The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides a template for generating and executing JavaScript code at runtime to automate browser tasks via Playwright. It involves creating a script to navigate to URLs, evaluate font loading, and capture screenshots.
[COMMAND_EXECUTION]: The agent is instructed to run shell commands to start local development servers (e.g., npm run dev, pnpm dev) and monitor console output for active ports.
[INDIRECT_PROMPT_INJECTION]: The skill processes external project files, manifests, and rendered UI components. It explicitly directs the agent to "read the screenshot back," which creates a vulnerability where instructions embedded in the rendered UI or source artifacts could influence the agent's logic.
Ingestion points: Reads .styleseed/bundles/<artifact-id>.md, project manifests, and visual screenshots of components.
Boundary markers: None identified; the agent evaluates the rendered content directly against visual rules without sanitization or boundary delimitations.
Capability inventory: The agent is granted Bash, Write, and Edit permissions, enabling it to modify code and execute shell commands based on findings from processed data.
Sanitization: There is no mention of sanitizing or escaping content retrieved from the rendered UI before it is processed by the agent.

ss-verify