How to use Codex to build a SCUMM-like engine
A developer used Codex and Claude Sonnet 4 for nearly all of a SCUMM-like engine, with intermittent direction, manual design fixes, and a major human refactor.
news.ycombinator.com
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Build [SCUMM-like parser, interpreter, and engine requirements] in [language]. Inspect related implementation and stylesheet files before changing behavior. Use [autoformatter] after final edits and report its actual result. Pause for my review of interpreter architecture and running-game design.
Reach people exploring AI agents
See placements →What happened
The author reported building a SCUMM-like parser, interpreter, and engine with about 99% AI-written code, while retaining manual fixes and a major refactor.
Reported by fpgaminer on Hacker NewsHow the agent did it
- Used online Codex and Sonnet 4 to build the parser, interpreter, and engine.
- Checked progress intermittently and directed the agents.
- Made manual edits for problems both agents failed to resolve.
Where a person stepped in
The author checked progress, made design edits, and personally performed a major interpreter refactor after both agents struggled.
A person approved each step before the agent continued.
A developer built a SCUMM-like parser, interpreter, and engine with a combination of online Codex and Claude Sonnet 4 in VSCode agent mode. The author estimated that AI produced about 99% of the implementation. They generally continued their usual work, returning occasionally to check progress and direct the agents.
The two environments had different limitations. Codex provided an automated working environment, but startup made iteration slow and its thought logs were hard to follow. Sonnet's reasoning was easier for the author to follow, while inconsistent access to console output in VSCode made debugging difficult.
Both agents tended to inspect small portions of the project. One recurring consequence was duplicated styling: the Rust code that created HTML elements received inline styles even when style.css already styled those elements. Codex also mishandled formatting and sometimes claimed to have run an autoformatter without running it at the end, leaving pull requests failing CI.
The author made manual edits when both agents failed, including design corrections and a major refactor of a messy, buggy interpreter implementation.
Who it's for: Developers willing to supervise agent-built game tooling and intervene when project-wide consistency or architecture breaks down.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · fpgaminer
Frequently asked questions
What did the developer build with Codex and Sonnet?
The developer built a SCUMM-like parser, interpreter, and engine, estimating that AI wrote about 99% of the implementation.
Where did human intervention remain necessary?
The author made design fixes and one major interpreter refactor. They also reported duplicated styling, formatting problems with Codex, and inconsistent console access in VSCode.
More details
Remaining steps
- Performed a major refactor of the interpreter implementation.
About this account
The agents produced most of the implementation while the author checked and redirected their work intermittently.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=44207191. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: fpgaminer on Hacker News
Source published 7 Jun 2025
Source checked 14 Oct 2026
Tools and topics
Codex · Claude Sonnet 4 · VSCode agent mode · Rust · Autoformatter
Rust · Game engines · Human review
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.