Simon Willison used a single prompt to have Claude Opus 5.5 add a four-player battle map to a web game
Django co-creator Simon Willison used a single English prompt to have Claude Opus 5.5 add a four-player battle map to his web game and rework the networking code to support four-player multiplayer. The prompt gave only four constraints—four-way rotational symmetry, at most four AI players, building on the existing WebSocket non-audio multiplayer mode, and empty slots filled by AI—without specifying files or lines of code. Based on this, the author argues that prompts should include fewer steps and pin down the boundary conditions.
Why read this
AI score 60
The original article breaks down the four constraints in Simon Willison's single prompt that got Claude to complete the map and multiplayer overhaul, which can serve as a reference for concise prompting.
Topics
Body · AI translation
Many people still write Prompts as long essays of several hundred words, even teaching the AI step by step to first do this and then do that. Django co-author Simon Willison today demonstrated a masterful example: With just one short English prompt, he got Claude Opus 5.5 to create an entirely new four-player battle map for his web game out of thin air, and along the way refactor all the underlying networking code to support four-player online play.
The prompt's original wording was extremely concise: Add a new four-player battle map—four-way rotational symmetry, up to four AI players, must work with the WebSocket-based non-audio online mode, allowing up to four human players to join (empty slots filled by AI).
There was not a single subheading in the whole thing, not a single procedural step, and nothing telling the model which file to look through or which line of code to change.
Yet Claude, like a seasoned senior engineer, found the map rendering function on its own, designed a symmetric compass map called Compass Rose, and then precisely refactored the WebSocket communication layer and the player matchmaking pool.
Why is it that your Prompts so often make large models hallucinate, while a top hacker's one sentence delivers flawlessly?
Because most people, when writing prompts, make one fatal mistake: Trying to use natural language to play the role of a compiler. The more detailed you write and the more granular the steps you give, the more noise there is in the large model's context, which instead locks down its own reasoning path.
The most formidable thing about Simon's sentence is that throughout, he input only four absolutely clear hard boundaries: First, a geometric constraint: explicitly requiring four-way rotational symmetry, which directly pins down the map's mathematical logic. Second, a participant cap: at most four AI and four human players in a mixed battle. Third, a technical reuse protocol: naming the existing WebSocket non-audio module, strictly forbidding the model from inventing a new network architecture on its own. Fourth, edge-case fallback logic: vacant seats are automatically filled by AI.
So the essence of a good prompt has never been teaching the AI how to do the work, but welding shut the guardrails it must not cross.
How to write the steps is the reasoning ability the model excels at; nailing down the boundaries is the business intent that humans should deliver.
The cleaner you write the constraints, the more impeccable the agent's creativity and engineering execution become. Cut the micro-instructions, give sufficient boundary conditions—this is the most valuable communication skill for the new generation of large models.