The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to manage design assets on the local filesystem.
Evidence: In workflows/text-to-design.md, the instructions suggest: "Use curl -o via run_command or similar" to download HTML and screenshots to the .stitch/designs directory.
[EXTERNAL_DOWNLOADS]: The skill fetches content from remote URLs provided by the MCP tool components.
Evidence: workflows/generate-design-md.md describes a step to "Metadata fetch" and then "Asset download" using the read_url_content tool on URLs derived from htmlCode.downloadUrl.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where untrusted data (external HTML) is ingested and analyzed to influence the agent's internal state and future outputs.
Ingestion points: In workflows/generate-design-md.md, the agent uses read_url_content to fetch HTML code from external projects for the purpose of design system synthesis.
Boundary markers: No delimiters or instructions to ignore embedded content are specified when processing the retrieved HTML code.
Capability inventory: The skill has the capability to write files to the local directory (Write), execute shell commands (run_command), and invoke generative design tools (StitchMCP).
Sanitization: There is no evidence of sanitization, filtering, or validation of the fetched HTML content before it is analyzed to generate the .stitch/DESIGN.md documentation.

stitch-design