The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/inspect-threejs-canvas.mjs dynamically resolves and imports the @playwright/test and pngjs modules from the local project's file system using require.resolve and import(). Additionally, it uses page.evaluate to execute code within the browser context to interact with game test hooks.
[COMMAND_EXECUTION]: The instructions guide the agent to perform shell operations, including executing the inspection script with node and running npx playwright test to facilitate playtests and visual regression checks.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external game URLs and browser contexts (console logs and global variables). This creates a vulnerability surface where malicious game content could attempt to influence the agent's behavior or reports.
Ingestion points: Browser console output, page errors, and the window.__THREE_GAME_TEST_HOOKS__ object accessed in scripts/inspect-threejs-canvas.mjs.
Boundary markers: No specific delimiters or safety instructions are used when interpolating browser-originating data into logs or reports.
Capability inventory: Shell command execution (Node.js) and file system write access for saving screenshots and JSON reports.
Sanitization: The script implements regex-based validation for the --state and --run-id command-line arguments to ensure they are safe identifiers.

threejs-qa-release