Ibrahim-3d/orchestrator-supaconductor

SupaConductor – an AI‑driven “software‑engineer‑as‑a‑service"

What it is – SupaConductor is a plugin for Anthropic’s Claude Code (and, via adapters, for Codex, Cursor, Google Antigravity, Windsurf and GitHub Copilot). It lets a coding‑agent act like a tiny development team. You give a single natural‑language goal (e.g. “Add Google‑OAuth login”) and the plugin automatically:

1. Writes a detailed spec.
2. Breaks the work into a dependency‑graph of tasks.
3. Executes those tasks (parallelising when possible).
4. Runs a set of built‑in evaluators (UI/UX, code‑quality, integration, business‑logic).
5. Fixes any failures and repeats the loop until everything passes.

All of this is orchestrated through slash commands that start with /orchestrator‑supaconductor:.

---

Core concepts

Concept	Role
Orchestrator	Central controller that runs the Evaluate‑Loop (plan → evaluate plan → execute → evaluate execution → fix → complete).
Agents / Skills	Small autonomous workers (15 agents, 39 skills) that perform planning, coding, testing, and evaluation.
Evaluators	Four quality‑check modules (UI/UX, code quality, integration, business logic).
Board of Directors	Five virtual executives (Architect, Product, Security, Ops, Experience) that can deliberate on high‑impact decisions.
Lead Engineers	Virtual specialists that give architecture, product, QA, and tech advice.
conductor/ folder	Project‑local metadata (specs, plans, track registry, decision logs, knowledge patterns).

---

How you use it

1. Install – either via the Claude Code marketplace or by cloning the repo and running scripts/build-platform-adapters.py to generate adapters for other hosts.
2. Initial setup – run /orchestrator‑supaconductor:setup once per project. It creates the conductor/ directory and asks a few questions about your codebase and tech stack.
3. Give a goal – e.g. /orchestrator‑supaconductor:go Add Stripe payment integration with webhooks.
4. Watch the loop – the plugin plans, executes, evaluates, and fixes automatically. You can check progress with /orchestrator‑supaconductor:status.
5. Control mode – edit conductor/config.json to choose agentic (fully autonomous) or human‑in‑the‑loop (pauses for your input at key steps).

---

Main commands (quick reference)

Command	What it does
:go <goal>	Starts a full end‑to‑end track for the described feature/bug.
:setup	Bootstraps the conductor/ folder and initial sprint.
:status	Lists all active tracks and their current step.
:implement	Manually step the Evaluate‑Loop forward on the current track.
:board‑meeting <topic>	Runs a full virtual board deliberation (written rationale, vote).
:board‑review <topic>	Quick board assessment (faster, less detail).
:cto‑advisor, :ux‑designer, …	Ask a virtual executive for advice on architecture, design, etc.
:loop‑planner, :loop‑executor, :loop‑fixer	Run individual phases of the loop for fine‑grained control.

---

Architecture at a glance

/go <goal> → Orchestrator → (Plan → Eval‑Plan → Execute → Eval‑Exec → Fix) × n → Complete

• Each phase is handled by a dedicated agent/skill.
• Parallel‑dispatcher runs independent tasks concurrently.
• Evaluators feed results back to the orchestrator, which may trigger up to 5 fix cycles or 3 plan revisions.
• Decision logs and knowledge files are stored as plain Markdown/JSON under conductor/ for human auditability.

---

Costs & performance

• Uses Claude’s API; planning/evaluation runs on Opus, execution on Sonnet.
• Roughly 3‑5× the token usage of a single manual Claude conversation because multiple agents are active, but the system automatically optimises cost (e.g., cheaper model for routine coding).
• Context usage is modular – only the active skill loads its ~5 k token prompt, keeping the main conversation under ~4 k tokens.

---

Who it’s for

• Teams that want to automate repetitive feature development or bug‑fix cycles.
• Solo developers who prefer a “set‑and‑forget” workflow for multi‑file changes.
• Organizations experimenting with AI‑augmented development pipelines and looking for a plug‑and‑play orchestration layer.

---

Limitations

• Not ideal for tiny one‑file edits – the overhead outweighs the benefit.
• Requires a supported coding‑agent host (Claude Code is the reference; others need the generated adapters).
• Higher API cost than manual prompting; you may need to monitor usage.

---

Community & contribution

• Discussions, issues, and roadmap are hosted on GitHub.
• Licensed under AGPL‑3.0; bundles the MIT‑licensed Superpowers library.

---

Bottom line – SupaConductor is a real, open‑source orchestration layer that turns a coding‑agent into a miniature, self‑managing engineering team. It’s built for Claude Code but ships adapters for several other AI‑coding assistants, letting you describe a feature once and let the AI handle planning, implementation, testing, and quality assurance automatically.