The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to process external documents retrieved from Virtual Data Rooms (VDRs) and tool-generated extractions.
Ingestion points: Data enters the agent context through the VDR inventory and the outputs of tools like Luminance or Kira in the 'QA the output' and 'Judgment layer' steps.
Boundary markers: There are no explicit delimiters or instructions to the agent to ignore potentially malicious instructions embedded within these external legal documents.
Capability inventory: The skill utilizes file-reading for configuration and matter-specific files, and file-writing to store findings in local directories.
Sanitization: There is no evidence of sanitization, escaping, or filtering of the content ingested from the external documents before it is processed by the agent's judgment layer.
File System Interaction: The skill interacts with the local file system to load practice-level configurations and matter-specific data.
Evidence: It accesses and writes to paths such as ~/.claude/plugins/config/claude-for-legal/corporate-legal/CLAUDE.md and ~/.claude/plugins/config/claude-for-legal/corporate-legal/matters/<matter-slug>/.
Context: While these paths are specific to the skill's application environment, the ability to read and write to the local directory is a consideration for data exposure, particularly if user-controlled input like matter slugs were to influence file paths through path traversal.

ai-tool-handoff