How to use Claude to compare chess-building skills
Claude built three chess prototypes from one prompt; changing installed skills altered rules, layout, and architecture in an informal artifact comparison.
alsade.me
Original prompt
Shared by the source author. See source notes for context.
Build me a playable chess game.
Reach people exploring AI agents
See placements →What happened
Three playable chess artifacts showed distinct layout, feature, and architecture choices under different installed skills; the account remains a small informal comparison.
Reported by Al Sade — Personal blogHow the agent did it
- Sade ran the same chess request with no skill, web-design guidelines, and a frontend-design skill.
- He inspected the resulting code structure, rule features, and interface differences.
- He published all three playable artifacts so readers could compare the actual outputs.
Where a person stepped in
Sade selected each skill configuration, kept the prompt and model constant, inspected the code and visual results, and published the comparison.
Al Sade compared three Claude Code chess builds while keeping the task prompt and model the same. One run had no additional skill, another used Vercel’s web-design guidelines, and the third used Anthropic’s frontend-design skill. The original post links each playable result.
The baseline produced legal-move handling, game-end detection, notation, and undo controls in a straightforward procedural implementation. The guidelines run changed the information layout, replacing the move-history sidebar with captured pieces and adding a fifty-move-rule check. Those were functional differences, not merely alternate colors.
The design-oriented run introduced a class-based game structure, separate engine and interface responsibilities, responsive panels, a promotion dialog, design tokens, and piece animation. Sade inspected the generated implementations and compared the visible artifacts. The article’s line-count descriptions are inconsistent, so this account does not use them as a performance measure.
The useful experiment is the controlled input with inspectable outputs: installed instructions changed architecture, interface, and feature selection. It remains a single author’s comparison. Three attractive browser games do not establish complete chess-rule correctness, general skill superiority, or future multiplayer readiness.
Prompt
Original prompt
Source: Personal blog · Al Sade
Frequently asked questions
What input stayed the same?
The author reports the same task prompt and model across the three runs.
Did only the styling change?
No. Feature selection and procedural versus class-based structure differed.
Is this a complete chess correctness benchmark?
No. The post compares three artifacts, not an exhaustive rules test.
More details
About this account
Sade ran the same chess request with no skill, web-design guidelines, and a frontend-design skill. He inspected the resulting code structure, rule features, and interface differences.
Original prompt preserved verbatim. Opened original comparison and linked playable artifacts where available. Omit contradictory line counts and avoid the author’s unsupported broad superiority claims. Original prompt excerpt.
Credit: Al Sade — Personal blog
Source published 8 Mar 2026
Source checked 14 Oct 2026
Tools and topics
Claude Code · frontend-design skill · web-design-guidelines skill · browser
game prototypes · skill comparison · chess
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.