The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from user-provided URLs and local action files which are then reflected in the agent's context through logs and state descriptions.
Ingestion points: The web_game_playwright_client.js script loads content from external URLs and reads local JSON files to drive browser interactions.
Boundary markers: The skill lacks explicit instructions or delimiters to isolate content retrieved from the web, which could allow malicious instructions embedded in a game's HTML or console logs to influence the agent.
Capability inventory: The agent has the ability to write to the file system (fs.writeFileSync) and execute commands via the Playwright client.
Sanitization: There is no evidence of sanitization or filtering applied to the content retrieved from the browser before it is processed by the agent.
[DYNAMIC_EXECUTION]: The Playwright client utilizes dynamic script injection to facilitate testing hooks within the browser environment.
The script uses page.addInitScript and page.evaluate to inject a makeVirtualTimeShim that overrides standard browser timing APIs like setTimeout and requestAnimationFrame to ensure deterministic execution.
[EXTERNAL_DOWNLOADS]: The documentation instructs the agent to install the @playwright/mcp package globally using npm install if the tool is not already available in the environment.
[COMMAND_EXECUTION]: The skill workflow relies on executing a Node.js script (web_game_playwright_client.js) with various command-line arguments to automate game testing and state capture.

develop-web-game