Arena

Other keel skills named here sit beside this one at ${CLAUDE_SKILL_DIR}/../<name>/SKILL.md, and a principle at principle-<name>. The Skill tool refuses them, so read the file and follow it.

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 the arena runners line. Read ${CLAUDE_CONFIG_DIR:-$HOME/.claude}/keel-models.md if it exists; use the role's line, else the default, one runner each on opus, fable, and sonnet. Three different Claude models give the panel its diversity. An inherit, inherit-parent, or auto entry in this line means the parent model, so omit model for it. If the Agent tool rejects a configured entry, run that seat on the default at the same position and say so. If it rejects a default, omit model for that seat and say so. Spawn more runners when the arena covers multiple design directions. Use the same model 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 lead skill.
4. Give each candidate its own output location (a git worktree where possible, via isolation: "worktree" on its Agent call, otherwise /tmp/arena-<slug>/candidate-<n>/), per the separate-before-serializing-shared-state principle skill.

Phase B: Fan out

Spawn all N subagents in one message, each with 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.

Subagents run in the background and the parent is notified as each completes, so never poll or sleep-wait. If a candidate produces no output, proceed with N-1 and note the dropout.

Phase C: Cross-judge

After all Phase B candidates complete, choose one model from the arena cross-judge pool line. Read ${CLAUDE_CONFIG_DIR:-$HOME/.claude}/keel-models.md if it exists; use the role's line, else the default pool of opus, fable, and sonnet. Spawn one read-only judge subagent, its prompt opening with "Read-only: do not edit files." The judge is never on the parent's model. Try only the entries that name a different model from the parent's, in pool order. An inherit, inherit-parent, or auto entry is the parent model, so it never judges. If the Agent tool rejects an entry, move to the next one. If no entry runs, try the default pool's entries other than the parent's model the same way. If none of those runs, skip the judge, pick the base alone in Phase D, and say so. If you do not know the parent's model, try the pool's entries in order and say that the parent-model check did not run. The judge 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 spawn 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.