The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to download an additional skill named 'performance' from an external GitHub repository (https://github.com/addyosmani/web-quality-skills). This introduces a dependency on code hosted outside the primary skill environment.
[REMOTE_CODE_EXECUTION]: By installing and invoking the external 'performance' skill via 'npx skills add', the agent is configured to execute logic and tools defined in a remote repository. This creates a vector for executing third-party code within the agent's context.
[COMMAND_EXECUTION]: The performance loop workflow requires the agent to execute shell commands to build and serve the application locally (e.g., 'npm run build', 'npm run start', 'vite build', 'vite preview'). While these are standard development tasks, they involve direct shell interaction with the user's project environment.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection by processing external data from performance traces and diagnostic insights (e.g., 'performance_analyze_insight').
Ingestion points: Performance traces recorded via 'performance_start_trace' and analysis results from tools like Lighthouse or Chrome DevTools MCP.
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore instructions embedded within the analyzed performance data.
Capability inventory: The agent has the capability to execute shell commands (build/start) and filesystem operations based on the analysis of these traces.
Sanitization: There is no evidence of sanitization or validation of the trace data before it is used to inform the agent's next optimization steps.

react