The Agent Skills Directory

Indirect Prompt Injection Surface: The skill ingests detailed user-provided facts regarding worker arrangements (e.g., job descriptions, control factors, economic terms) to execute classification tests. This creates a surface where external data could potentially contain instructions aimed at influencing the agent's reasoning. The skill lacks explicit sanitization or boundary markers for this untrusted input.
Ingestion points: Step 1 Information Gathering (The work, Control, Economics, etc.).
Boundary markers: Absent for user-provided facts.
Capability inventory: Writes analysis results to the local filesystem at ~/.claude/plugins/config/claude-for-legal/employment-legal/matters/.
Sanitization: Absent.
Local File System Access: The skill is designed to interact with a specific local directory structure (~/.claude/plugins/config/claude-for-legal/) to load jurisdictional footprints, practice policies, and matter-specific context. While this supports the workflow's state management, it involves reading and writing configuration files which should be monitored within the host environment.
Safety Gates: The skill incorporates two positive security and safety patterns: a 'Prospective-only hard gate' to prevent out-of-scope liability assessments, and a 'Consequential-action gate' that restricts final recommendations based on the user's role. These mechanisms serve to reduce the risk of misuse in sensitive legal contexts.

worker-classification