customize
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- User-Driven Configuration Updates: The skill modifies local configuration files located in
~/.claude/plugins/config/based on natural language instructions. This creates a data ingestion surface where user-provided content is stored and later interpreted by other plugins, which is a potential vector for indirect prompt injection. - Safety Marker Protection: The agent is explicitly instructed to refuse requests to disable load-bearing guardrails, such as
[review]or[verify]tags. This instruction acts as a defensive measure to maintain output integrity. - Change Confirmation: The process requires the agent to display the intended change and its consequences before writing to disk, providing a manual checkpoint to mitigate the risk of unintended configuration modifications.
Audit Metadata