Launch: Pre-flight Checklist for ML Training Jobs

A five-step pre-flight checklist — config diff, run naming, path verification, monitoring, and ETA — plus restart/kill cleanup for long GPU training runs.

Data & AnalyticsAdvanced★ 406⑂ 34AI score 8/10Last updated: Sep 16, 2026

What it does

Config diff: locates a recently modified reference config, diffs it against the intended one, and reviews every line for silent regressors (num_workers, batch_size, learning_rate, optimizer betas, mixed_precision, grad accumulation, seed).
Run-name discipline: rejects session-local labels like run-1 or phase2-internal and enforces a descriptive <dataset/task>-<model>-<key-config>-<recipe> pattern.
Path verification: shells out to ls/cat to prove the dataset, pretrained checkpoint, output parent dir, and config actually exist before GPUs are committed.
Monitoring setup: auto-detects wandb / neptune / mlflow / tensorboard from env vars and launcher imports, then confirms project, entity, tags and groups.
ETA: computes wall-clock hours and reports finish time in your local timezone, flagging runs that straddle meetings or sleep.
Restart & kill cleanup: purges local checkpoints, remote artifacts, tracker runs, and scheduler reservations (scancel, cron, cloud reservations) in a fixed order to avoid ghost state.

Who it's for

Researchers and ML engineers running multi-hour or multi-day training jobs
Anyone submitting multi-GPU work via torchrun, accelerate launch, deepspeed, or sbatch
Teams who have lost a day to a missing dataset path or a stale framework default

Example uses

"Kick off this training command" → the skill diffs against your baseline config, flags num_workers=8 on an NFS mount and an unscaled learning rate, then prints a copy-ready launch command.
"Restart the run that died overnight" → cleans local and remote checkpoint dirs, deletes the stale tracker run, cancels the lingering SLURM job, then relaunches.
"Review this sbatch script" → verifies all paths, proposes a self-describing run name, and reports expected completion time in your local timezone.

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/fcakyon/phd-skills/HEAD/plugin/skills/launch/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)

Download the ZIP with the button below.
In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.

↓ Download ZIP

Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the plugin/skills/launch folder from the GitHub repo fcakyon/phd-skills into my ~/.claude/skills/fcakyon-launch/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/fcakyon/phd-skills.git && mkdir -p ~/.claude/skills && cp -r phd-skills/plugin/skills/launch ~/.claude/skills/launch

⚠ This is a third-party skill. Check the source repository before installing.

Open a terminal and change into a working directory.
Clone the repo: git clone https://github.com/fcakyon/phd-skills.git
Create the skills folder: mkdir -p ~/.claude/skills
Copy just this skill: cp -r phd-skills/plugin/skills/launch ~/.claude/skills/launch
Restart Claude Code and confirm launch appears in your skills list.
From your training project directory, say "pre-flight check this training launch" to trigger the checklist.
(Optional) Also install the repo's destructive_path_guard.sh and timezone_scrub.sh hooks so the cleanup steps get their safety net.

View source on GitHub ↗License: MIT