totolford/project-control-center

A Windows mission control for multi-agent vibe coding with Claude Code.
> NEXUS is the working name of the app (repository: project-control-center).
Pick a project folder, give a mission in plain language, and a permanent Central
agent plans it, creates specialised worker agents, splits the work into tasks with
dependencies, routes messages between agents, reviews results and keeps the
project's memory up to date. Every agent is a real Claude Code session; every task,
message and file change is real and persisted.
It is not an IDE with a chatbot: it is a control center where you watch a swarm of
agents work on your project live — tiled agent panels with their real terminals, the
Central agent's coordination view, connections such as Roblox Studio, missions,
memory and a timeline. Use your usual editor alongside it.
## Features
AI World on the real AI Town - your agents live in an embedded
a16z-infra/ai-town (MIT) world: each agent
is a character that walks to the building of what it is really doing, talks when it
really sends a message and raises its hand when it waits for your approval. Runs on
a local Convex backend (no account; installed only after you confirm).
Central chat + command bar - the right panel is the real Claude Code
conversation with Central (or any agent); the bottom bar takes quick asks and
slash commands (/mission, /agent, /skill, /ssh, /model...).
Missions & Skill Market - queued and prioritized missions with an automatic
analysis and skill recommendations; a Skill Market over Claude Code marketplaces,
the NEXUS catalog and GitHub with a security analysis before any install.
Swarm workspace — a tiling workspace of panels (agent terminals, Central,
missions, task board, Roblox Studio / connection panels, diffs, memory, activity):
drag & drop, resize, maximize, minimize, pin, open in a new tab. Several workspace
tabs per project; the layout is saved in .agent-project/settings/workspace.json.
Claude Control Center — everything your installed Claude Code really exposes,
read live from it: models, context and rate-limit usage, MCP servers, skills,
plugins, settings, and a Command Center generated from its own --help.
MCP & Skills managers — test MCP servers (stdio, HTTP/SSE: tools, resources,
prompts, latency), add/remove/toggle them, import them for agents; create, edit
(with diff), disable, duplicate, test and export skills.
CLAUDE UNLOCKED — a maximum-autonomy mode that answers Claude Code's permission
prompts per your rules (never bypassing them), with a full decision journal,
power presets, Emergency Stop and Revoke-all.
Continuous improvement, Raw Terminal, Environment Inspector — scheduled
improvement missions, a real ConPTY terminal (Claude Code, PowerShell, CMD, WSL)
and a view of the machine, tools and project stack.
"Fais-moi ça" — give Central an intent ("connect to my Pi, check Docker,
update the server and test it"): it discovers its capabilities, reuses or creates
connections (SSH, MCP from a claude mcp add line, GitHub...), asks you only for
approvals, secrets (stored in the OS credential store) or sign-ins, and
journals every command.
NEXUS MASTER CONTROL — open domains (PC, GitHub, MCP, SSH, skills) to Central