Gilfoyle
Multi-agent orchestration for Claude Code. Plan, execute, evaluate, and fix — automatically.
Conductor turns Claude Code into a structured engineering system. Instead of ad-hoc prompting, every task flows through a rigorous Evaluate-Loop with specialized agents, quality gates, and automatic recovery. It ships with 16+ agents, 36 skills, a 5-member Board of Directors for architectural decisions, DAG-based parallel execution, and a learning system that improves across projects.
Built on top of obra/superpowers v4.3.0.
Table of Contents
Key Features
Evaluate-Loop Workflow — Every task follows Plan → Evaluate Plan → Execute → Evaluate Execution → Fix. No code ships without passing quality gates.
16+ Specialized Agents — Orchestrator, planners, executors, evaluators, fixers, code reviewers, and executive advisors. Each agent has a focused role and clear boundaries.
Board of Directors — Five expert directors (Chief Architect, CPO, CSO, COO, CXO) deliberate on major decisions through a structured assess → discuss → vote → resolve protocol.
Parallel Execution — Plans include dependency graphs (DAGs). Independent tasks run simultaneously through ephemeral worker agents coordinated via a file-based message bus.
Lead Engineer System — Architecture, Product, Tech, and QA leads make autonomous decisions within defined guardrails, reducing interruptions while maintaining control over high-impact choices.
Learning System — A knowledge manager loads relevant patterns before planning. A retrospective agent extracts learnings after completion. The system gets better at your codebase over time.
Quick Start
1. Install
claude plugin marketplace add AhmedElhadarey/Gilfoyle
claude plugin install gilfoyle
2. Initialize Your Project
In Claude Code, run:
/conductor setup
This creates the conductor/ directory in your project with track registry, decision log, knowledge base, and workflow documentation.
Alternatively, /gilfoyle auto-initializes your project on first use -- no separate setup step needed.
3. Start Working
/gilfoyle <your goal>
That's it. The system analyzes your goal, creates a track with a spec and execution plan, evaluates the plan, executes tasks (in parallel where possible), evaluates the results, and fixes any issues automatically.
Examples
/gilfoyle Add Stripe payment integration
/gilfoyle Fix the login bug where users get logged out after 5 minutes
/gilfoyle Build an admin dashboard with user analytics
/gilfoyle Refactor the auth system from sessions to JWT
What Happens Behind the Scenes
- Goal Analysis — Parses intent (feature, bugfix, refactor), estimates complexity, extracts requirements
- Track Detection — Checks for existing matching tracks to resume, or creates a new one
- Spec Generation — Writes a
spec.md with requirements and acceptance criteria
- Plan Creation — Generates
plan.md with phased tasks, dependencies, and a DAG for parallel execution
- Plan Evaluation — Validates scope, checks for overlap with completed work, verifies dependencies. For major tracks, convenes the Board of Directors
- Execution — Runs tasks through specialized workers. Updates
plan.md after every task. Commits at logical checkpoints
- Evaluation — Dispatches specialized evaluators (UI/UX, code quality, integration, business logic) to verify the work
- Fix Cycle — If evaluation fails, creates targeted fix tasks and re-evaluates (max 3 cycles before escalating to you)
- Completion — Syncs business docs if needed, runs retrospective, updates track registry
The Evaluate-Loop
The core workflow. Every track follows this loop without exception.
AhmedElhadarey