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

Add Reward Score

by verl-project in verl-project/verl-omni

Guide for adding a new reward scorer to verl-omni and wiring it into a run.

Apache-2.0Auto-check passedAI & LLM Engineering

Install Add Reward Score

skills CLI
$ npx skills add verl-project/verl-omni --skill add-reward-score -a claude-code

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

GitHub CLI
$ gh skill install verl-project/verl-omni add-reward-score --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/verl-project/verl-omni.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.agents/skills/add-reward-score .claude/skills/add-reward-score && 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
add-reward-score
GitHub stars
1.2k
Token cost
~648 tokens
SKILL.md length
220 words
Files
1
Skills in repo
7
Repo updated
First seen
Licence
Apache-2.0

At a glance

Guide for adding a new reward scorer to verl-omni and wiring it into a run.

  • Works in 4 steps: Copy the closest existing scorer → Write it → Select it from a run → …
  • Adding a reward function
  • SKILL.md covers Step 1 — Copy the closest…, Step 2 — Write it, Step 3 — Select it from a run and Step 4 — Test
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Add Reward Score is an agent skill from verl-project/verl-omni. Guide for adding a new reward scorer to verl-omni and wiring it into a run. Use when adding a reward function or reward model for image, video, or multimodal RL (flow-GRPO, DanceGRPO, DPO), including preference models and remote HTTP scorers.

Its SKILL.md is about 650 tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It sits in AI & LLM Engineering, covering Fine-tuning and Reinforcement learning. The repository describes itself as: Multimodal RL training framework for diffusion & omni models. The licence is Apache-2.0.

When your agent uses it

  • Adding a reward function
  • Reward model for image
  • Multimodal RL (flow-GRPO
  • Including preference models and remote HTTP scorers

Example prompts

  • “/add-reward-score”

Requirements

  • Python 3

Workflow steps

4 steps, taken from the step headings in SKILL.md.

  1. Copy the closest existing scorer
  2. Write it
  3. Select it from a run
  4. Test

What it can do on your machine

Read from SKILL.md and the folder at commit 17393f3. 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

    No scripts in the folder and no shell commands in SKILL.md (its code samples are bash and python).

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

  • Network

    No URLs in SKILL.md.

    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

Add Reward Score loads about 648 tokens when it runs. Until then it costs about 65 tokens; SKILL.md has 220 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~65
When it runs · the whole SKILL.md, loaded when a task matches
~648

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 verl-project/verl-omni at commit 17393f3, republished under its Apache-2.0 licence (© verl-project). 220 words, ~648 tokens.

Download SKILL.mdSave it as .claude/skills/add-reward-score/SKILL.md (or your agent's skills folder).
name
add-reward-score
description
Guide for adding a new reward scorer to verl-omni and wiring it into a run. Use when adding a reward function or reward model for image, video, or multimodal RL (flow-GRPO, DanceGRPO, DPO), including preference models and remote HTTP scorers.

Add a Reward Scorer

One new file under verl_omni/utils/reward_score/, then select it from the run config — there is no code registration step.

The signature, return, failure, and caching contracts live in the reward rule; read it first, this skill does not repeat it. For the machinery around the scorer see docs/algo/async_reward.md (reward-loop workers, resource pools, config reference) and docs/start/http_scorer.md (the request/response protocol for a remote scorer service).

Step 1 — Copy the closest existing scorer

bash
grep -rn "^async def compute_score\|^def compute_score" verl_omni/utils/reward_score/*.py

genrm_ocr.py::compute_score_ocr is the most-copied remote pattern and the one most examples/ scripts use; jpeg_compressibility.py is the minimal rule-based one; hpsv3_reward.py is the reference for a locally-loaded preference model. reward_utils.py holds the tensor/PIL conversion helpers — use them rather than re-deriving the conversion.

Step 2 — Write it

Apache 2026 header, a module docstring crediting upstream if the score is adapted, and one compute_score / compute_score_<name> entrypoint matching the rule's keyword contract. Nothing else is required of the file.

Step 3 — Select it from a run

The config keys and their semantics (custom_reward_function vs reward_functions.<key> with weight / required, and how extra keys become kwargs) are in the reward rule.

Copy a working invocation from examples/ rather than assembling one by hand — examples/flowgrpo_trainer/sd35/run_sd35_medium_drm_lora.sh is the multi-reward reference.

Step 4 — Test

tests/utils/reward_score/test_<name>_on_cpu.py, asserting the score is finite and in range for a synthetic tensor; mock the transport for an async scorer (run-cpu-tests).

python
def test_score_is_finite_on_cpu():
    image = torch.randint(256, (3, 64, 64), dtype=torch.uint8)
    out = <name>.compute_score(solution_image=image)
    assert 0.0 <= out["score"]
<!--
MAINTAINER GUIDE — keep this procedural; contracts belong in rules/reward.md. Update
Step 1's prose when the set of reference scorers changes materially.
-->

© verl-project, 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

Just SKILL.md in .agents/skills/add-reward-score of verl-project/verl-omni.

Open the folder on GitHubat commit 17393f3

Compare with similar skills

Add Reward Score 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.

Add Reward Score compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Add Reward Score this skillverl-project/verl-omni1.2k—~648Automated 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 verl-project/verl-omni

  • Add Pipeline

    verl-project/verl-omni

    Router for adding a diffusion or omni pipeline to verl-omni.

    1.2k GitHub stars~1k tokensUpdated yesterday
    Auto-check passed

  • Profile

    verl-project/verl-omni

    Route a verl-omni performance investigation to the right tool and capture a usable trace.

    1.2k GitHub stars~1.1k tokensUpdated yesterday
    Auto-check passed

  • Run Cpu Tests

    verl-project/verl-omni

    How to write and run verl-omni CPU tests (testoncpu.py) that exercise adapters, rewards, and configs without a GPU or model weights.

    1.2k GitHub stars~673 tokensUpdated yesterday
    Auto-check passed

  • Commit And PR

    verl-project/verl-omni

    verl-omni commit message + PR conventions and the mandatory contribution policy.

    1.2k GitHub stars~1.4k tokensUpdated yesterday
    Auto-check passed

  • Self Review

    verl-project/verl-omni

    Review your own verl-omni branch against the project rubric before opening or updating a PR.

    1.2k GitHub stars~2k tokensUpdated yesterday
    Auto-check passed

  • Train Infer Consistency

    verl-project/verl-omni

    Route verl-omni training/inference consistency checks through MindStudio's MSProbe collection and root-cause analysis skills.

    1.2k GitHub stars~860 tokensUpdated yesterday
    Auto-check passed

Categories

Questions about Add Reward Score

What does Add Reward Score do?

Guide for adding a new reward scorer to verl-omni and wiring it into a run. Add Reward Score is an agent skill from verl-project/verl-omni. Guide for adding a new reward scorer to verl-omni and wiring it into a run.

When should I use Add Reward Score?

Add Reward Score fits situations like: adding a reward function; reward model for image; multimodal RL (flow-GRPO; including preference models and remote HTTP scorers.

How do I install Add Reward Score in Claude Code?

Run `npx skills add verl-project/verl-omni --skill add-reward-score -a claude-code`. Or copy the skill folder (.agents/skills/add-reward-score in verl-project/verl-omni) into .claude/skills/add-reward-score in your project. Claude Code loads it when a task matches its description.

How do I install Add Reward Score in Codex?

Run `npx skills add verl-project/verl-omni --skill add-reward-score -a codex`. Or copy the skill folder (.agents/skills/add-reward-score in verl-project/verl-omni) into .agents/skills/add-reward-score in your project. Codex loads it when a task matches its description.

Can I use Add Reward Score 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 verl-project/verl-omni --skill add-reward-score -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/add-reward-score, .gemini/skills/add-reward-score, .github/skills/add-reward-score and .opencode/skills/add-reward-score in your project.

What does Add Reward Score need to run?

SKILL.md names no scripts, command-line tools or credentials: Add Reward Score is instructions for the agent only. Our summary lists: Python 3.

Does Add Reward Score access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Add Reward Score 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 Add Reward Score use?

Add Reward Score 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 Add Reward Score use?

About 648 tokens (SKILL.md is roughly 2.6k 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 Add Reward Score?

Skills that share tags, products or a category with Add Reward Score: 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 Add Reward Score?

verl-project (a GitHub organization) maintains it in verl-project/verl-omni, which has 1,222 GitHub stars. The repository holds 7 skills in this directory. The repository was last updated on October 10, 2026.

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