The Agent Skills Directory

[COMMAND_EXECUTION]: The skill workflow requires executing local shell commands such as npm install, npm run build, and npm run dev. This behavior is expected for a development-focused skill but involves executing scripts found within the local project directory.
[DATA_EXFILTRATION]: The skill provides instructions for deploying code to external hosting platforms (e.g., Vercel, Netlify) and using tunneling tools. While the skill explicitly instructs the agent to seek user consent and verify that no secrets or .env files are included, these actions represent a data exfiltration surface.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing data from a prototype path.
Ingestion points: The agent reads the 'QA'd prototype' files from a local path specified by the user.
Boundary markers: The instructions do not define specific delimiters or isolation techniques for the content of the prototype files.
Capability inventory: The agent has the capability to execute subprocesses (npm commands) and perform network uploads via deployment CLIs.
Sanitization: The skill includes manual sanitization steps, instructing the agent to 'check for secrets/env files' and 'confirm host' before proceeding with high-impact actions.

design-share