The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions in workflows/text-to-design.md and workflows/edit-design.md direct the agent to fetch HTML and image assets from remote URLs provided by the Stitch MCP tool output.
[COMMAND_EXECUTION]: Documentation within the workflows suggests using run_command to execute curl for downloading assets. However, run_command is not included in the allowed-tools list in SKILL.md, which represents a discrepancy between instructions and defined capabilities.
[INDIRECT_PROMPT_INJECTION]: The workflow for generating design system documentation (workflows/generate-design-md.md) involves reading HTML content from external URLs and using that data to write a .stitch/DESIGN.md file. This ingestion of external data combined with the capability to write to the local filesystem creates a surface for indirect prompt injection.
Ingestion points: Data is fetched via read_url_content from screenshot.downloadUrl and htmlCode.downloadUrl returned by the Stitch API.
Boundary markers: None identified in the synthesis instructions.
Capability inventory: The skill uses the Write tool to create local files and instructions suggest shell execution.
Sanitization: No explicit sanitization or validation of the retrieved HTML content is specified before it is used to generate the design system document.
[NO_CODE]: The skill consists of Markdown-based instructions and configuration without any embedded Python or JavaScript code, which reduces the potential for direct code-based attacks.

stitch-design