matter-workspace
Pass
Audited by Gen Agent Trust Hub on Jul 3, 2026
Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by design, as it ingests and displays user-controlled matter information.
- Ingestion points: Data provided during the intake interview (client names, slugs, key facts) is stored in
matter.mdandhistory.mdand subsequently read into the agent's context. - Boundary markers: The skill uses Markdown templates with clear headers and bold labels to delimit content, although it lacks explicit warnings to ignore embedded instructions.
- Capability inventory: The skill is restricted to file management operations (read, write, list, move) within the
~/.claude/plugins/config/claude-for-legal/directory. - Sanitization: Content is interpolated directly into templates without specific sanitization or validation of the user-provided text.
- [NO_CODE]: The skill is implemented entirely through natural language instructions and Markdown templates, relying on the agent's built-in file system tools rather than custom scripts or executables.
Audit Metadata