View the Project on GitHub cse110-sp26-group08/one-arm-slot-machine-2
This is task-specific but having an Agents.md is a given when working with Agents. One should have an template they use for what information to distribute to the agent, but this is very team dependant. Example being, what code structure are you looking for? What is the theme of the app your building? This should go in the Agents.md file. What kind of tests should the agent write? From now on we will use Codex web snce it allows for parrallelization to solve various issues in github simultaneously.
By connecting GitHub repository to codex, we can create detailed issues on github and ask codex to implement them. Codex can simultaneously work on multiple issues using multiple tasks. Once codex finishes the task, we can ask it to create pull request which we can then review and merge the changes to main.
Allows our model to become an expert in one particular thing Full documentation Agent Skills – Codex | OpenAI Developers https://developers.openai.com/codex/skills?utm_source=chatgpt.com
Example, when we have different instances of gpt-5.4 we should give each instance a different skill. For example Frontend or backend Skills should go in SKILLS.md file and will be only used when that specific skill needs to be invoked.
The main thing we should keep in mind for our project is utilizing the ready MCP’s that can allow codex to view the browser, access figma for designing an interesting UI
Important: We should not overwhelm codex with MCP’s as too many mcps just become cumbersome in the long run.
Useful @playwright/mcp - npm playwright mcp, allows the model to look at browsers.
ChromeDevTools/chrome-devtools-mcp: Chrome DevTools for coding agents useful for if our browser is chrome.