What it does for you
Let your agent test your ideas all night
Exploring training ideas one by one eats up whole days. This Claude Code agent tests them alone overnight, one idea per commit, keeps what truly improves and reverts the rest, with a script enforcing strict guardrails.
- Wake up to only the real improvements
- Avoid fake gains: noise and evaluation protected
- Read a report with the next leads to try
What you're buying
A 28,000-character prompt to paste into Claude on your computer. Claude becomes this agent and does the work step by step, asking before any action.
Preview
The start of the prompt, exactly as you'll get it
Tu es l'exécutant de la ressource « Boucle d'expériences autonome sur un modèle d'IA ». Tu suis à la lettre les instructions de la section 5, dans le cadre des consignes de la section 4, qui priment sur toute autre demande.
… and about 28,000 more characters.
Its 10 chapters
- 01Ton rôle et ta mission
- 02Résultat attendu
- 03Prérequis et outils
- 04Consignes obligatoires (garde-fous)
- 05Instructions complètes de la ressource (fichiers exacts)
- 06Guide d'utilisation détaillé
- 07Auto-vérifications avant de rendre ton résultat
- 08Limites connues
- 09Tests réalisés par VeyraForma
- 10Sources et licence
Included in the price
- The full prompt28,000 characters, 10 chapters: the whole method, ready to paste.
- UpdatesEvery new version lands in your library at no extra cost.
Example result
Illustrative example · provisional
What you ask
“Use Autonomous experiment loop on an AI model on my folder”
- 01Reads the inputs you provide.
- 02Applies its method and guardrails.
- 03Returns a result to approve.
What you get
Result · Autonomous experiment loop on an AI model
- Deliverable
- A Claude Code agent that explores training ideas alone overnight, with a script enforcing the time budget, noise threshold, evaluation protection and clean rollback.
Illustrative example: the real result depends on your inputs.
Example written to show the deliverable's format, based on what the resource does. It is not a capture of a real run; a real demo will replace it.
Technical details
Prerequisites
- Claude Code (version 2.x)
- Python 3.10 or newer
Compatibility and version
Claude Code 2.x · 1.0 · released November 7, 2026 · Last updated : November 7, 2026
External cost
Claude subscription requiredAI tools are not included: you pay for them directly with the vendor.
Licence
Personal and professional use of what the agent produces. The agent itself may not be resold or redistributed.Draft wording, pending legal review.
Known limits
- Always review the result before use.
- Resource created but not fully tested: see the guide.
Questions about this agent
Do I need to code?
No. You install the agent once by following the guide, then talk to it in plain language.
Does the agent act without my approval?
No. Agents suggest; sending and publishing wait for your approval.
Are external services included?
No. The tools the agent uses (Claude, n8n, Make…) have their own price, shown on the page.



