recursive-arena

A workflow that repeatedly improves an answer using several language models and judge feedback. Each round compares generated answers, then refines the best one.

In plain words
What is it for?
Use it for tasks where multiple model opinions and repeated refinement may improve the final answer.
Why use it?
It reduces the need to manually compare drafts and repeat critique cycles. The workflow keeps the highest-scoring result across rounds.

Skill for Claude CodeCodex

Part of the recursive-reasoning plugin — 3 skills, 3 commands, 2 agents shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/lollipopkit/cc-plugins/recursive-arena
Any agent
npx skills add lollipopkit/cc-plugins --skill recursive-arena
Clone the repo
git clone --depth 1 https://github.com/lollipopkit/cc-plugins

Made for: Claude Code, Codex.

Or install recursive-reasoning, the plugin that ships this one along with the rest of its 3 skills, 3 commands, 2 agents.

Per session 41 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 399 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00041 $0.00399
Opus 5 $0.00020 $0.00199
Sonnet 5 $0.00008 $0.00080
Haiku 4.5 $0.00004 $0.00040

Measured 3d ago against content hash c91a396a1300, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

recursive-arena scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

The scan reads SKILL.md. This mod also ships 1 executable file (scripts/recursive_arena.py), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

plugins/recursive-reasoning/skills/recursive-arena/SKILL.md · 53 lines

What it actually says

Recursive-Arena

Use the orchestrator: python3 "${CLAUDE_PLUGIN_ROOT}/skills/recursive-arena/scripts/recursive_arena.py".

How to run

python3 "${CLAUDE_PLUGIN_ROOT}/skills/recursive-arena/scripts/recursive_arena.py" \
  --prompt "<task>" --iters 4 --arena-iters 3 --json

Common flags: --max-judges, --temperature, --max-tokens, --timeout.

Iteration Loop

For each outer iteration:

  1. Run multi-model to generate a best candidate.
  2. Use judge summaries as critique input.
  3. Refine the prompt with the current best answer.
  4. Keep the global best by score and continue.

Configuration

Reuses multi-model .env configuration:

  • ARENA_MODELS
  • ARENA_OPENAI_BASE_URL or ARENA_PROVIDER_<NAME>_BASE_URL
  • Optional API keys (ARENA_OPENAI_API_KEY, ARENA_PROVIDER_<NAME>_API_KEY)

Optional orchestration env:

  • RLM_ARENA_ARENA_ITERS default inner arena iterations
  • RLM_ARENA_MAX_JUDGES default judge cap

Output and Safety

  • Final answer is the best outer-iteration result.
  • When useful, show a compact evolution table:
    • iteration
    • winner_model_id (numeric ID only)
    • avg_judge_score
    • refinement_applied
  • Never disclose provider/model names.
  • Never print secrets from .env.
Files

What ships with it

1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 53 lines · 41 tokens per session scan A c91a396a1300

Subscribe to this mod's changes

recursive-arena is a skill published in the GitHub repository lollipopkit/cc-plugins (7 stars, last pushed 5mo ago), licensed MIT. It adds 41 tokens to every session and 399 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

metaprompt

Generate a complete, ready-to-use prompt for a target model and harness. Triggers: metaprompt, generate a prompt for, write me a prompt, create a system prompt, prompt engineer this, optimize this prompt.

JairoTorregrosa/jaiskills · 48 tokens

agent-platform-prompt-management

Manages and orchestrates prompts in Agent Platform. Use when you need to create, list, retrieve, version, or delete managed prompts in Agent Platform. Don't use for model training, model deployment to endpoints, or managing non-Agent Platform prompts.

google/skills · 55 tokens

gemini-api-dev

Use this skill when writing code that calls the Gemini API for text generation, multi-turn chat, multimodal understanding, image generation, video generation, streaming responses, background research tasks, function calling, structured output, or migrating from the old generateContent API. Covers SDK usage and best…

google-gemini/gemini-skills · 73 tokens

semantic-kernel

Build AI-enabled .NET applications with Semantic Kernel using services, plugins, prompts, and function-calling patterns that remain testable and maintainable. USE FOR: adding AI-driven prompts, plugins, or orchestration to a .NET app; reviewing kernel construction, service registration, or plugin usage; building…

managedcode/dotnet-skills · 116 tokens

prompt-version-management

Manage prompt versions, run A/B tests across agent prompts, track performance regressions, and safely roll out prompt changes in production. Covers prompt diffing, semantic versioning, canary releases, and automated evaluation.

cosmicstack-labs/mercury-agent-skills · 47 tokens

realistic-prompt-generation

Generate natural, controlled prompt variants from a target-blind design brief and prompt architecture while preserving approved jobs, acts, journeys, constraints, roles, locales, proximity bands, and evidence language. Use after architecture design and before contamination or semantic QA.

elvisun/newsjack · 55 tokens