do-competitively — Claude Code skill: competitive generation
Runs three agents on the same task in parallel, scores their output with judges, then synthesizes the best result. Built for work where quality beats speed.
npx skills add NeoLabHQ/context-engineering-kit --skill do-competitively -a claude-codeWhat do-competitively does
The skill follows the Generate-Critique-Synthesize pattern. Three generator agents each solve the same task independently, running a draft-critique-revise loop. A meta-judge runs alongside them: it reads the codebase before the generators touch it and writes a YAML evaluation spec with rubrics and scoring criteria.
Three judges then score the solutions against that spec, each verifying and revising its own report. An adaptive step picks the strategy: polish a clear winner, synthesize a split decision, or redesign when every solution scores low. Reports land in .specs/reports, while the solutions stay in their output locations.
The orchestrating agent deliberately does not read the generators' or judges' files, so its context stays small. Generators are barred from mutating git commands unless the task explicitly allows them. Expect this to cost more than a single agent run, since seven agents work on one task.
What it helps with
- get several independent solutions to one task and pick the best
- compare implementation approaches against shared criteria
- merge the strong parts of competing solutions into one
- rerun generation when every attempt comes out weak
- keep evaluation reports separate from the solutions themselves
Example prompts for the agent
- do-competitively: design the API schema for orders
- do-competitively: write the auth module, three variants to compare
- do-competitively: propose an architecture for the notification service
How to install do-competitively
1Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-competitively -a claude-code2Manually: put the files in place
~/.claude/skills/do-competitively/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/do-competitively/
1Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-competitively -a codex2Manually: put the files in place
.agents/skills/do-competitively/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-competitively -a cursor2Manually: put the files in place
.cursor/skills/do-competitively/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install NeoLabHQ/context-engineering-kit do-competitively2Manually: put the files in place
.github/skills/do-competitively/SKILL.md1Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-competitively -a gemini-cli2Platform CLI
gemini skills install https://github.com/NeoLabHQ/context-engineering-kit --path skills/do-competitivelyManually: put the files in place
.devin/skills/do-competitively/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add NeoLabHQ/context-engineering-kit --skill do-competitively -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about do-competitively
How do I install do-competitively in Claude?
Run npx skills add NeoLabHQ/context-engineering-kit --skill do-competitively -a claude-code. The skill file then sits at ~/.claude/skills/do-competitively/SKILL.md.
What does do-competitively do?
It launches three generator agents on one task, has three judges score their solutions against a meta-judge spec, and assembles the final result using the chosen strategy.
How is do-competitively different from a normal agent run?
Seven agents work on the task: three generators, a meta-judge and three judges. You get independent solutions and comparative scoring, at a higher cost.