md-improve
Pass
Audited by Gen Agent Trust Hub on Mar 16, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill performs legitimate analysis of user-agent interactions to optimize the CLAUDE.md instruction files.
- [PROMPT_INJECTION]: The skill processes conversation history which presents an indirect prompt injection surface. Ingestion points: Conversation history analysis in Phase 1. Boundary markers: Absent. Capability inventory: 'Edit' tool for filesystem writes and 'bunx' for command execution. Sanitization: Mitigated by mandatory user approval in Phase 3 (AskUserQuestion) and verification in Phase 5.
- [COMMAND_EXECUTION]: The skill runs 'bunx prettier --check' to verify formatting. This is an execution of a well-known technology tool and is documented neutrally.
- [SAFE]: File access is limited to instruction-specific files (CLAUDE.md) at the global and project levels, which is required for the skill's primary objective.
Audit Metadata