stitch-manage-design-system

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to parse .stitch/DESIGN.md files to extract design system tokens such as colors, fonts, and layout rules from YAML frontmatter. This process introduces a surface for indirect prompt injection if the design file contains malicious instructions aimed at the agent.
  • Ingestion points: Reads .stitch/DESIGN.md from the local file system.
  • Boundary markers: None identified; the agent is instructed to directly parse and apply the extracted configurations.
  • Capability inventory: Includes execution of a local Python script and multiple Stitch-specific MCP tool calls for project and asset management.
  • Sanitization: No sanitization or validation of the markdown content is specified.
  • [COMMAND_EXECUTION]: Documentation for the skill includes a workflow step that executes a local Python script (scripts/upload_to_stitch.py) using python3. The script handles large file uploads and requires an API key passed via command-line arguments.
  • Evidence: The reference/tool-schema.md file provides the command: python3 <SKILL_DIR>/scripts/upload_to_stitch.py --project-id <PROJECT_ID> --file-path /path/to/DESIGN.md --api-key <API_KEY>.
  • Source: The skill and its scripts are identified as being imported from the google-labs-code/stitch-skills repository.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:52 AM
Security Audit — agent-trust-hub — stitch-manage-design-system