The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local shell commands including node, npm, and npx to execute its internal scripts (inspect-threejs-canvas.mjs) and run Playwright test suites. These are standard operations for a development and QA toolset.
[EXTERNAL_DOWNLOADS]: The skill recommends installing dependencies and browser binaries using npm install and npx playwright install chromium. These downloads originate from well-known, trusted package registries and browser providers.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it processes data from external sources.
Ingestion points: The scripts/inspect-threejs-canvas.mjs script reads information from the global window object of the web page under test, specifically __THREE_GAME_DIAGNOSTICS__ and __THREE_GAME_TEST_HOOKS__.
Boundary markers: There are no explicit markers used to separate the content of the tested web page from the agent's instructional context.
Capability inventory: The agent can perform filesystem operations (writing to artifacts/) and execute shell commands (npm, node).
Sanitization: The data retrieved from the web page is passed to the agent without validation or escaping, potentially allowing a malicious game site to include instructions in its diagnostic output.

threejs-qa-release