The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Python helper script (scripts/_skill_cli.py) that uses subprocess.run to execute commands. The implementation follows security best practices by passing arguments as a list and avoiding the shell, which prevents common command injection vulnerabilities. The skill also references a local utility script (.agents/skills/heyeddi/scripts/suggest_next_skill.py) for workflow orchestration.
[EXTERNAL_DOWNLOADS]: The skill configures several MCP servers that are executed via npx, including PrimeVue, Quasar, Playwright, and Penpot. These references target well-known open-source projects and developer tools within the web development ecosystem. The use of the @latest tag for Quasar and Playwright is noted as a common practice for maintaining up-to-date tooling in this context.
[INDIRECT_PROMPT_INJECTION]: The skill implements a feedback loop where the agent reads the stdout of CLI scripts and takes screenshots of a live Vite server using Playwright to perform visual QA. This processing of external data constitutes an attack surface where malicious content in the UI or command output could potentially influence agent behavior.
Ingestion points: stdout of terminal scripts, Playwright screenshots of the local development server.
Boundary markers: Not explicitly defined in the prompt instructions for data processing.
Capability inventory: File writing within the project root, command execution (via subprocess.run), and network access (via npx and browser tools).
Sanitization: No specific sanitization or filtering logic is described for the content captured from tool outputs.

vue-design-system