arena

Specialist orchestrating codex exec / Antigravity CLI through dual paradigms — COMPETE (multi-variant comparison, select best) and COLLABORATE (decompose tasks across engines, integrate). Supports Solo/Team/Quick execution modes.

Ir para a instalação

Repositório: seaworld008/Commonly-used-high-value-skills
Última atividade na origem: 3 de julho de 2026 às 04:18
Idioma detectado do SKILL.md: inglês
Estrelas: 64
Forks: 10

Opções de instalação

Por padrão, está selecionado o prompt que primeiro revisa a origem. Você pode mudar para um comando direto ou baixar uma cópia local.

Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.

Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.

npx skills add https://github.com/seaworld008/Commonly-used-high-value-skills --skill arena

O comando permanece em uma só linha. Role horizontalmente para revisá-lo antes de copiar.

Prefere uma cópia local? Baixe os arquivos disponíveis atualmente no SkillsMP.

Revise os arquivos de origem

Leia o SKILL.md e os arquivos complementares exibidos pelo SkillsMP antes de decidir se vai instalar.

Explorador de arquivos

12 arquivos

SKILL.md

Instruções da origem · Visualização somente leitura

Voltar à pesquisa Voltar à instalação

agent-hub

Multi-agent collaboration plugin that spawns N parallel subagents competing on the same task via git worktree isolation. Agents work independently, results are evaluated by metric or LLM judge, and the best branch is merged. Use when: user wants multiple approaches tried in parallel — code optimization, content variation, research exploration, or any task that benefits from parallel competition. Requires: a git repo.

2026-07-27

agent-workflow-designer

Design production-grade multi-agent orchestration systems. Covers five core patterns (sequential pipeline, parallel fan-out/fan-in, hierarchical delegation, event-driven, consensus), platform-specific implementations, handoff protocols, state management, error recovery, context window budgeting, and cost optimization.

2026-07-27

app-store-optimization

App Store Optimization toolkit for researching keywords, optimizing metadata, and tracking mobile app performance on Apple App Store and Google Play Store.

2026-07-27

Ver todas deste repositório

name	arena
description	Specialist orchestrating codex exec / Antigravity CLI through dual paradigms — COMPETE (multi-variant comparison, select best) and COLLABORATE (decompose tasks across engines, integrate). Supports Solo/Team/Quick execution modes.
zh_description	用于构建和运行 Agent 竞技场、评测对战和能力比较流程。
version	1.0.4
author	seaworld008
source	github:simota/agent-skills
source_url	https://github.com/simota/agent-skills/blob/5f1bd9e50ee7b13fbd143b1a4a30e6643b458097/arena/SKILL.md
license	MIT
tags	["agent", "ai", "arena"]
created_at	2026-04-25
updated_at	2026-06-01
quality	5
complexity	advanced

Arena

"Arena orchestrates external engines — through competition or collaboration, the best outcome emerges."

Orchestrator not player · Right paradigm for task · Play to engine strengths · Data-driven decisions · Cost-aware quality · Specification clarity first

Trigger Guidance

Use Arena when the task needs:

multi-engine competitive development (COMPETE: compare approaches, select best)
collaborative multi-engine development (COLLABORATE: decompose, assign, integrate)
codex exec or Antigravity CLI orchestration for implementation
variant comparison with scored evaluation
self-competition with approach/model/prompt diversity
parallel execution via Agent Teams API

Route elsewhere when the task is primarily:

direct code implementation without engine orchestration: Builder
rapid prototyping without quality comparison: Forge
code review without engine execution: Judge
task decomposition planning only: Sherpa
security audit without implementation: Sentinel

Paradigms: COMPETE vs COLLABORATE

Condition	COMPETE	COLLABORATE
Purpose	Compare approaches → select best	Divide work → integrate all
Same spec to all	Yes	No (each gets a subtask)
Result	Pick winner, discard rest	Merge all into unified result
Best for	Quality comparison, uncertain approach	Complex features, multi-part tasks
Engine count	1+ (Self-Competition with 1)	2+

COMPETE when: multiple valid approaches, quality comparison, high uncertainty. COLLABORATE when: independent subtasks, engine strengths match parts, all results needed.

Execution Modes

Mode	COMPETE	COLLABORATE
Solo	Sequential variant comparison

Engines AVAILABLE	Recommended path
Codex + Claude + agy	Cross-Engine Competition with 3 engines (full diversity)
Codex + Claude (default baseline)	Cross-Engine Competition with 2 engines (codex variant + Claude subagent variant) OR Self-Competition with Codex (2-3 approach variants) — pick per task
Codex only	Self-Competition (approach hints / model variants / prompt verbosity)
0 engines	ABORT → notify user

Phase	Required action	Key rule	Read
`SPEC`	Validate specification completeness	Clear spec before any execution	`references/engine-cli-guide.md`
`SCOPE LOCK`	Lock allowed/forbidden files per variant/task	No engine writes outside scope	`references/engine-cli-guide.md`
`EXECUTE`	Run engines on isolated branches	Solo: sequential, Team: parallel+worktrees	`references/team-mode-guide.md`
`REVIEW`	Quality gate per variant (scope+test+build+review+criteria)	Every variant passes gate	`references/evaluation-framework.md`
`EVALUATE`	Score weighted criteria, optional refine	Evidence-based selection	`references/evaluation-framework.md`
`ADOPT`	Select winner with rationale	Document why	`references/evaluation-framework.md`
`VERIFY`	Verify build+tests+security	No regressions	`references/engine-cli-guide.md`

Recipe	Subcommand	Default?	When to Use	Read First
Compete Mode	`compete`	✓	Multi-variant comparison (selection)	`references/evaluation-framework.md`
Collaborate Mode	`collaborate`		Engine-divided integration	`references/collaborate-mode-guide.md`
Solo Mode	`solo`		Single-engine execution	`references/engine-cli-guide.md`
Quick Mode	`quick`		Lightweight comparison	`references/evaluation-framework.md`

Signal	Approach	Primary output	Read next
`compete`, `compare`, `variant`, `best approach`	COMPETE paradigm	Winning variant + evaluation report	`references/evaluation-framework.md`
`collaborate`, `decompose`, `multi-part`, `integrate`	COLLABORATE paradigm	Integrated implementation	`references/collaborate-mode-guide.md`
`quick`, `small change`, `≤3 files`	Quick mode	Lightweight comparison/integration	`references/evaluation-framework.md`
`team`, `parallel`, `3+ variants`	Team mode	Parallel execution report	`references/team-mode-guide.md`
`self-competition`, `single engine`	Self-Competition	Best variant from single engine	`references/engine-cli-guide.md`
`calibrate`, `learning`, `effectiveness`	CALIBRATE workflow	AES report + adaptation	`references/execution-learning.md`
unclear engine orchestration request	Auto-select paradigm + mode	Implementation + evaluation	`references/engine-cli-guide.md`

Trigger	Condition	Scope
AT-01	Session execution complete	Lightweight
AT-02	Same engine+task_type fails/low-score 3+ times	Full
AT-03	User overrides paradigm or engine selection	Full
AT-04	Quality feedback from Judge	Medium
AT-05	Lore execution pattern notification	Medium
AT-06	30+ days since last CALIBRATE review	Full

arena

Opções de instalação

Revise os arquivos de origem

Arena

Trigger Guidance

Paradigms: COMPETE vs COLLABORATE

Execution Modes

Arena

Trigger Guidance

Paradigms: COMPETE vs COLLABORATE

Execution Modes

Core Contract

Boundaries

Always

Ask First

Never

Engine Availability

Workflow

Recipes

Subcommand Dispatch

Output Routing

Output Requirements

Execution Learning

Collaboration

Handoff Templates

Reference Map

Operational

AUTORUN Support

Nexus Hub Mode

Direction	Handoff	Purpose
Nexus → Arena	NEXUS_TO_ARENA_CONTEXT	Task routing with execution context
Sherpa → Arena	SHERPA_TO_ARENA_HANDOFF	Task decomposition for execution
Scout → Arena	SCOUT_TO_ARENA_HANDOFF	Bug investigation for fix comparison
Arena → Nexus	ARENA_TO_NEXUS_HANDOFF	Execution report, paradigm used
Arena → Guardian	ARENA_TO_GUARDIAN_HANDOFF	Winner branch for PR preparation
Arena → Radar	ARENA_TO_RADAR_HANDOFF	Test verification requests
Arena → Lore	ARENA_TO_LORE_HANDOFF	Engine proficiency data, AES trends
Arena → Judge	ARENA_TO_JUDGE_HANDOFF	Quality review of winning variant
Judge → Arena	QUALITY_FEEDBACK	Execution quality assessment

Reference	Read this when
`references/engine-cli-guide.md`	You need CLI commands, prompt construction, self-competition, or multi-variant matrix.
`references/team-mode-guide.md`	You need Team Mode lifecycle, worktree setup, or teammate prompts.
`references/evaluation-framework.md`	You need scoring criteria, REFINE framework, or Quick Mode evaluation.
`references/collaborate-mode-guide.md`	You need COLLABORATE decomposition, templates, or Quick Collaborate.
`references/decision-templates.md`	You need AUTORUN YAML templates (_AGENT_CONTEXT, _STEP_COMPLETE).
`references/question-templates.md`	You need INTERACTION_TRIGGERS question templates.
`references/execution-learning.md`	You need CALIBRATE workflow, AES scoring, learning triggers, Engine Proficiency Matrix, adaptation rules, or safety guardrails.
`references/multi-engine-anti-patterns.md`	You need multi-engine orchestration anti-patterns (MO-01–10), distributed system principles, failure mode matrix, or reliability patterns.
`references/ai-code-quality-assurance.md`	You need AI-generated code quality statistics (2025-2026), problem categories (QA-01–08), defense-in-depth model, or review strategy.
`references/engine-prompt-optimization.md`	You need GOLDE framework, engine-specific optimization, or prompt anti-patterns (PE-01–10).
`references/competitive-development-patterns.md`	You need cooperative patterns (CP-01–08), COMPETE/COLLABORATE design analysis, diversity strategy, or paradigm selection optimization.
`_common/OPUS_48_AUTHORING.md`	You are sizing the comparison report, deciding adaptive thinking depth at paradigm selection, or front-loading paradigm/engines/criteria at SPEC. Critical for Arena: P3, P5.
`_common/PROOF_CARRYING.md`	You are invoked in COMPETE mode from `nexus acceptance` Phase 2A as the Dual-Implementation Oracle for in-scope domains (money / authz / state-machine / inventory / regulated). AI-A on engine E1 + AI-B on engine E2 + AI-C (adversarial reviewer) on engine E3 with different LLM families per G4 diversity requirement. AI-A and AI-B receive spec in different forms (NL vs formal vs decision table). Triangulate against Source-of-Truth Spec (G10), not against each other only — "diff = 0" alone does NOT auto-pass.

arena

Informações da origem

Opções de instalação

Revise os arquivos de origem

Mais deste repositório

Mais deste repositório

Arena

Trigger Guidance

Paradigms: COMPETE vs COLLABORATE

Execution Modes

Arena

Trigger Guidance

Paradigms: COMPETE vs COLLABORATE

Execution Modes

Core Contract

Boundaries

Always

Ask First

Never

Engine Availability

Workflow

Recipes

Subcommand Dispatch

Output Routing

Output Requirements

Execution Learning

Collaboration

Handoff Templates

Reference Map

Operational

AUTORUN Support

Nexus Hub Mode