with one click
remote-experiment-execution
通过 compute-helper CLI 在远程服务器上自主执行、调试、迭代
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Menu
通过 compute-helper CLI 在远程服务器上自主执行、调试、迭代
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Search traceable academic papers, download legally accessible PDFs from arXiv and open-access sources, convert PDFs or page images to Markdown with a PaddleOCR layout-parsing API (or local pdfminer fallback), and organize the results into an AI-readable literature library. Use when Claude Code needs to build a paper corpus, batch OCR PDFs to Markdown, ingest real literature into a knowledge base, fetch arXiv or Hugging Face paper leads, or turn a directory of papers into structured Markdown plus metadata.
Delegate complex coding tasks to Claude Code CLI
Delegate coding tasks to OpenAI Codex CLI
Generates 2-4 candidate research directions from survey results, presents them with pros/cons for user selection, and converges to a publishable angle.
Academic research assistant for literature reviews, paper analysis, and scholarly writing.
# bioinformatics-init-analysis
| id | remote-experiment |
| name | Remote Experiment Execution |
| version | 2.0.0 |
| stages | [] |
| tools | ["bash","read_file","write_file"] |
| description | 通过 compute-helper CLI 在远程服务器上自主执行、调试、迭代 |
| summary | Autonomous remote code execution workflow — edit locally, sync, run remotely, analyze output, iterate |
| primaryIntent | remote_experiment_execution |
| capabilities | ["remote_code_sync","remote_command_execution","iterative_debugging","autonomous_optimization"] |
| domains | ["general","machine_learning","data_processing"] |
| keywords | ["experiment","remote","server","ssh","sync","compute-helper","iterate"] |
你可以通过 compute-helper CLI 在远程服务器上自主执行代码和命令。
compute-helper 路径和服务器信息在 system prompt 的
<compute_node>块中给出。 如果没有,在sidecar/bin/compute-helper.mjs查找。
| 命令 | 用途 | 何时用 |
|---|---|---|
node <helper> ssh "<cmd>" | 仅远程执行命令 | 检查环境、查看文件、不涉及代码改动时 |
node <helper> sync up --cwd <root> | 同步本地代码到服务器 | 手动同步(通常不需要,run 自动同步) |
node <helper> run "<cmd>" --cwd <root> | 同步代码 + 远程执行 | 修改代码后需要在服务器运行时 |
node <helper> sync down --cwd <root> --files "logs/ results/" | 从服务器拉回文件 | 需要查看结果文件时 |
node <helper> info | 查看节点配置 | 确认连接信息时 |
修改代码后 必须立即 run 远程执行验证,不要修改完就停下来等用户确认。
❌ 错误: "我已经修改了代码,你可以运行看看。"
✅ 正确: 修改代码 → 立即 run → 分析输出 → 汇报结果
远程执行报错时,立即分析错误输出 → 修改代码 → 再次 run,形成自修复循环。除非遇到无法判断的问题才向用户求助。
❌ 错误: "执行出错了,错误信息如下:..."(等用户处理)
✅ 正确: 分析错误 → 修改代码 → 再次 run → 如果还是失败 → 换策略再试
首次操作远程服务器时,先用 ssh 检查环境:
which python / python3 --version — Python 是否可用nvidia-smi — GPU 状态(如果需要)pip list | grep <package> — 依赖是否安装ls <workdir> — 工作目录状态run vs sshrun(自动 sync + 执行)sshsync up每次迭代只修改一个变量或一个方面,方便定位问题。同时改多处出错时无法判断哪个改动导致。
当用户要求在服务器上运行/测试/实验:
1. ssh 检查环境 ──→ 缺依赖?安装
│
2. 修改本地代码 ◄──────┘
│
3. run 远程执行 ──→ 成功?
│ ├─ 是 → 汇报结果,问下一步
│ └─ 否 → 分析错误 → 回到步骤 2
│
4. (可选) sync down 拉回结果文件
每次执行后简要说明:
📊 执行结果
━━━━━━━━━━
改动: 将 batch_size 从 32 改为 64
命令: python train.py --batch_size 64
结果: ✅ 训练完成,loss 从 0.45 降到 0.32
下一步: 尝试增大学习率到 3e-4