design-sync-upload
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external, user-supplied data such as design documents, tokens, and image assets to either upload them via a tool (MCP) or prepare them in a staging folder. This ingestion of untrusted data combined with file-write and network-write capabilities creates a potential attack surface where malicious instructions hidden in design files could influence the agent's behavior.
- Ingestion points: Processes local files including
DESIGN.md, token files (JSON/CSS), and image assets. - Boundary markers: The skill instructions do not explicitly define delimiters to isolate the content of the design files from the agent's instructions, though it does provide a manual upload guide with safety warnings.
- Capability inventory: The skill uses the
DesignSyncMCP tool to perform network operations (listing projects, writing files, deleting files) and writes new files to the local file system (staging directory and upload guide). - Sanitization: The skill includes positive practices such as instructing users to anonymize sensitive data (customer data, revenue) and verify font licenses before including assets in the sync process.
Audit Metadata