The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface identified in the design system synthesis workflow where the agent processes untrusted external data.
Ingestion points: In workflows/generate-design-md.md, the agent is instructed to use read_url_content to fetch HTML code from remote URLs for analysis.
Boundary markers: Absent. The instructions do not include delimiters or warnings to ignore embedded instructions within the fetched HTML content.
Capability inventory: The skill utilizes the Write tool to create local files (.stitch/DESIGN.md) and interacts with the StitchMCP service, which could be influenced by malicious instructions in the ingested data.
Sanitization: Absent. There is no mention of validating or filtering the remote HTML content before processing it for design synthesis.
[EXTERNAL_DOWNLOADS]: The skill performs remote downloads of assets to the local filesystem.
Evidence: Workflows in workflows/text-to-design.md and workflows/edit-design.md instruct the agent to download HTML and screenshots from URLs provided by the Stitch MCP tool output and save them to the .stitch/designs directory.
[COMMAND_EXECUTION]: The skill instructs the agent to use shell commands for file management.
Evidence: In workflows/text-to-design.md, the instructions explicitly suggest using curl -o via run_command to perform the asset downloads to the local project environment.

stitch-design