Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/redai-infra/Relaxnpx agentmods add skills/redai-infra/relax/verl-to-relaxWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/redai-infra/relax/verl-to-relax)<a href="https://agentmods.dev/skills/redai-infra/relax/verl-to-relax"><img src="https://agentmods.dev/badge/skills/redai-infra/relax/verl-to-relax/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/redai-infra/relax/verl-to-relax"><img src="https://agentmods.dev/badge/skills/redai-infra/relax/verl-to-relax.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high YARA Match · line 69 YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).Fix: Remove offensive tool references and exploit code. Legitimate agent skills should not contain penetration testing tools, exploit frameworks, or reconnaissance utilities.
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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00078 | $0.10443 |
| Opus 5 | $0.00039 | $0.05221 |
| Sonnet 5 | $0.00016 | $0.02089 |
| Haiku 4.5 | $0.00008 | $0.01044 |
Grade A, and why
verl-to-relax 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 13d ago.
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.
How it starts
The opening of the file, as written. The whole thing — 813 lines — stays where its author put it; the contents beside it link to each section on GitHub.
verl → Relax Recipe Migration
This skill guides migration of RL training recipes (reward functions, tool environments, multi-turn rollouts, training scripts) from the verl framework to Relax.
For detailed import/code mapping tables and transformation templates, see references/migration_mapping.md.
Migration overview
A verl recipe typically consists of:
| verl Component | verl Location | Relax Equivalent | Relax Location |
|---|---|---|---|
Reward function (compute_score) |
verl/utils/reward_score/<dataset>.py or custom file |
Async reward_func(args, sample) |
examples/<algo>/reward_<algo>.py via --custom-rm-path |
Tool class (BaseTool) |
verl/tools/<tool>.py |
BaseInteractionEnv subclass |
examples/<algo>/env_<algo>.py |
| Multi-turn config YAML | examples/sglang_multiturn/config/ |
Custom config YAML | examples/<algo>/<algo>_config.yaml |
| Training launch script | examples/<recipe>/run_*.sh |
Shell script (python3 relax/entrypoints/train.py) |
examples/<algo>/run_<algo>.sh |
| Dataset class | verl/utils/dataset/rl_dataset.py or custom |
Parquet + CLI args | --prompt-data, --input-key, etc. |
| Hydra YAML config | verl/trainer/config/ppo_trainer.yaml |
CLI argparse flags | relax/entrypoints/train.py args |
| RewardManager | verl/workers/reward_manager/naive.py |
RewardExecutor + custom-rm-path |
relax/engine/rewards/ |
Reward 两层机制说明：Relax 的 reward 系统分为两层。
- 内置 reward（
relax/engine/rewards/）：通过--rm-type deepscaler|math|dapo|...直接使用，无需写 Python 代码。如果 verl 的compute_score恰好等价于某个内置类型（如简单数学答案校验），可直接使用--rm-type而不必迁移代码。- 自定义 reward（
--custom-rm-path）：当--custom-rm-path被设置时，RewardExecutor会优先加载用户函数，跳过内置分发。verl 的compute_score通常包含算法特定的打分逻辑，属于自定义范畴，因此迁移目标是examples/<algo>/reward_<algo>.py，通过--custom-rm-path examples.<algo>.reward_<algo>.reward_func注册。
The algorithm code lives under examples/<algo>/ in Relax — not inside the framework core.
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.
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.
- 13d ago First seen · 813 lines · 78 tokens per session scan A d1eb674deedd
verl-to-relax is a skill published in the GitHub repository redai-infra/Relax (580 stars, last pushed 15d ago), licensed Apache-2.0. It adds 78 tokens to every session and 10,443 once invoked, about $0.0004 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 2026-08-30.
Other skills, from other repositories
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.