Two Claude Code agents formed an RPG party and found co-op bugs the solo dev missed
tschilpi · reddit · 2026-10-26
Solo developer tschilpi spent three years building Lore & Legends Maker, a browser RPG with an AI Dungeon Master. He let each Claude Code session launch its own backend, database, Redis and frontend, open the game and play—two agents then partied up in co-op and found party bugs he'd missed testing alone.
Workflow highlights:
- Visual feedback loops: dedicated lab pages for spells, cards, maps and voices; Claude builds an effect, captures frames with Playwright, inspects and retries before human review
- Agents operating production tools: Claude runs Midjourney in-browser with a fixed style reference, traces clickable areas onto painted scenes, works with ElevenLabs for audio
- Using agents to investigate real players: one player haggled with a merchant for 25 minutes and claimed to move "like the shadows"—the merchant accused him of drawing steel while he held a stick. Successful builds never surface such bugs; feeding session logs to Claude reveals narration/state mismatches
- Decision notes: e.g. "never route on word lists over player text, add a field to the compiler instead"
Core lesson: give agents something concrete to look at—isolated environments, visual labs, session replay, funnel analytics—and let them actually use what you build.
More from coding & agent
- Claude Code rebuilt its own ComfyUI pipeline to fix prop continuity in a 2-min AI film — Odd_Definition_9724 · 2026-10-26
- Claude Code audit finds 214,856 "permanently deleted" photos lurking in Android hidden cache — Loo_Atreides · 2026-10-26
- Someone sent the first customer fax via ChatGPT over Onetime Fax's new MCP server — tdhopper · 2026-10-26
- China Telecom's XingChen lab open-sources Xing4.0-29B-Intent, a 29B MoE with 4B activated params — FellMentKE · 2026-10-26
- Codebases sized to fit one context window dramatically improve agent outcomes — julianharris · 2026-10-26
- 42-Minute Video Breaks Down Why AI Agents Fail: Mostly Not the Model's Fault — JohnSavill · 2026-10-26