Drawing
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions for the agent to use
openclawCLI commands (openclaw models status,openclaw models set-image) to manage image model configurations. These are documented platform commands for the target environment. - [INDIRECT_PROMPT_INJECTION]: The skill implements a memory system that reads preferences and previous successful prompts from local files in
~/Clawic/data/drawing/. - Ingestion points: Files including
memory.md,winning-prompts.md, andstyle-notes.mdare used to customize agent behavior. - Boundary markers: The prompt templates use structured headers (e.g., 'Subject:', 'Style:', 'Composition:') which help maintain context separation.
- Capability inventory: The agent can execute platform-specific CLI commands and generate prompts for third-party image models.
- Sanitization: No specific sanitization of the retrieved preference data is documented, though the data is generated by the agent itself based on user interaction.
Audit Metadata