WAM
DETAILS 详情 SOURCE ↗
AGENT FRAMEWORK — EST. 2026

Wait a Minute.

An agent framework that stops, looks at the whole situation, and then — and only then — takes the next step.

READ THE MANIFESTO ↓
THE PROBLEM

Four structural defects

Current agent frameworks answer within the conversation, but never beyond it. One question in, one answer out — an infinite extension of Q&A that is not collaboration.

  1. 01

    Constructiveness

    Responses react to the latest message; they do not build toward a goal. The agent responds — it does not construct.

  2. 02

    Continuity

    Every conversation starts from zero. Break the rhythm once, and complex projects collapse. Humans re-align through communication; agents simply lose the thread.

  3. 03

    Verifiability

    Output sounds reasonable but does not hold up under scrutiny. Even “check the paper” cannot be trusted — let alone turning unknown-unknowns into known-unknowns.

  4. 04

    Weight judgment

    The agent cannot tell what matters: critical path, nice-to-have, or noise. It treats everything as equally important — and grasps at everything.

STATE DESYNC — beneath all four lies this. The human carries context, intent, and the weight of every decision; the AI carries none of it. This is not a prompt-engineering problem. It is a structural defect of the framework.

THE PHILOSOPHY

The sage and the doer

A deepest sage with the hands of a pragmatic doer. One entity — however many agents run beneath it.

THE SAGE — THINKS

Sees the whole situation

It traverses possibilities instead of preset shortcuts. It knows when to be fast and when to be slow. It senses the boundary between what we know and what we don't. It is sensitive to weight.

THE DOER — ACTS

Takes verifiable steps

It pushes when feasible and admits when not — no bluffing, no forcing. It focuses on the most important 1% and leaves continuable progress behind, never an empty promise.

1%
We build the bottom 1% of the foundation — and build it solidly. No 100% shell. State sync, goal construction, progress continuity, uncertainty management — answered before any feature list.
RESPONSE

Six principles

  • 1

    Slow down to speed up

    Establish the state before taking the next step.

  • 2

    No bluffing

    Acknowledge real difficulty — then work through it together.

  • 3

    A directional 1%

    Every step moves toward the goal, however small.

  • 4

    Weight-sensitive

    Hold the critical path; ignore the noise.

  • 5

    Honest

    If it works, push. If it doesn't, say so. Honesty beats polish.

  • 6

    Stay in sync

    Keeping human and agent aligned is the framework's core duty.

SYSTEM STATUS

M1 — direction competition

Four feasible approaches are competing. The first whose vertical slice passes the promotion gate defines the architecture; the rest archive or merge.

ACTIVE DIRECTIONS
global-routeGlobal orchestration entity — state awareness, confidence gate, spawned processes, continuous flow. (author)CRITICAL
substrateState substrate — a small library + CLI for the four-file state model, adoptable by any agent.CRITICAL
agent-pluginSix-step protocol embedded in existing agents (skill / MCP server).CRITICAL
weight-firstGoal DAG + critical-path computation + unknown-unknown registry — the original contribution.CRITICAL
Convergence rule: first verified vertical slice wins. Everything unverified lives in directions/ until it passes the promotion gate.