The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Electron and Playwright/CDP for background verification of GUI components. This involves executing repository-specific test scripts and managing ephemeral browser processes to capture screenshots and validate layout behavior.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external documents, logs, and file previews through components like DocReader and MarkdownDocument.
Ingestion points: Processes Markdown documents, file previews, logs, and review prose in SKILL.md.
Boundary markers: Not explicitly defined in instructions, though the use of shared primitives for rendering is mandated.
Capability inventory: File system access via an authorized daemon read path and shell execution for automated testing via Playwright.
Sanitization: Explicitly instructs the agent to reuse the shared MarkdownDocument parser rather than mounting independent renderers, which helps maintain a controlled execution environment.
[DATA_EXPOSURE]: The skill provides strong proactive security guidance, specifically prohibiting the agent from saving or restoring the user's system clipboard and ensuring that browser storage or sensitive profile data is never included in logs or task artifacts.

harness-gui