuration paths (~/.claude/plugins/config/claude-for-legal/), and the skill is explicitly prohibited from overwriting source policy documents, instead writing proposals to new, dated files.

policy-redraft