skills/arena/SKILL.md
Spawn N parallel candidates at the same task, pick a base, and graft the strongest parts of the losers into it. Use for arena this, throw it in the arena, or when one attempt at a non-trivial artifact would lock in the wrong shape.
npx skillsauth add shipshitdev/library arenaInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
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.
Companion to swarm, which covers different slices or races and
returns a report. Arena synthesizes one artifact.
Inputs:
Outputs:
Creates/Modifies:
.tmp/arena-<slug>/candidate-<n>/External Side Effects:
Confirmation Required:
Delegates To:
worktree when candidates need isolated checkoutsverification-before-completion before claiming the result worksOpen a todo list: Frame, Fan out, Cross-judge, Pick, Graft, Verify.
State the artifact. Derive a rubric of 3-6 concrete gradeable criteria. Candidates see the task, not the rubric.
Pick runners across capability tiers and, when possible, different families. Same-tier N times only when the work is generation-bound.
Assign each candidate its own output path. Shared writes fail Separate Before Serializing Shared State.
Spawn all N subagents in one message. Each produces the artifact and a rationale naming alternatives it rejected. A dropout proceeds as N-1.
After candidates complete, spawn one read-only judge on a different tier or family from the parent. It scores each criterion and recommends a base.
Read every candidate end to end. Score criterion by criterion. Compare against the cross-judge. Prefer the cleaner boundary when two feel tied.
Walk each loser once. Port one or two things, by hand, so the result stays coherent. Record rejections. Wild divergence means Phase A was under-specified. Reframe and re-run.
The synthesized artifact holds up under the same scrutiny as any other output. A miss here is a re-frame or a missed graft, not a paper-over.
development
Coordinates a weekly engineering review of board accuracy, recent code changes, operational health, and scoped cleanup. Use for a recurring repository health review or a review of the last several days.
testing
Audits project board configuration and prepares explicitly requested setup, copy, or normalization changes while preserving the existing workflow and provider boundaries. Use when inspecting a board's fields, columns, scope, or configuration.
testing
Reconciles a project board with current work and delivery evidence, reports incomplete coverage and metadata gaps, and applies only approved provider-supported field changes. Use when auditing board drift, reviewing blocked work, or assessing upcoming delivery.
development
Walk through how a subsystem works. Use for "how does X work", code walkthroughs before changing something, and placement or ownership questions. Explains architecture, runtime flow, and onboarding mental models. Can critique architecture. Use why for motivation.