scenario-brand-kit
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The SKILL.md file instructs the agent to use the curl command to download SVG assets from URLs generated by platform tools to facilitate local recoloring and variant creation. It also mentions recommending the use of npx commands to the user for installing supplemental skills from the author's repository.
- [INDIRECT_PROMPT_INJECTION]: The skill architecture in SKILL.md presents an attack surface for indirect prompt injection as it ingests and evaluates external data. 1. Ingestion points: The skill processes user-defined brand specifications and SVG assets retrieved via the asset_get tool in SKILL.md. 2. Boundary markers: The instructions emphasize a 'spec sheet' for consistency but do not provide technical delimiters to isolate external data from agent instructions. 3. Capability inventory: The skill utilizes tools for network downloads (curl), asset analysis (asset_analyze), and file management (upload_asset) across its operations in SKILL.md. 4. Sanitization: No explicit validation or sanitization of the SVG XML content or specification text is described in SKILL.md before these items are processed.
Audit Metadata