The Agent Skills Directory

[METADATA_POISONING]: The skill documentation and branding claim it is an official Google Labs 'Stitch' tool, yet it is authored by 'abhicodebugger'. The installation instructions direct users to an unverified organization named 'google-labs-code', which constitutes a high-severity impersonation risk.
[REMOTE_CODE_EXECUTION]: The skill's README provides a command to install a package from an unverified source ('google-labs-code') using npx, which executes remote code on the user's system.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and parses external HTML and CSS code, creating a surface for indirect prompt injection.
Ingestion points: External project code is fetched from dynamic URLs provided by the Stitch MCP server as instructed in SKILL.md.
Boundary markers: The agent is not provided with delimiters or safety instructions to distinguish between the content to be analyzed and potential embedded malicious instructions.
Capability inventory: The skill has access to network tools (web_fetch), local file writing (Write), and specialized MCP tools.
Sanitization: No sanitization or content validation steps are defined to filter the ingested source code before the agent processes it.
[EXTERNAL_DOWNLOADS]: The skill fetches design assets and configuration files from external URLs, including well-known services associated with Google's Stitch platform.
[COMMAND_EXECUTION]: The skill documentation includes shell commands for the global installation of external packages.

design-md