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 skills add LigphiDonk/Oh-my--paper --skill remote-experimentgit clone --depth 1 https://github.com/LigphiDonk/Oh-my--paperWrote 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/ligphidonk/oh-my--paper/remote-experiment)<a href="https://agentmods.dev/skills/ligphidonk/oh-my--paper/remote-experiment"><img src="https://agentmods.dev/badge/skills/ligphidonk/oh-my--paper/remote-experiment/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/ligphidonk/oh-my--paper/remote-experiment"><img src="https://agentmods.dev/badge/skills/ligphidonk/oh-my--paper/remote-experiment.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00022 | $0.01066 |
| Opus 5 | $0.00011 | $0.00533 |
| Sonnet 5 | $0.00004 | $0.00213 |
| Haiku 4.5 | $0.00002 | $0.00107 |
Grade A, and why
Remote Experiment Execution 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.
What it actually says
Remote Experiment Execution
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。
命令速查
| 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 |
核心行为规则
规则 1：主动执行，不等催促
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
规则 2：失败后自主分析修复
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
规则 3：先探测环境
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
规则 4：区分 run vs ssh
- 改了本地代码 → 用
run（自动 sync + 执行） - 只想在服务器上执行某个命令（查看进程、安装包、看日志） → 用
ssh - 仅需同步代码不执行 → 用
sync up
规则 5：单变量修改
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。
迭代工作流
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
进度汇报
每次执行后简要说明：
- 做了什么改动、为什么
- 执行结果（成功/失败 + 关键输出）
- 下一步计划
📊 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
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 · 115 lines · 22 tokens per session scan A 7321e3ce868a
Remote Experiment Execution is a skill published in the GitHub repository LigphiDonk/Oh-my--paper (721 stars, last pushed 4mo ago), licensed MIT. It adds 22 tokens to every session and 1,066 once invoked, about $0.0001 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
tensorpool-gpu-cloud
Safely inspect and operate TensorPool GPU clusters and jobs using the current tp CLI, with explicit approval before any billable or destructive action.
lambda-gpu-cloud
Safely inspect and operate Lambda Cloud GPU instances through the documented Cloud API and SSH, with explicit approval before billable or destructive actions.
runpod-gpu-cloud
Safely inspect and operate RunPod resources with runpodctl, live product data, and explicit approval before paid or destructive actions.
vast-ai-gpu-cloud
Safely inspect and operate Vast.ai marketplace instances with the vastai CLI, live offer data, and explicit approval before paid or destructive actions.
sys-info
Retrieve system information and list running processes on a remote host via QuickDesk. Use when the user asks to check server health, monitor CPU or memory usage, view disk space, list running processes, or query OS and uptime details on a remote machine.
geoserver-cloud
Use when deploying GeoServer on Kubernetes as cloud-native microservices with auto-scaling, service discovery, and centralized configuration. GeoServer Cloud: break monolithic GeoServer into independently scalable WMS/WFS/WCS services.