A command that starts training for the current machine-learning experiment branch. It checks the branch, prepares a remote computing instance, synchronises the code, and launches training in tmux, a persistent terminal session.

Use it to provision an experiment machine, choose a training configuration, sync the experiment code, and start training.

It combines the setup steps needed to begin a training run and prevents training from being started from the main branch. It also handles the remote environment instead of requiring the training to run on the developer’s computer.

exp-run — Command for Claude Code by zhyx12/projtool

Install

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.

Clone the repo

git clone --depth 1 https://github.com/zhyx12/projtool

Made for: Claude Code.

Wrote 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.

README.md

[![agentmods](https://agentmods.dev/badge/commands/zhyx12/projtool/exp-run/github.svg)](https://agentmods.dev/commands/zhyx12/projtool/exp-run)

Your own site

<a href="https://agentmods.dev/commands/zhyx12/projtool/exp-run"><img src="https://agentmods.dev/badge/commands/zhyx12/projtool/exp-run/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.

Your own site · 80×15

<a href="https://agentmods.dev/commands/zhyx12/projtool/exp-run"><img src="https://agentmods.dev/badge/commands/zhyx12/projtool/exp-run.svg" alt="Reviewed on agentmods" width="80" height="20"></a>

Per session 27 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.

When invoked 548 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.

Origin original No closer match found in the catalogue.

Token cost

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	$0.00027	$0.00548
Opus 5.5	$0.00011	$0.00219
Sonnet 5.5	$0.00005	$0.00110
Haiku 4.5	$0.00003	$0.00055

Measured 2mo ago against content hash 21c0d65706db, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

exp-run 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 2mo 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.

src/projtool/assets/project_templates/claude/commands/exp-run.md · 50 lines

How it starts

The opening of the file, as written. The whole thing — 50 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are running the user's /exp-run command. Treat this as a request to launch training, end-to-end, with minimal back-and-forth. Read .proj-tool/project.toml for paths and entrypoint conventions.

Steps

Verify branch + cleanliness.
- Call mcp__projtool__worktree_status. If the current worktree is the main branch, refuse — training only happens on exp/* branches. Tell the user to mcp__projtool__new_experiment first.
- If dirty: ask the user whether to commit-and-train or abort. Don't pass allow_dirty=true unless the user explicitly says so.
Provision the instance.
- Call mcp__projtool__instance_up with the experiment_id from the branch name (exp/<topic> → exp-<topic>).
- The pre-tool hook may warn if a binding already exists; act on its output.
Pick a config.
- List .proj-tool/project.toml [entrypoint].config_dir.
- If the user provided a config in the slash-command argument, use it. Otherwise ask which config to run.
Start training.
- Call mcp__projtool__start_training with experiment_id and config_path. The tool flushes the code sync, writes the manifest stub, and starts a tmux session on the remote.
- Report the run_id and the predicted duration (read it from the config if present, otherwise say "unknown — I'll poll").
Hand back to the user.
- Tell them whether you're going to poll proactively or wait for them to come back.
- Do NOT block the conversation by polling in a tight loop.

On failure

start_training returning a hook block: explain the block (dirty worktree / unpushed commit / no instance), do not retry blindly.
instance_up failing on quota: tell the user, don't loop.
sync_flush taking >2min: ask the user whether to continue waiting or diagnose with mcp__projtool__diagnose.

Reminder

The full workflow rules live in .claude/skills/ml-research-workflow/. Load that skill if anything below the surface is unclear.

Read the full file on GitHub · 50 lines

Changes

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.

2mo ago First seen · 50 lines · 27 tokens per session scan A 21c0d65706db

Subscribe to this mod's changes

exp-run is a command published in the GitHub repository zhyx12/projtool (1 stars, last pushed 2mo ago), licensed MIT. It adds 27 tokens to every session and 548 once invoked, about $0.0001 per session on Opus 5.5. 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 2026-08-31.

Data and AI

exp-run next to 6 commands from other repositories
Mod	Type	Stars	Last push	Licence	Scan	Tokens	Works with
exp-runzhyx12/projtool	Command	1	2mo ago	MIT	A	27	Claude Code
bootmicrosoft/PromptKit	Command	113	9d ago	MIT	A	14	GitHub Copilot
music-suno-promptfrankxai/Starlight-Intelligence-System	Command	10	yesterday	MIT	A	23	Claude Code
audit-promptOwl-Listener/ai-design-skills	Command	199	5mo ago	MIT	A	13	Claude Code
laravel-ai-sdkjpcaparas/superpowers-laravel	Command	157	4mo ago	MIT	A	27	Claude Code
vlm-ocr-evaluationscdenney/open-science-skills	Command	69	yesterday	none	A	0	Claude Code
dare-llm-integrationdewtech-technologies/dare-method	Command	5	3mo ago	MIT	A	0	Claude Code