Borrowing it
Nothing to install: this file belongs to redai-infra/Relax. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/redai-infra/Relax/main/.opencode/agents/launcher-expert.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/agents/redai-infra/relax/launcher-expert)<a href="https://agentmods.dev/agents/redai-infra/relax/launcher-expert"><img src="https://agentmods.dev/badge/agents/redai-infra/relax/launcher-expert/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/agents/redai-infra/relax/launcher-expert"><img src="https://agentmods.dev/badge/agents/redai-infra/relax/launcher-expert.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00038 | $0.01052 |
| Opus 5 | $0.00019 | $0.00526 |
| Sonnet 5 | $0.00008 | $0.00210 |
| Haiku 4.5 | $0.00004 | $0.00105 |
Grade A, and why
launcher-expert 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 10d 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 — 105 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Launcher & Orchestration Expert
Relax 的服务编排、部署生命周期、资源分配和健康管理。For project-level rules see AGENTS.md. Ray 底层细节见 ray-expert.
不用于：RL 算法 (algorithm-expert)、Megatron (megatron-expert)、FSDP (fsdp-expert).
三层架构
| 层 | 类 | 位置 | 职责 |
|---|---|---|---|
| Controller | Controller |
relax/core/controller.py |
顶层编排、训练循环 |
| Service | Service |
relax/core/service.py |
生命周期、placement groups |
| Implementation | Actor, Rollout, etc. |
relax/components/ |
具体训练/推理组件 |
Controller 初始化
Controller.__init__():
_initialize_data_system()— TransferQueue- 创建 DCS coordinator
- 部署 Metrics Service（可选）
- 注册所有 Ray Serve 服务
- 启动健康监控
Service 部署
每个 Service 创建 placement group → serve.run() 部署 → 返回 handle。
服务角色：actor · critic · rollout · advantages · genrm · actor_fwd · agent_loop
资源分配
--resource '{"actor": [1, 8], "rollout": [1, 8], ...}' # [num_serves, num_gpus]
--colocate # Actor/Rollout 共享 GPU
Colocate 模式：共享 PG + sleep/wake 机制切换训练/推理，需 --offload-train.
RolloutManager
位置: relax/distributed/ray/rollout.py
管理 SGLang 推理引擎：
- 引擎类型:
regular·prefill·decode·placeholder - 生命周期: 启动 SGLang → 健康探测 → 生成样本 → 权重更新 → 可选重启/缩放
关联: relax/distributed/ray/actor_group.py (RayTrainGroup)
健康监控
位置: relax/utils/health_system.py → HealthManager
- 周期性 ping 所有已注册服务
- 不健康时触发
on_unhealthy回调自动恢复 - RolloutManager 使用
concurrency_groups隔离健康检查 RPC
数据管道
RolloutDataSource → RolloutManager → SGLang → 奖励计算
→ TransferQueueController → SimpleStorageUnit
→ TransferQueueClient → TrainRayActor
存储后端: ray_storage_client (默认) · mooncake_client · yuanrong_client
采样器: grpo_group_n_sampler · rank_aware_sampler · sequential_sampler
故障排除
| 症状 | 可能原因 | 首要步骤 |
|---|---|---|
| Job 启动失败 | Ray 集群未初始化 | ray status 检查 |
| GPU 分配错误 | GPU 不足或 PG 冲突 | 对比 GPU 总数 vs 请求量 |
| Service 超时 | 初始化慢或 OOM | 增大超时；检查 GPU 内存 |
| Rollout 引擎崩溃 | SGLang 服务失败 | 检查 SGLang 日志；验证模型路径 |
| 权重同步超时 | NCCL 通信失败 | 检查网络；尝试 --colocate |
| TransferQueue 空 | Rollout 未产出数据 | 验证 rollout 服务健康 |
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.
- 10d ago First seen · 105 lines · 38 tokens per session scan A 6bcc841e7e9a
launcher-expert is an agent published in the GitHub repository redai-infra/Relax (580 stars, last pushed 12d ago), licensed Apache-2.0. It adds 38 tokens to every session and 1,052 once invoked, about $0.0002 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 agents, from other repositories
service-mesh-expert
Expert service mesh architect specializing in Istio, Linkerd, and cloud-native networking patterns. Masters traffic management, security policies, observability integration, and multi-cluster mesh configurations. Use PROACTIVELY for service mesh architecture, zero-trust networking, or microservices communication…
serv
Designs serverless architectures for Lambda, Cloud Functions, and Cloud Run — cold start mitigation, event-driven wiring, cost modeling, and IaC via SAM or Serverless Framework. Use when building or auditing serverless workloads. Trigger with "design a serverless architecture", "optimize my Lambda cold starts".
hono-builder
Expert Hono developer for implementing routes, middleware, handlers, and integrations with Cloudflare Workers bindings.
integration-engineer
Integration engineer. Connects the chatbot to messaging channels (Slack, KakaoTalk, web) and implements integration with external APIs and databases. Responsible for deployment and infrastructure.
drupal-ddev
Use this agent for managing DDEV-based Drupal 11 projects — installing/updating modules, running Drush commands, managing configuration, database operations, cache management, and Composer workflows.
render-expert
Expert in Render.com — deploying web services, workers, cron jobs, managed databases, environment variables, render.yaml (Infrastructure as Code), and production best practices. Use to configure, optimize, or debug deployments on Render.