--- name: arena description: "Run several candidates at one task, pick a base, graft in the best of the rest, and return one artifact. Use for /arena, \"경쟁시켜줘\", \"여러 버전 만들어서 비교해줘\", or a design where one attempt would lock in the wrong shape. Split-up work ending in a report is swarm." --- # Arena Fan out N parallel attempts at the same task. Read every candidate end to end. Pick the strongest as the base. Graft the best ideas from the others into it. Verify the synthesized result. A **** principle skill below is `references/principle-.md` in the **principles** skill. ## Phase A: Frame The N candidates will receive the same prompt, so the prompt is the contract. 1. State the artifact each candidate is producing. 2. Derive the rubric. State what success looks like for *this* task, then turn it into 3-6 concrete gradeable criteria. The rubric is the picker's tool in Phase D. Candidates only see the task. 3. Pick the runners. Default to one per row; a caller such as **architect** may pass its own set. Spawn more when the arena covers multiple design directions. Use the same model N times when the work is generation-bound rather than judgment-sensitive. | Runner | How to run it | |--------|---------------| | Claude (opus) | `Agent` tool, `subagent_type: "general-purpose"`, `model: "opus"`, `run_in_background: true` | | Claude (fable) | Same, with `model: "fable"` | | Codex | `node /scripts/codex-companion.mjs task --background "$(cat )"`, the configured default model; add `--model gpt-6-astra --effort medium` only for the hardest design question in the work. Then wait with `status --wait --timeout-ms 1800000` as a Bash call with `run_in_background`, which wakes you when the job ends, and read `result `. Ending the turn while that call runs is how you wait; ending it with nothing running in the background ends an unattended run for good. Without `--write` it is read-only, so it returns the artifact in its result and you save it. Find the script with `ls ~/.claude/plugins/cache/openai-codex/codex/*/scripts/codex-companion.mjs`. | Run one Codex job at a time; two concurrent jobs kill each other. If a model name is rejected, use the closest available tier of the same family and say which one ran. 4. Assign output paths. Each candidate writes to its own location (a git worktree through `isolation: "worktree"` where the artifact lives in the repo, otherwise its own folder in the scratchpad), per the **separate-before-serializing-shared-state** principle skill. ## Phase B: Fan out Launch every candidate in one message, each with the task, the path to the shared grounding, its own output path, and instructions to produce both the artifact and a short rationale. Each rationale names the alternatives the candidate considered and what it rejected. If a candidate fails to produce output, proceed with N-1 and note the dropout in the synthesis record. ## Phase C: Cross-judge After every Phase B candidate completes, spawn one read-only judge from a model family other than yours: Codex `task` when you are Claude. It sees the rubric and the candidates by path label, scores each criterion, and recommends a base with rationale, while you do your own reading in Phase D. ## Phase D: Pick a base Read every candidate end to end before picking. Score each candidate against the rubric criterion by criterion, not on holistic feel. Compare against the cross-judge. Agreement on the base confirms the pick. Disagreement means one of you is biased or the rubric was ambiguous. Read both rationales before deciding. Pick the base on which candidate a future maintainer can extend most easily without breaking invariants. Prefer the cleaner boundary or smaller API when two feel tied, per the **laziness-protocol** principle skill. Record the pick and the reason in a short synthesis note alongside the base artifact, including the cross-judge's verdict. ## Phase E: Graft Walk each losing candidate once more and identify what is worth porting into the base. The signal is usually one or two things per candidate, not most of it. Fold each graft in by hand, per the **redesign-from-first-principles** principle skill. Don't paste mechanically. The result has to remain coherent under one mental model. Record what was grafted, from which candidate, and what was rejected and why. When N candidates converge on the same shape, that is a strong agreement signal. Note the convergence in the record and ship the consensus shape. No graft is needed. When N candidates wildly diverge, Phase A was under-specified. Reframe and re-run rather than averaging the divergence. ## Phase F: Verify The synthesized artifact has to hold up under the same scrutiny as any other output, per the **prove-it-works** principle skill. If verification surfaces a problem the arena did not catch, either Phase A was wrong (re-frame and re-run) or one candidate caught it and you missed the graft (go back to Phase E). Don't paper over. ## Outputs One synthesized artifact. One short synthesis note alongside, naming the base, the grafts (with source candidate), the rejections, the dropouts if any, and the verification result.