reespec

From intent to outcome.

A framework for working with AI agents.

01 discover

One question at a time until the problem is sharp.

02 plan

Brief, design, specs — everything inspectable before work begins.

03 execute

Each task completes and verifies before the next one starts.

04 evaluate optional

Blind review: did the output match the intent?

npm install -g reespec
GitHub →

You've been here.

The gap between intent and outcome

Your AI delivered something — just not what you meant. The gap appears somewhere, but you can't pinpoint where.

Sessions that start from scratch

Every conversation resets. Decisions made yesterday aren't remembered. Context dissolves between sessions.

Work that loses the thread

Complex tasks drift. The AI loses the thread halfway through, confidently producing something that missed the point.

No way to know where it went off track

You can't inspect the work in progress. By the time you see the output, it's too late to course-correct.

Four phases. One closed loop.

Each phase directly addresses one of the ways AI work goes wrong. Use all four, or start with just discover and plan.

01

discover

Form the intent before the work starts.

fixes → Intent never fully formed upfront

Your AI asks you one question at a time, drawing out what you actually want — not what you think you want. Every branch of the decision tree gets resolved. You decide when there's enough clarity to move forward.

02

plan

Turn intent into a verifiable contract.

fixes → Work not broken into inspectable units

A structured set of artifacts — brief, design, specs — is produced before any work begins. Every task has a clear before and after state that anyone can inspect. Nothing is left to interpretation. You review and approve before execution starts.

03

execute

One verifiable step at a time.

fixes → AI loses the thread on long tasks

Work proceeds one task at a time. Each step has a clear starting condition and a clear completion check. The AI pauses and reports on blockers rather than guessing. Progress is always visible.

04

evaluate optional

Did the output match the intent?

fixes → No check that output matched intent

An adversarial review reads only the original contract and the actual outputs — never the implementation notes. It surfaces gaps between what was asked and what was delivered, giving you a structured verdict per capability.

How reespec compares

Most frameworks assume you're building software. reespec works for any complex task where intent needs to be expressed, work needs to be structured, and outcomes need to be verified.

  reespec BMad Method superpowers
Audience Anyone using AI agents Software developers Software developers
Weight Lightweight Heavy Medium
Non-code tasks ✓ First-class ✗ Dev-only ~ Partial
Structured discovery ~ Partial
Verifiable outputs ✓ All tasks ✓ Code only ~ Partial

Ready to close the gap?

Install reespec globally, initialise it in your project, and ask your AI agent to use the reespec skills. Three commands and you're in.

# install globally
$ npm install -g reespec
# initialise in your project
$ reespec init
# scaffold a new request
$ reespec new request "my-first-request"
Made with reespec and ♥ in EU