compete
Runs 3 competing implementations of a feature using different strategies, then picks the best one. Swiss cheese layer 1 — diversity of approaches.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Runs 3 competing implementations of a feature using different strategies, then picks the best one. Swiss cheese layer 1 — diversity of approaches.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
| name | compete |
| description | Runs 3 competing implementations of a feature using different strategies, then picks the best one. Swiss cheese layer 1 — diversity of approaches. |
| user-invocable | true |
| argument-hint | <spec-file-path or feature description> |
You orchestrate a competition between 3 independent implementation attempts. Each uses a different strategy. A neutral judge picks the winner. The loser branches are deleted.
Read the spec or feature description from: $ARGUMENTS
BASE_BRANCH.BASE_BRANCH:
compete/minimal — Strategy Acompete/thorough — Strategy Bcompete/creative — Strategy CFor EACH strategy below, checkout the corresponding branch and use the implementer agent to execute. Pass the strategy directive along with the original spec.
compete/minimal)Implement the spec with the fewest lines of code possible. Prefer reusing existing utilities, composing existing functions, and leveraging framework conventions. No new abstractions. No new files if avoidable. The best code is code you didn't write.
compete/thorough)Implement the spec with comprehensive edge case handling from the start. Validate all inputs. Handle every error path. Add more tests than the spec requires — boundary values, empty inputs, malformed data, concurrent access if applicable. Prefer robustness over brevity.
compete/creative)Implement the spec using an unconventional approach. Look for a different data structure, a different algorithm, a different decomposition than the obvious one. Challenge assumptions in the spec if they constrain a better solution. The goal is a fresh perspective — not cleverness for its own sake.
After each implementation, return to BASE_BRANCH before starting the next.
BASE_BRANCH.{{verify_command}} on each, eliminates failures, and scores the survivors.BASE_BRANCH: git merge {winning-branch}git branch -D compete/minimal
git branch -D compete/thorough
git branch -D compete/creative
{{verify_command}} one final time on the merged result.Print a summary:
## Competition Complete
Winner: [strategy name] (compete/[branch])
Score: [X/25]
### Why it won
[Judge's reasoning]
### Cherry-picked from runners-up
- [Any improvements merged from other branches, or "None"]
### Final verification
[pass/fail]