Shura
A multi-agent orchestration system for coordinating work across multiple repositories simultaneously. Inspired by the Arabic consultative council concept — each repository gets its own autonomous Product Owner who leads a specialist team, while a Program Manager coordinates all teams toward a shared mission.
Installation
claude plugin install https://github.com/mm0rsy/shura
Requires Claude Code. After installing, the /shura, /init, /add-repo, /goal, /get-manager, /status, and /recover commands are available in any Claude Code session.
Optional: graphify (strongly recommended)
graphify builds a knowledge graph of each repo so agents can understand the full codebase in a single read — architecture, key files, entry points, dependencies — instead of exploring blind with dozens of find/cat calls.
uv tool install graphifyy && graphify install
Install before running /add-repo. If graphify is present when you add a repo, it runs automatically and the resulting GRAPH_REPORT.md is injected into every agent's context at dispatch time. Without graphify, agents still work — they just explore the codebase manually, which is slower and less precise.
Optional: External Skill Repos (recommended for specialist teams)
Shura integrates with external Claude Code skill plugins to give your agent teams domain expertise. When installed, Product Owners can hire specialists backed by proven skill prompts rather than generic instructions.
claude-skills — 313+ engineering, product, ML, and compliance skills across QA, RAG architecture, ML pipeline, DevOps, and more.
everything-claude-code — 232 skills, 60 specialized agents, TDD workflows, architecture review, security scanning, database design.
Install each as a Claude Code plugin:
claude plugin install https://github.com/alirezarezvani/claude-skills
claude plugin install https://github.com/affaan-m/everything-claude-code
Then declare them when you run /init — Shura will use them to build the specialist catalogue for each repo's stack. Without skill repos, teams consist of Developer only.
How Shura Works
The core idea
You describe a mission. Shura splits the work across your repositories — each repo gets its own Product Owner (PO) agent who runs a full stakeholder meeting with you, breaks the epic into tasks, hires specialists as needed, and pushes the finished branch. A Program Manager (PM) coordinates all POs and is your only touchpoint throughout.
The dispatch chain
/goal → stakeholder meeting (inline, you + PM)
→ epics confirmed
→ [PO dispatched per repo] ──── [PO dispatches Developer(s)]
└── [PO dispatches Specialists on-demand]
(Tester, Architect, Security Reviewer, ...)
Every agent receives its full context at dispatch time — repo path, branch, epic, decisions log, knowledge graph path, skill catalogue. Agents do not share state files directly; context is injected into each prompt.
What runs where
| Layer | Who | Visibility |
|---|
| User session | Program Manager (inline) | You talk to PM directly |
| Per-repo terminal | Product Owner | One visible terminal per repo |
| Below PO (hidden) | Developer(s), Specialists | Subagents — not separately visible |
Commands
| Command | When to use |
|---|
/init | First step — create the project directory and configure skill repos |
/add-repo | Add each repository (local worktree or remote clone); auto-detects stack; auto-indexes with graphify if installed |
/goal | State the mission; interactive stakeholder meeting with PM; auto-launches all teams |
/get-manager | Talk to the Program Manager at any time — check progress, give new direction |
/status | Live dashboard — every repo's status, branch, epic, and recent decisions |
/recover | Re-launch teams manually after a failure or incomplete /goal run |
Agent Hierarchy
User ─── /get-manager ──► Program Manager (PM)
│
┌───────────────┼───────────────┐
▼ ▼ ▼
Product Owner Product Owner Product Owner
(repo A) (repo B) (repo C)
│
┌─────────┼──────────┐
▼ ▼ ▼
Developer Tester Architect
(on-demand specialists — hired by PO when task warrants it)
Communication rules
