The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to run npx @booplex/stitch-kit for a "Quick install." This command downloads and executes a package from a third-party scope (@booplex) that does not match the skill author's name (gabelul) or any established trusted provider.
[REMOTE_CODE_EXECUTION]: The skill guides the user to install a plugin directly from a remote Git repository using /plugin marketplace add https://github.com/gabelul/stitch-kit.git. This repository is owned by the skill vendor.
[COMMAND_EXECUTION]: In the troubleshooting section, the skill suggests executing a local script using bash scripts/fetch-stitch.sh "[url]" "temp/output.html" to handle file downloads.
[EXTERNAL_DOWNLOADS]: The skill makes outbound connections to official Google services, including stitch.googleapis.com, storage.googleapis.com, and accounts.google.com for API communication, file storage, and authentication.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its orchestration workflow.
Ingestion points: The skill processes user-supplied design requests (e.g., "Use Stitch to design a simple login screen") in SKILL.md.
Boundary markers: None are present in the provided examples to separate user instructions from system commands.
Capability inventory: The skill uses Bash, Read, and Write tools, along with MCP tools like create_project and generate_screen_from_text.
Sanitization: There is no evidence of input validation or sanitization before passing user data to the generation tools.

stitch-setup