Flagship project / Agent system

CLI Harness

A local-first autonomous AI development environment designed around director and worker agents, extensible plugins, tool use, and long-running Wiggum loops.

Operation
Local first
Control
Director / worker
Extension
Plugin based
Loop mode
Wiggum

01 / Architecture

One director, many ways to work

The director coordinates work across specialized workers, plugins, and project tools. Outputs feed an inspect-and-improve loop that can continue under human review.

This first diagram is static by design. Its component boundary leaves room for richer run visualization later without putting client JavaScript into the foundation build.

CLI/HARNESS · director-worker topologyRouting

02 / Capabilities

A system for directed autonomy

The harness joins coordination, project-aware tooling, and iterative review without treating autonomy as a single opaque action.

CAP-01

Director / worker agents

A coordinating director can delegate focused work to specialized workers or agents.

CAP-02

Wiggum loops

Long-running loops plan, create, inspect, improve, review, and continue.

CAP-03

Plugin system

Project-specific capabilities can be introduced without making every workflow core.

CAP-04

Tool use

Agents can be exposed to development and creative tools through controlled workflows.

CAP-05

Local models

The project is oriented toward local AI experimentation and local-first operation.

CAP-06

Review systems

Planning and inspection remain visible parts of the loop instead of hidden implementation detail.

03 / Wiggum loop

Work becomes a visible cycle

Wiggum loops are autonomous or semi-autonomous runs that repeatedly move through a structured improvement cycle.

  1. PlanEstablish the next bounded objective.
  2. CreateProduce code, media, scenes, or project artifacts.
  3. InspectRead the result and collect concrete evidence.
  4. ImproveAddress the most useful next change.
  5. ReviewEvaluate quality, constraints, and direction.
  6. ContinueRepeat while the run remains useful and authorized.

04 / Workflows & plugins

Project-aware creative tooling

The plugin boundary gives different kinds of work their own tools and review patterns. These descriptions name intended workflows, not unverified performance claims.

PLG-BLEND

Blender

Generate or manipulate scenes through an agent-directed Blender workflow.

PLG-IMAGE

Image generation

Expose image-generation systems to iterative agent workflows and review.

PLG-RENPY

Ren'Py design

Support visual-novel planning and project-specific creation workflows.

05 / Field media

Real outputs will live here

These explicit placeholders reserve space for future screenshots and run captures without presenting stock art as project evidence.