PLUGINS & ORCHESTRATION

BeadsHive

Multi-role orchestration for Claude Code workers. Conductor, Gatekeeper, and Planner roles coordinate development across multiple sessions.

Roles

100%

Shell

5 MB

Size

Jan 2026

Created

Project Overview

BeadsHive solves the multi-agent coordination problem with a simple but powerful idea: give each AI worker a defined role with specific permissions and responsibilities. Instead of every worker having full access, roles constrain what each session can do.

The entire system is implemented in pure shell scripts -- no Python, no Node, no compiled binaries. This makes it instantly portable to any Unix system with bash and tmux. The role definitions, access controls, and coordination protocols are all declarative shell configurations.

Tech Stack

Shell 100%

Bash

Role definitions, access control, coordination

tmux

Multi-session management backend

Git Worktrees

Isolated branches per worker role

Beads

Issue tracking and task assignment

Architecture

[C]

Conductor

Orchestrates the overall development flow. Plans work, assigns tasks to workers, monitors progress, and merges results.

spawn assign merge monitor

[G]

Gatekeeper

Reviews and validates work before it reaches the main branch. Enforces quality standards, runs tests, and approves merges.

review approve reject test

[P]

Planner

Analyzes requirements, breaks down features into tasks, estimates complexity, and creates the execution roadmap.

plan decompose estimate prioritize

Planner decomposes → Conductor assigns → Workers execute → Gatekeeper reviews

Key Features

Role-Based Access

Each Claude worker session gets a defined role with specific permissions. No worker can exceed its role boundaries.

Coordinated Sessions

Multiple Claude sessions work in parallel on different aspects of a codebase, coordinated through the role hierarchy.

Pure Shell Implementation

Zero external dependencies beyond bash and tmux. Runs anywhere Unix runs, with no compilation or package management.

Quality Gates

The Gatekeeper role ensures code review happens before merging, maintaining standards across multi-agent output.

Task Decomposition

The Planner role breaks large features into parallelizable subtasks, maximizing throughput across workers.

Beads Integration

Every task is tracked as a Beads issue, providing audit trails and progress visibility across the hive.

Stats

Defined Roles

100%

Shell Scripts

5 MB

Repository Size

Dependencies

Lessons Learned

Multi-agent coordination patterns -- how to partition work across AI sessions without conflicts or duplicated effort
Role separation as a design constraint -- limiting what each worker can do actually improves output quality
Orchestration as shell scripts -- bash is surprisingly capable as a coordination layer when the primitives (tmux, git) are already there
Quality gates in AI workflows -- automated review roles catch issues that single-agent workflows miss entirely