The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute scripts/fetch-stitch.sh, which wraps curl to download design assets. While the script properly quotes its variables, the invocation involves passing remote URLs as arguments, which could lead to command injection if the agent does not properly escape the input.
[EXTERNAL_DOWNLOADS]: The skill fetches HTML designs and screenshot images from Google Cloud Storage URLs provided by the stitch:get_screen MCP tool. These downloads are performed at runtime to facilitate the design-to-code workflow.
[REMOTE_CODE_EXECUTION]: Phase 4 of the skill instructions permits the agent to run npm install to set up validation dependencies and npx react-native start to launch development packagers. The skill includes a safety gate requiring explicit user permission before executing these commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external Stitch projects to extract theme tokens and map HTML to components, creating a surface for indirect prompt injection.
Ingestion points: Phase 1 and Phase 2 involve downloading and reading HTML files from external URLs to parse tailwind.config blocks and DOM structures.
Boundary markers: Absent; there are no specific instructions to treat the downloaded HTML content as untrusted data or to use delimiters to prevent the agent from following instructions embedded within the design files.
Capability inventory: The agent has access to Bash (shell execution), Write (file system modification), and web_fetch tools.
Sanitization: The skill provides scripts/validate.js, which uses the @swc/core AST parser to statically analyze generated code for anti-patterns, providing a layer of validation for the agent's output.

stitch-react-native