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.
npx agentmods add instructions/redai-infra/relax/agents-mdgit clone --depth 1 https://github.com/redai-infra/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/instructions/redai-infra/relax/agents-md)<a href="https://agentmods.dev/instructions/redai-infra/relax/agents-md"><img src="https://agentmods.dev/badge/instructions/redai-infra/relax/agents-md.svg" alt="Measured on agentmods" height="20"></a>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 | $0.01627 | $0.01627 |
| Opus 5 | $0.00813 | $0.00813 |
| Sonnet 5 | $0.00325 | $0.00325 |
| Haiku 4.5 | $0.00163 | $0.00163 |
Grade A, and why
Relax AGENTS.md 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 6d 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 — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Relax
Relax 是一个基于 Ray Serve 的大模型强化学习训练框架，支持 Megatron 训练后端、SGLang 推理引擎、以及 GRPO/PPO/DAPO 等算法族。
Tech stack: Python 3.8+ · PyTorch · Megatron · SGLang · Ray / Ray Serve
Project Structure
relax/ 核心框架
├── core/ 编排层 — 训练循环、服务基类、全局注册表
├── components/ 组件层 — RL 服务组件（Ray Serve Deployment）
├── engine/ 引擎层 — Rollout 数据生成、奖励计算、请求路由
├── backends/ 后端层 — Megatron 训练后端、SGLang 推理引擎
├── distributed/ 分布式层 — Ray 集群管理、分布式 Checkpoint
├── entrypoints/ 入口层 — 训练入口脚本
└── utils/ 基础设施 — 工具函数、指标监控、多模态处理
tests/ 测试（镜像 relax/ 层级）
├── backends/megatron/ Megatron 后端测试（权重转换等）
├── distributed/ray/ 分布式 / Ray 测试（弹性伸缩等）
├── engine/rewards/ 奖励函数测试
├── engine/rollout/ Rollout 引擎测试（预取、数据源等）
└── utils/ 工具函数测试（HTTP、指标、流式数据集等）
transfer_queue/ 分布式数据传输队列
examples/ 用户级示例（deepeyes、OPD 等）
scripts/ 训练启动脚本 & 模型配置
configs/env.yaml 运行时环境配置
修改代码时重点关注：relax/utils/utils.py、relax/components/、relax/core/controller.py、relax/entrypoints/train.py。
Code Standards
- Ruff 格式化，行宽 119，
isort管理导入（配置见pyproject.toml） relax/下所有.py须含版权头：# Copyright (c) 2026 Relax Authors. All Rights Reserved.- pre-commit 对
transfer_queue/排除大部分检查 - 日志统一用
relax.utils.logging_utils.get_logger(__name__)，禁止print/logging.getLogger - 显式类型注解；禁止通配导入（
from x import *）；重型可选依赖放函数内导入 - 热路径禁止 GPU-CPU 同步（
.item()、.tolist()、print(tensor)） - 组合优于继承，层级 ≤ 2；偏好委托
Naming Conventions
| 类型 | 模式 | 示例 |
|---|---|---|
| Service 类 | @serve.deployment |
Actor, Rollout, Critic, GenRM |
| 后端 Actor | XxxTrainRayActor |
MegatronTrainRayActor |
| Actor 组 | RayTrainGroup |
管理一组训练 Ray Actor |
| Manager | XxxManager |
RolloutManager, GenRMManager |
| Reward 函数 | engine/rewards/xxx.py |
deepscaler.py, math_utils.py |
| 数据源 | XxxDataSource |
RolloutDataSource |
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.
- 6d ago First seen · 119 lines · 1,627 tokens per session scan A 38b8d985315a
Relax AGENTS.md is an instructions file published in the GitHub repository redai-infra/Relax (580 stars, last pushed 8d ago), licensed Apache-2.0. It adds 1,627 tokens to every session, about $0.0081 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 instructions, from other repositories
hands-on-modern-rl AGENTS.md
AGENTS.md instructions for walkinglabs/hands-on-modern-rl, covering repository guidelines, project structure, commands, editing rules and tikz → svg workflow.
theartifact-cli CLAUDE.md
Claude Code instructions for beast-bruiser/theartifact-cli, covering claude.md, commands, build, run with local backend and one-shot commands.
openrouter-mcp-multimodal AGENTS.md
AGENTS.md instructions for stabgan/openrouter-mcp-multimodal, covering agent instructions, before you ship, releasing (read this before publishing), short version and version files (must all match package.json).
pi-tinker AGENTS.md
Instructions for gvkhosla/pi-tinker, covering agent guide for pi-tinker, product promise, agent compatibility, best first commands for users and what is possible.
pi-tinker copilot-instructions.md
Instructions for gvkhosla/pi-tinker, a project described as: Fine-tune open-source models with Tinker from inside Pi — managed improve loops, data prep, evals, smoke tests, deploy snippets, and checkpoint chat.
AI-System-Design-Consultant CLAUDE.md
Claude Code instructions for deepanshu2711/AI-System-Design-Consultant, covering claude.md, what this is, running it, load-bearing typos — do not "fix" these paths and agent node pattern.