The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and process update proposals generated by the playbook-monitor agent. These proposals may contain instructions that could influence the behavior of the legal practice profile. However, the skill explicitly requires a human review flow and attorney confirmation for all edits and diffs before they are applied to the configuration, significantly mitigating the risk of unintended behavior.
Ingestion points: Processes data from the pending proposals file generated by a separate agent.
Boundary markers: Utilizes a step-by-step approval flow (Accept, Reject, Edit, Defer) to ensure logical separation between processing and execution.
Capability inventory: Possesses the ability to read, write, and archive files within the ~/.claude/plugins/config/ directory.
Sanitization: Requires showing the exact diff to the user for validation before application, providing a visual sanitization step for the operator.
File System Interaction: The skill performs file read and write operations on specific configuration files located in ~/.claude/plugins/config/claude-for-legal/commercial-legal/CLAUDE.md. These operations are central to the skill's purpose of maintaining practice guidelines and are restricted to specific, non-system paths.

review-proposals