Package guide

Quick start

Estimated time to the first AI response: 2 minutes. Time to the first ready candidate: depends on what you are building, not the skill.

This guide assumes you have already installed the Gauntlet Loop. If you haven't installed it yet, see the installation guide first.

1. Open the contract and fill in three fields

Inside the package, in templates/contrato-do-loop.md, there are ten sections. Only three are mandatory to start:

  1. Objective and deliverable. What exactly will be produced. An artifact, not an intention.
  2. Gating criteria. What disqualifies a candidate. Each line must be verifiable by inspection, counting, or measurement. "Good", "professional", and "premium" are not criteria.
  3. Budget. Time, number of cycles, or money. If you can't measure it in money, use time and cycles.

You don't need to fill out the other seven sections to run the first cycle. They exist for when the work justifies more control. If you want to see the expected level of detail, the package includes a complete example.

2. Paste this command into the chat

With the skill installed and the three fields filled in, paste something close to this:

Use a Gauntlet Loop to create [what you want]. My contract is below.

[paste the three fields you filled in here]

3. What to expect in the first few minutes

AI doesn't start producing. Before any draft, it confirms what is missing and asks at what depth the loop should run:

  • Essential, the default. Up to three gating criteria, one reference, a maximum of two cycles. No formal contract or separate scorecard. Works for most jobs.
  • Controlled. Full contract, per-cycle scorecard, explicit budget. For higher-impact deliverables.
  • Organizational. Everything in Controlled, plus tracking of amendments and named approvals. For work that needs to be audited by third parties.

If you don't choose, it defaults to Essential. After the first version, it checks the criteria itself before presenting the candidate as ready. If any criterion fails, it states which one and why, instead of delivering it as if everything is fine.

4. When to stop

The loop ends on the first of these events:

  • Every gating criterion passes and you approve the result.
  • The budget (time, cycles, or money) runs out.
  • Two rounds in a row repeat the same failure without new evidence.
  • You decide to stop, with or without a stated reason.

Stopping because the budget ran out is not a failure of the method. It is the method working. A real project using this skill reached seven cycles because the budget was increased midway, with a recorded decision. The complete case show how this looks in practice, including what hasn't improved.

If you want to record the result

After closing a loop, registration is voluntary: tell what you built. Help improve the material for whoever comes next.

Methodology created by Matt Shumer. Skill and materials adapted and structured by Israel Degasperi. Licensed under CC BY 4.0.