New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

02 Rl Reward

by agentscope-ai in agentscope-ai/OpenJudge

Build RL reward signals using the OpenJudge framework. An agent skill from agentscope-ai/OpenJudge.

Apache-2.0Auto-check passedAI & LLM Engineering

Install 02 Rl Reward

skills CLI
$ npx skills add agentscope-ai/OpenJudge --skill 02-rl-reward -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install agentscope-ai/OpenJudge 02-rl-reward --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/agentscope-ai/OpenJudge.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/openjudge-core/02-rl-reward .claude/skills/02-rl-reward && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
02-rl-reward
GitHub stars
871
Token cost
~1.6k tokens
SKILL.md length
339 words
Files
3
Skills in repo
19
Repo updated
First seen
Licence
Apache-2.0

At a glance

Build RL reward signals using the OpenJudge framework. An agent skill from agentscope-ai/OpenJudge.

  • Building reward models
  • SKILL.md covers When to Use This Skill, Step 1 — Choose Your Reward…, Sub-documents — Read When… and Install, plus 3 more sections
  • Calls pip
  • Scoring rollouts for GRPO/REINFORCE

What it does

02 Rl Reward is an agent skill from agentscope-ai/OpenJudge. Build RL reward signals using the OpenJudge framework. Covers choosing between pointwise and pairwise reward strategies based on RL algorithm, task type, and cost; aggregating multi-dimensional pointwise scores into a scalar reward; pairwise tournament reward for GRPO on subjective tasks (net win rate across group rollouts); generating preference pairs for DPO/RLAIF; and normalizing scores for training stability. Use when building reward models, scoring rollouts for GRPO/REINFORCE, generating preference data for…

Its SKILL.md is about 1.6k tokens, which your agent loads only when the skill is triggered. The skill folder holds 2 other files (for example `pairwise.md` and `pointwise.md`).

It sits in AI & LLM Engineering, covering Fine-tuning and Reinforcement learning. The repository describes itself as: OpenJudge: A Unified Framework for Holistic Evaluation and Quality Rewards. The licence is Apache-2.0.

When your agent uses it

  • Building reward models
  • Scoring rollouts for GRPO/REINFORCE
  • Generating preference data for DPO
  • Doing Best-of-N selection

Example prompts

  • “/02-rl-reward”

Requirements

  • Python 3

What it can do on your machine

Read from SKILL.md and the folder at commit d1e0642. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • pip

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use pip, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

02 Rl Reward loads about 1.6k tokens when it runs. Until then it costs about 141 tokens; SKILL.md has 339 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~141
When it runs · the whole SKILL.md, loaded when a task matches
~1.6k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from agentscope-ai/OpenJudge at commit d1e0642, republished under its Apache-2.0 licence (© agentscope-ai). 339 words, ~1,599 tokens.

Download SKILL.mdSave it as .claude/skills/02-rl-reward/SKILL.md (or your agent's skills folder). This skill also uses 2 other files; get the full folder from GitHub.
name
02-rl-reward
description
Build RL reward signals using the OpenJudge framework. Covers choosing between pointwise and pairwise reward strategies based on RL algorithm, task type, and cost; aggregating multi-dimensional pointwise scores into a scalar reward; pairwise tournament reward for GRPO on subjective tasks (net win rate across group rollouts); generating preference pairs for DPO/RLAIF; and normalizing scores for training stability. Use when building reward models, scoring rollouts for GRPO/REINFORCE, generating preference data for DPO, or doing Best-of-N selection.

RL Reward Construction with OpenJudge

Build reward signals for reinforcement learning from human feedback (RLHF) and reinforcement learning from AI feedback (RLAIF) using the openjudge library.

When to Use This Skill

  • Building scalar rewards for GRPO / REINFORCE rollout scoring
  • Generating (chosen, rejected) preference pairs for DPO / IPO
  • Best-of-N candidate selection
  • Multi-dimensional reward shaping (correctness + safety + format)
  • Replacing or bootstrapping a reward model with LLM-as-judge

Step 1 — Choose Your Reward Strategy

Use this decision tree before writing any code:

RL Algorithm + Task type?
│
├── GRPO / REINFORCE — Verifiable task (math, code, structured output)
│   └── → POINTWISE  ✅  (FunctionGrader, exact score, zero LLM cost)
│
├── GRPO / REINFORCE — Subjective task (instruction following, dialogue, summarization)
│   └── → PAIRWISE TOURNAMENT  ✅  (compare each rollout vs all others in group,
│                                    reward = net win rate within group)
│
├── DPO / IPO / SLiC — need (chosen, rejected) pairs
│   └── → PAIRWISE  ✅  (two-way comparison, return winner/loser)
│
└── Best-of-N / reranking — rank N candidates
    └── → LISTWISE  ✅  (single call ranks all N at once)
Cost constraint?
├── Low budget
│   └── FunctionGrader (free) → pointwise; or pairwise with small judge model
│
├── Medium budget
│   └── Pointwise: 2–3 LLM graders + WeightedSumAggregator
│   └── Pairwise tournament: 1 LLM judge, N*(N-1)/2 comparisons per group
│
└── High quality / no cost limit
    └── Pointwise voting (3–5 calls) or pairwise with strong judge + debiasing

Sub-documents — Read When Relevant

TopicFileRead when…
Pointwise multi-dim rewardpointwise.mdGRPO on verifiable tasks; multi-dimension scoring
Pairwise rewardpairwise.mdGRPO on subjective tasks (tournament); DPO/RLAIF preference pairs

Read the relevant sub-document before writing any code.

Install

bash
pip install py-openjudge

Strategy Comparison

StrategyOutputReward signalTypical useCost
Pointwisescalar per responsedirect reward r(x, y)GRPO on verifiable tasks, filteringLow–Medium
Pairwise Tournamentnet win rate per responserelative reward within groupGRPO on subjective tasksMedium (N²/2 calls)
Pairwisewinner/loser pairimplicit preference y+ > y-DPO, IPO, RLAIF preference dataMedium
Listwiserank over N responsesordinal reward / rerankingBest-of-N, rerankingMedium–High

Score Normalization

All graders return scores on different scales. Always normalize before feeding into RL:

python
def normalize(score: float, min_score: float, max_score: float) -> float:
    """Map [min_score, max_score] → [0.0, 1.0]."""
    if max_score == min_score:
        return 0.0
    return (score - min_score) / (max_score - min_score)

# LLM graders (common/*) return 1–5 → normalize to 0–1
reward = normalize(result.score, min_score=1, max_score=5)

# FunctionGrader / text graders already return 0–1 → no normalization needed

Evaluation Strategies

Evaluation strategies control how many times a grader is called and how results are aggregated. They are independent of the grader itself.

Choose Your Strategy
Grader type?
│
├── Deterministic (FunctionGrader, StringMatch, CodeExecution, etc.)
│   └── → Direct  (zero variance, no need for aggregation)
│
├── LLM grader — Pointwise scoring
│   │
│   ├── Budget limited / speed critical
│   │   └── → Direct  (accept variance, 1× cost)
│   │
│   ├── Discrete scores (1–5 integer, pass/fail, binary)
│   │   └── → Voting  (majority vote, robust to outliers, N× cost)
│   │
│   └── Continuous / fine-grained scores (need precise ranking)
│       └── → Average  (mean, preserves signal, N× cost)
│
└── LLM grader — Pairwise GRPO tournament
    └── → GRPOTournament  (all-pairs comparison, net win rate)
StrategyAggregationBest forCost
DirectEvaluationStrategyNoneDeterministic graders; low budget1×
VotingEvaluationStrategyMajority voteDiscrete / integer LLM scoresN×
AverageEvaluationStrategyMeanContinuous LLM scoresN×
GRPOTournamentEvaluationStrategyNet win ratePairwise GRPO on subjective tasksN²/2×

All strategies are imported from openjudge.evaluation_strategy.

Pointwise — Noise Reduction with Voting / Average

For high-variance LLM judges, wrap any grader with VotingEvaluationStrategy to run N calls and take the majority vote:

python
from openjudge.evaluation_strategy import VotingEvaluationStrategy

grader = CorrectnessGrader(
    model=model,
    strategy=VotingEvaluationStrategy(num_votes=3, tie_breaker="closest_to_mean"),
)
# Now each call internally runs 3 LLM evaluations and returns the most common score

Use odd num_votes (3, 5) to avoid ties.

Pairwise — GRPO Tournament

For GRPO on subjective tasks, use GRPOTournamentEvaluationStrategy to run all-pairs comparison and compute net win rate per rollout:

python
from openjudge.evaluation_strategy import GRPOTournamentEvaluationStrategy

strategy = GRPOTournamentEvaluationStrategy(debiased=False)
results = await strategy.execute(
    pairwise_grader.aevaluate,
    query="Write a haiku about the ocean.",
    responses=["rollout_1", "rollout_2", "rollout_3", "rollout_4"],
)
rewards = [r.score for r in results]  # net win rates in [-1.0, 1.0]

Set debiased=True to run each pair in both orders and only count consistent results (doubles LLM calls but mitigates position bias).

© agentscope-ai, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 2 other files in skills/openjudge-core/02-rl-reward of agentscope-ai/OpenJudge.

  • SKILL.md
  • pairwise.md
  • pointwise.md

Open the folder on GitHubat commit d1e0642

Compare with similar skills

02 Rl Reward next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

02 Rl Reward compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
02 Rl Reward this skillagentscope-ai/OpenJudge871—~1.6kAutomated safety check: PassApache-2.0
Train RlOpenPipe/ART11k—~2.4kAutomated safety check: PassApache-2.0
Qwopus27b Rl TrainingR6410418/Jackrong-llm-finetuning-guide1.7k—~830Automated safety check: PassApache-2.0
Fine Tuning With TrlOrchestra-Research/AI-Research-SKILLs13k6 repos~2.9kAutomated safety check: PassMIT
Optim AgentOptim-Agent/optim-agent801—~1.3kAutomated safety check: PassMIT
Safactory WorkflowsAI45Lab/SAfactory236—~1.8kAutomated safety check: PassNone

Similar skills

  • Train Rl

    OpenPipe/ART

    RL training reference for the ART framework. An agent skill from OpenPipe/ART.

    11k GitHub stars~2.4k tokensUpdated today
    AI & LLM EngineeringAuto-check passed

  • Qwopus27b Rl Training

    R6410418/Jackrong-llm-finetuning-guide

    Prepare, validate, launch-plan, monitor, resume, and stop configurable Qwopus 27B reinforcement-learning workflows for GRPO or GSPO.

    1.7k GitHub stars~830 tokensUpdated 3 mo ago
    AI & LLM EngineeringAuto-check passed

  • Fine Tuning With Trl

    Orchestra-Research/AI-Research-SKILLs

    Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training.

    13k GitHub starsUsed in 6 repos~2.9k tokens
    AI & LLM EngineeringAuto-check passed

  • Optim Agent

    Optim-Agent/optim-agent

    A skill your agent uses when the user wants to optimize configurable system parameters against a measurable scalar objective, especially for model training, inference, quantitative strategies…

    801 GitHub stars~1.3k tokensUpdated 1 mo ago
    AI & LLM EngineeringAuto-check passed

  • Safactory Workflows

    AI45Lab/SAfactory

    Integrate a benchmark or custom environment into SAfactory using fixed adapter templates and local contract tests, optionally run Docker/RJob evaluation, or prepare GRPO/RL training.

    236 GitHub stars~1.8k tokensUpdated 17 days ago
    AI & LLM EngineeringAuto-check passed

  • Hugging Face LLM Trainer

    huggingface/skills

    Official

    Trains or fine-tunes language and vision models with TRL or Unsloth on Hugging Face Jobs cloud GPUs, then converts the results to GGUF.

    11k GitHub starsUsed in 1 repo~7.2k tokens
    AI & LLM EngineeringAuto-check passed

More from agentscope-ai/OpenJudge

All 19 skills in this repo

  • Align Human

    agentscope-ai/OpenJudge

    A skill your agent uses when the user has a judge/grader and human-labeled data, and wants to measure how well the judge agrees with humans, detect systematic biases, determine whether automatic…

    871 GitHub stars~3.1k tokensUpdated 1 mo ago
    Auto-check passed

  • Prompt Regression

    agentscope-ai/OpenJudge

    A skill your agent uses when the user has changed a prompt (system prompt, RAG template, agent instruction, etc.) and wants to know whether the candidate is better or worse than the baseline.

    871 GitHub stars~2.8k tokensUpdated 1 mo ago
    Auto-check passed

  • RAG Eval

    agentscope-ai/OpenJudge

    A skill your agent uses when the user has a RAG (Retrieval-Augmented Generation) system and wants to evaluate its quality — separating retrieval issues from generation issues.

    871 GitHub stars~2.4k tokensUpdated 1 mo ago
    Auto-check passed

  • Claude Authenticity

    agentscope-ai/OpenJudge

    Detect whether an API endpoint is backed by genuine Claude (not a wrapper, proxy, or impersonator) using 9 weighted rule-based checks that mirror the claude-verify project.

    871 GitHub starsUsed in 1 repo~5k tokens
    Auto-check passed

  • Eval Design

    agentscope-ai/OpenJudge

    A skill your agent uses when the user needs to design evaluation datasets, create test cases, stratify samples, generate adversarial examples, extract eval dimensions from traces/specs, or build a…

    871 GitHub stars~2.8k tokensUpdated 1 mo ago
    Auto-check: warnings

  • Find Skills Combo

    agentscope-ai/OpenJudge

    Discover and recommend combinations of agent skills to complete complex, multi-faceted tasks.

    871 GitHub starsUsed in 1 repo~4.6k tokens
    Auto-check: warnings

Categories

Questions about 02 Rl Reward

What does 02 Rl Reward do?

Build RL reward signals using the OpenJudge framework. An agent skill from agentscope-ai/OpenJudge. 02 Rl Reward is an agent skill from agentscope-ai/OpenJudge. Build RL reward signals using the OpenJudge framework.

When should I use 02 Rl Reward?

02 Rl Reward fits situations like: building reward models; scoring rollouts for GRPO/REINFORCE; generating preference data for DPO; doing Best-of-N selection.

How do I install 02 Rl Reward in Claude Code?

Run `npx skills add agentscope-ai/OpenJudge --skill 02-rl-reward -a claude-code`. Or copy the skill folder (skills/openjudge-core/02-rl-reward in agentscope-ai/OpenJudge) into .claude/skills/02-rl-reward in your project. Claude Code loads it when a task matches its description.

How do I install 02 Rl Reward in Codex?

Run `npx skills add agentscope-ai/OpenJudge --skill 02-rl-reward -a codex`. Or copy the skill folder (skills/openjudge-core/02-rl-reward in agentscope-ai/OpenJudge) into .agents/skills/02-rl-reward in your project. Codex loads it when a task matches its description.

Can I use 02 Rl Reward in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add agentscope-ai/OpenJudge --skill 02-rl-reward -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/02-rl-reward, .gemini/skills/02-rl-reward, .github/skills/02-rl-reward and .opencode/skills/02-rl-reward in your project.

What does 02 Rl Reward need to run?

Going by SKILL.md and its folder, 02 Rl Reward needs the command-line tools its instructions call (pip). Our summary lists: Python 3.

Does 02 Rl Reward access the network?

SKILL.md contains no URLs. Its commands use pip, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is 02 Rl Reward safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does 02 Rl Reward use?

02 Rl Reward is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does 02 Rl Reward use?

About 1.6k tokens (SKILL.md is roughly 6.4k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to 02 Rl Reward?

Skills that share tags, products or a category with 02 Rl Reward: Train Rl (OpenPipe/ART, 11k stars), Qwopus27b Rl Training (R6410418/Jackrong-llm-finetuning-guide, 1.7k stars), Fine Tuning With Trl (Orchestra-Research/AI-Research-SKILLs, 13k stars) and Optim Agent (Optim-Agent/optim-agent, 801 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains 02 Rl Reward?

agentscope-ai (a GitHub organization) maintains it in agentscope-ai/OpenJudge, which has 871 GitHub stars. The repository holds 19 skills in this directory. The repository was last updated on September 11, 2026.

Source: agentscope-ai/OpenJudge on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.