AGENT SKILL

arena

Spawn N parallel candidates at the same task, pick a base, graft the strongest parts of the losers into it. Use for /arena, 'arena this', 'throw it in the arena', or when one attempt at a non-trivial artifact would lock in the wrong shape.

License 许可
MIT
最近更新
2026-11-09
Stars
★ 574
Forks
90

01 前言 Front Matter

name
arena
description
Spawn N parallel candidates at the same task, pick a base, graft the strongest parts of the losers into it. Use for /arena, 'arena this', 'throw it in the arena', or when one attempt at a non-trivial artifact would lock in the wrong shape.

03 SKILL.md 正文

Arena

Dispatch contract. Read provider-dispatch.md before fan-out. Configured values are provider-qualified descriptors, not host-native model slugs. The parent starts native and external lanes directly; children never route themselves. On Codex, resolve the remaining Claude tool names via codex-tools.md.

Open a todolist with one entry per phase (Frame, Fan out, Cross-judge, Pick, Graft, Verify) before launching anything.

Phase A: Frame

Every candidate gets the same prompt, so the prompt is the contract.

  1. State the artifact each candidate is producing.
  2. Derive the rubric, 3-6 concrete gradeable criteria for what success looks like on this task. It is the picker's tool in Phase D. Candidates only see the task.
  3. Pick the runners. Use arena runners from the current harness's pstack model sheet when present. Otherwise use its role default, one runner per entry. Spawn more runners when the arena covers multiple design directions. Use the same descriptor N times when the work is generation-bound rather than judgment-sensitive. Pick each subagent's effort for the job, per the Subagents section of the poteto-mode skill.
  4. Give each candidate its own output location (a git worktree where possible, otherwise /tmp/arena-<slug>/candidate-<n>/), per the separate-before-serializing-shared-state principle skill.

Phase B: Fan out

Start all N lanes in one fan-out phase through the provider-dispatch contract. Native lanes are background subagents. External lanes are direct background launcher processes with retained task/session handles, never foreground calls and never subagents supervising subprocesses. Give every lane the task, the path to the shared grounding, its own output path, and instructions to produce the artifact and a short rationale naming the alternatives it considered and what it rejected.

An external lane counts only when its receipt says complete and carries either a matching provider-report or Codex's exact pinned-argv evidence; a native lane counts when its tool transcript returns the assigned model's result. If a candidate fails, proceed with N-1 and note the exact dropout in the synthesis record. Never replace it with another provider silently.

Phase C: Cross-judge

After all Phase B candidates complete, choose the judge from arena cross-judge pool in the current harness's pstack model sheet when present, otherwise from the runner defaults above. The judge is never from the parent's provider. Walk the pool in order and take the first entry from another provider that runs. An auto or inherit-parent entry is the parent model, so it never judges. An entry that does not run is a dropout: do not try a substitute slug or the default pool for it. If no entry runs, skip the judge, pick the base alone in Phase D, and say so in the output. Dispatch one read-only judge through the provider contract. It sees the rubric and the candidates by path label, scores each criterion, and recommends a base with rationale. It runs in parallel with your reading in Phase D. Don't dispatch it while candidates are still writing.

Phase D: Pick a base

Read every candidate end to end before picking. Score each against the rubric criterion by criterion, not on holistic feel, and compare with the cross-judge when Phase C ran one. 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 a future maintainer can extend most easily without breaking invariants. When two feel tied, prefer the cleaner boundary or smaller API, per the Laziness Protocol.

Phase E: Graft

Walk each losing candidate once more for what is worth porting into the base, 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, so the result stays coherent under one mental model. Don't paste mechanically.

If the candidates converge on one shape, that is strong agreement. Note it and ship the consensus shape. No graft is needed. If they wildly diverge, Phase A was under-specified. Reframe and re-run rather than averaging the divergence.

Phase F: Verify

Verify the synthesized artifact like any other output, per the prove-it-works principle skill. If verification finds a problem the arena missed, either Phase A was wrong (re-frame and re-run) or a candidate caught it and you missed the graft (go back to Phase E). Don't paper over it.

Outputs

One synthesized artifact, with a short synthesis note alongside naming the base and why, the cross-judge's verdict or that Phase C skipped the judge, each graft with its source candidate, what was rejected and why, any convergence or dropouts, and the verification result.

04 同仓库更多 Skills