claude-sisyphus-grad: Command for Claude Code

.claude/commands/train.md

train is a command for Claude Code from Hwiyeon/claude-sisyphus-grad. It costs 31 tokens per session (1,489 once invoked), scanned A, original, MIT.

A command for automating repeated deep-learning training experiments from a script and configuration file.

In plain words
What is it for?
Use it to launch training loops, set experiment limits and environments, and evaluate runs against goals or decision criteria.
Why use it?
It organizes experiment runs and can repeat them with review and decision steps, reducing manual trial-and-error.

Command for Claude Code

Written for Claude Code: $ARGUMENTS substitution. Also seen: mentions CLAUDE.md; mentions subagents.

This is Hwiyeon/claude-sisyphus-grad's own configuration. It tells Claude Code how to work on claude-sisyphus-grad itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything claude-sisyphus-grad configures →

Needs its repository: it runs a file that does not travel with it, so clone the repository first. The line is ./scripts/train-loop.sh "script={script} config={config} max_experiments={max_experiments} experiment_title={experiment_title} env={env} analyze={analyze} subse.

Reuse

Borrowing it

Nothing to install: this file belongs to Hwiyeon/claude-sisyphus-grad. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/Hwiyeon/claude-sisyphus-grad/main/.claude/commands/train.md
Clone the repo
git clone --depth 1 https://github.com/Hwiyeon/claude-sisyphus-grad

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.

agentmods badge for train

README.md
[![agentmods](https://agentmods.dev/badge/commands/hwiyeon/claude-sisyphus-grad/train/github.svg)](https://agentmods.dev/commands/hwiyeon/claude-sisyphus-grad/train)
Your own site
<a href="https://agentmods.dev/commands/hwiyeon/claude-sisyphus-grad/train"><img src="https://agentmods.dev/badge/commands/hwiyeon/claude-sisyphus-grad/train/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.

agentmods 80×15 button for train

Your own site · 80×15
<a href="https://agentmods.dev/commands/hwiyeon/claude-sisyphus-grad/train"><img src="https://agentmods.dev/badge/commands/hwiyeon/claude-sisyphus-grad/train.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 31 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,489 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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00031 $0.01489
Opus 5 $0.00015 $0.00745
Sonnet 5 $0.00006 $0.00298
Haiku 4.5 $0.00003 $0.00149

Measured 11d ago against content hash ddb0f4956bcc, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-10, from the pricing page.

Security

Grade A, and why

train 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 11d 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.

.claude/commands/train.md · 151 lines

How it starts

The opening of the file, as written. The whole thing — 151 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Experiment Automation Launcher (External Script Loop)

This command parses arguments, creates the initial state, and runs scripts/train-loop.sh in the background to automatically repeat the experiment loop. Detailed orchestrator rules are defined in .claude/prompts/train-orchestrator.md.


Step 1: Parse Input Arguments ($ARGUMENTS)

Parse arguments in the format below. Use defaults for unspecified items:

/train script=train.py config=configs/myexp.yaml max_experiments=20 experiment_title="baseline_v1" env=myenv parallel=1 instructions="adjust lr based on previous run report before starting"
argument description default
script path to training script required
config path to config file for training required
experiment_title experiment title (used as folder name) required
max_experiments maximum number of experiments 20
review_cycles number of G Research Brief → round 2 → round 3 → Judge decision cycle repetitions 1
goal experiment termination condition (metric-based) optional
decision_criteria decision criteria (free text or @filepath) optional

| env | Python virtual environment name (training runs after conda activate {env}) | required |

| parallel | number of models to train simultaneously. If >=2, each runs as a separate parallel process | 1 | | instructions | pre-instructions for the orchestrator before training. Free text or @filepath to reference previous reports. Multiple files: @path1 @path2 | optional | | analyze | before training, analyze best model performance, training reports, and research logs to derive and apply improvements. May include fundamental changes to architecture, loss, data pipeline, etc. | false | | subset | if true, first validate with 20% subset, then switch to full dataset if promising | false | | circuit_breaker | if the same Judge decision type occurs N times consecutively, pass pattern detection to the Judge during round 2 review to strengthen decision direction. null disables this | null |

| lang | output language for all records, messages, and reports (ko, en, etc.). The project's CLAUDE.md defines the default | project default |

Read the full file on GitHub · 151 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.

  1. 11d ago First seen · 151 lines · 31 tokens per session scan A ddb0f4956bcc

Subscribe to this mod's changes

train is a command published in the GitHub repository Hwiyeon/claude-sisyphus-grad (59 stars, last pushed 3mo ago), licensed MIT. It adds 31 tokens to every session and 1,489 once invoked, about $0.0002 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.