COLLECTIVE AGENT / RESEARCH DOCUMENTATIONAll docs
SYSTEMS / RESEARCH PREVIEW

Collective Agent

An experimental agent harness for work that can be inspected, independently tested, and accepted only with explicit authority.

Current status: experimental

The private project is not a public execution service or a certified unattended coding system. Individual tests do not imply universal security or general accuracy.

From task to evidence

01 / SCOPEFreeze task

Bound the source, tests, and budget.

02 / PROPOSEWorker

Generate a reviewable candidate.

03 / CHECKVerify

Run independent acceptance checks.

04 / DECIDEReview

Operator controls next action.

Documentation topics

6 topics found

01 / GUIDE

Introduction

What the system does, where its boundaries lie, and what is still experimental.

02 / GUIDE

Getting started

The evidence-first sequence from a frozen task to a reviewable candidate.

03 / GUIDE

Architecture

How workers, restricted execution, checking, and approval are separated.

04 / GUIDE

Permissions

Why child privileges are narrow and how explicit authorization works.

05 / GUIDE

Verification

What the measurements establish—and what they do not.

06 / GUIDE

Roadmap

The experiments required before expanding autonomy or execution authority.

Read the evidence, not the claim.

Experiments are labeled with their limitations. A failed held-out test remains a failed task even if earlier visible checks passed. Deployment is not automatic.