The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process external, untrusted content such as vendor agreements, NDAs, and SaaS subscriptions provided via file paths, links, or text pasting. This creates a potential surface where instructions hidden within a document could influence the agent's behavior.
Ingestion points: The skill accepts data from external files, Google Drive links, and raw text input as defined in the SKILL.md workflows.
Boundary markers: The instructions do not specify explicit delimiters or 'ignore embedded instructions' warnings to help the agent distinguish between the document content being analyzed and the review logic.
Capability inventory: The skill has the capability to read local configuration files (~/.claude/plugins/config/...), invoke other review skills, and generate output memos or CLM records.
Sanitization: There are no explicit sanitization or validation steps mentioned for the ingested document content before processing.
Local File System Access: The skill reads from a specific local directory (~/.claude/plugins/config/claude-for-legal/commercial-legal/CLAUDE.md) to load its 'playbook' and user preferences. While this is necessary for its functionality, it establishes a dependency on the integrity and privacy of that local configuration file.

review