The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes project-specific build and development scripts including build-storybook, npm start, and npm dev to render UI states.
[COMMAND_EXECUTION]: It invokes a local script scripts/screenshot-states.mjs and manages headless Chromium/Playwright processes to capture UI snapshots.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted ticket acceptance criteria and user specifications verbatim into the judging agent's context.
Ingestion points: Project AGENTS.md and user-provided ticket/specification text in SKILL.md.
Boundary markers: None identified; instructions are passed directly to the secondary agent.
Capability inventory: Subprocess execution (npm), file system writes (committing to repository), and agent spawning.
Sanitization: No explicit escaping or validation of external specification content is mentioned.
[DATA_EXFILTRATION]: Automatically commits binary screenshot data to the repository under docs/pr-screenshots/ and generates raw GitHub URLs for embedding in pull requests.

visual-verify