How to use Claude to rebuild a rotation-battle prototype
Claude rebuilt a Godot rotation-battle prototype after its creator found it had used chat notes instead of specs, exposing a polished, incorrect first screen.
note.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Build this Godot rotation-battle prototype from the three specification files at these exact paths. Treat the Gemini chat archive as nonauthoritative background. Summarize ally counts, enemy counts, rotation/stay controls, and stat rules before implementation. Let me inspect the first battle screen, then repair any mismatch against the approved specification.
Reach people exploring AI agents
See placements →What happened
A rebuilt Godot battle prototype reflected the intended ally/enemy structure and rotation rules after the owner corrected the agent’s source-context mistake.
Reported by たまごやきゲームス (@tamagoyaki42dev) — noteHow the agent did it
- The creator wrote three battle specification files after preliminary design conversations.
- Human play-screen review found missing rotation controls and discovered Claude had read informal notes instead.
- Claude reread the exact specifications and rebuilt the battle screen for seven allies and one enemy.
Where a person stepped in
The creator designed mechanics, inspected the first screen, challenged the discrepancy, and explicitly selected the authoritative files for the rebuild.
たまごやきゲームス wanted a Godot battle prototype inspired by a rotation-based tactical game. The creator discussed ideas with Gemini, then spent substantial time defining battle rules, enemy data, and ally templates in three specification files before asking Claude Code to implement them.
The first screen looked plausible but violated the intended design. It had extra enemies and defense statistics, while the rotation and stay controls were missing. Human inspection caught the mismatch. When questioned, Claude acknowledged that it had used saved Gemini conversation notes instead of reading the actual specification.
The creator then directed it to the correct files and rebuilt the prototype. The revised battle screen had seven allies and one enemy and reflected the planned rotation mechanic more closely. The post announces a playable battle-screen demonstration and links the public source.
The useful result is both the repaired prototype and a concrete explanation of its initial failure. A polished interface concealed a wrong-context implementation until the owner checked mechanics. The author’s lesson was to separate informal notes from authoritative files and provide exact paths. This is prototype development, not a claim that the whole intended game was completed or released commercially.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: note · たまごやきゲームス (@tamagoyaki42dev)
Frequently asked questions
Why was the first implementation wrong?
Claude used saved chat notes rather than the actual specification files.
How was the mismatch found?
The creator inspected enemy counts, added statistics, and missing rotation controls.
Was the complete game finished?
No. The account announces a battle-screen prototype demonstration.
More details
About this account
The creator wrote three battle specification files after preliminary design conversations. Human play-screen review found missing rotation controls and discovered Claude had read informal notes instead.
Firsthand Japanese creator account; translated version and original URL opened. Gemini’s conversational planning is not mapped as the implementation agent. Prototype result only. Original task wording is paraphrased rather than quoted, so prompt is suggested. Suggested prompt written by Agents In Use; no original task prompt was published in the verified account.
Credit: たまごやきゲームス (@tamagoyaki42dev) — note
Source published 5 Jul 2026
Source checked 14 Oct 2026
Tools and topics
Claude Code · Godot · GDScript · GitHub
specification review · battle systems · wrong-context repair
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.