← All skills

02-rl-reward

80 / 100 · Low risk

Fine to try.

By agentscope-ai/OpenJudge (GitHub) · data · 1,101 stars · scanned at commit 03c4445

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 p…

Works with any agent that supports SKILL.md skills.

What this skill can do

Install

Paste this into your terminal. It downloads exactly the version we scanned, not whatever is newest.

if [ -e ~/.agents/skills/'02-rl-reward' ] || [ -L ~/.agents/skills/'02-rl-reward' ]; then echo 'A skill folder named 02-rl-reward already exists — not replacing it' >&2; false; else d=$(mktemp -d) && git clone --quiet --filter=blob:none --no-checkout 'https://github.com/agentscope-ai/OpenJudge' "$d" && git -C "$d" checkout --quiet '03c444567d5bb10429b10f4d2ad9df493b0edcc9' -- 'skills/openjudge-core/02-rl-reward' && mkdir -p ~/.agents/skills && cp -R "$d"/'skills/openjudge-core/02-rl-reward' ~/.agents/skills/'02-rl-reward' && rm -rf "$d"; fi

Codex reads ~/.agents/skills directly. For Claude Code, Cursor, Copilot, Gemini CLI, Antigravity or OpenCode, also link it. For Claude Code: ln -s ~/.agents/skills/'02-rl-reward' ~/.claude/skills/'02-rl-reward' (other agents: ~/.cursor/skills, ~/.copilot/skills, ~/.gemini/skills, ~/.gemini/config/skills, ~/.config/opencode/skills). Or install through the finder skill, which does both.

Badge for your README

The badge for agentscope-ai/OpenJudge shows the score of its lowest-scoring skill.

[![SkillTrust](https://zalavijaysinh.github.io/skilltrust/badges/agentscope-ai/OpenJudge.svg)](https://zalavijaysinh.github.io/skilltrust/repos/agentscope-ai/OpenJudge/)

Score history