claude-design-sync

Pass

Audited by Gen Agent Trust Hub on Jul 22, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting and processing local files like HTML and instruction templates (prompt.md).
  • Ingestion points: Local project files and source HTML.
  • Boundary markers: No explicit markers are used to delimit external data.
  • Capability inventory: File reading and writing via tools like write_files and get_file (SKILL.md).
  • Sanitization: The skill contains specific instructions to sanitize sensitive data such as company names and NDA details.
  • [COMMAND_EXECUTION]: The skill invokes file management tools (DesignSync, finalize_plan, write_files) to automate the synchronization and manifest updating process.
  • [DATA_EXFILTRATION]: The skill moves local files to an external destination (Claude Design). This is a core functionality and is managed through a mandatory confirmation step and instructions to strip sensitive identifiers prior to the transfer to this well-known service.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 22, 2026, 01:14 AM
Security Audit — agent-trust-hub — claude-design-sync