skills.rest
zniihgnexy
zniihgnexyCOMMUNITY
Explore all skills by this author →
GitHub

monitor-experiment

Collects and summarizes remote experiment results from screen sessions over SSH.

Updated Apr 1, 2026
Search Tags:#machine learning#SSH#Feishu#screen#baseline comparison#experiment monitoring#results collection
One-click install
npx skills add https://github.com/zniihgnexy/vid_tokenizer --skill monitor-experiment-zniihgnexy
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: monitor-experiment
Source: https://github.com/zniihgnexy/vid_tokenizer/tree/main/.claude/skills/monitor-experiment
Command: npx skills add https://github.com/zniihgnexy/vid_tokenizer --skill monitor-experiment-zniihgnexy

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running ML experiments on remote servers makes it hard to know whether training finished, what the metrics are, and how they compare to baselines without manually logging in and digging through logs. ## Core Features & Use Cases - Remote Session Inspection: Lists active screen sessions on a server and captures recent output from each one via SSH. - Structured Result Collection: Finds and parses JSON result files, then presents metrics in a comparison table with deltas against the correct baseline. - Anomaly Flagging & Notification: Detects negative deltas, NaN values, or divergence, and optionally sends a Feishu notification when experiments complete. - Use Case: You launched three codec training runs on a GPU server last night. Ask the assistant to check results, and it fetches each screen's output, parses the JSON summaries, and reports which method beat the baseline. ## Quick Start Ask the assistant to check the experiment results on your GPU server and summarize how each run compares to the baseline.

Frequently Asked Questions about monitor-experiment

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I check if my remote training experiment is finished?▼

Use SSH to list active screen sessions with screen -ls, then capture recent output from each session using screen hardcopy. If no sessions remain and JSON result files exist in the results directory, the experiment has completed.

How to collect experiment metrics from a remote server over SSH?▼

Fetch the latest JSON result files by listing them with ls -lt and reading them with cat over SSH. Parse the metrics and present them in a comparison table showing deltas against the matching baseline configuration.

What should I do if screen hardcopy fails to capture output?▼

If hardcopy fails, look for log files or tee output written by the training process instead. Check the results directory for JSON summaries, which often exist even when terminal capture is unavailable.

Why do my experiment results show NaN or negative deltas?▼

NaN values or negative deltas usually indicate training divergence, misconfigured baselines, or incomplete runs. Check the training logs for errors and confirm you are comparing against the baseline with the same configuration before concluding.

Can I get a notification when a remote experiment completes?▼

Yes, if a ~/.claude/feishu.json config exists, an experiment_done notification with the results summary and baseline delta is sent via Feishu. If the config is absent or set to off, notification is skipped entirely.

Related Skills

Explore More Software Engineering Skills→
zjunlpzjunlp
75

monitor-experiment

Monitor remote GPU experiments, collect results, and finalize cost manifests over SSH.

Official
Advanced
ShiienShiien
7

experiment-analyze

Parse experiment metrics and logs to assess significance against baselines.

Community
Intermediate
goupup-aigoupup-ai
2

training-check

Monitor WandB training metrics to detect NaN loss and divergence.

Community
Intermediate
NousResearchNousResearch
227.8k

weights-and-biases

Track and visualize machine learning experiments with Weights & Biases.

Official
Intermediate