The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the system-level curl command to download HTML content from URLs provided in Stitch project metadata (e.g., screenshot.downloadUrl).
[COMMAND_EXECUTION]: A PowerShell utility script (scripts/stitch-to-react.ps1) performs automated environment setup, including directory creation, file writing, and execution of package manager commands.
[REMOTE_CODE_EXECUTION]: The skill invokes package managers (npm, npx) to download and execute code from public registries for project initialization and component installation (e.g., npm create vite@latest, npx shadcn@latest).
[EXTERNAL_DOWNLOADS]: Fetches external screen assets, HTML, and design data via the Stitch MCP server and network-enabled CLI tools.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by design as it processes content from untrusted or external sources to drive its autonomous "Build Loop" workflow.
Ingestion points: Reads task instructions from next-prompt.md (the "baton") and HTML/JSON data retrieved via the Stitch MCP tools (SKILL.md, Part 3).
Boundary markers: The instructions do not define boundary markers or include safety directives to ignore embedded instructions within the processed data.
Capability inventory: The skill has access to shell command execution (curl), file system manipulation (stitch-to-react.ps1), and remote package installation/execution (npm, npx).
Sanitization: There is no evidence of input validation or sanitization for content retrieved from external design screens or local task files.

stitch-design