Skip to content

ORCHESTRATION.md

Latest commit

 

History

History
149 lines (118 loc) · 5.22 KB

ORCHESTRATION.md

File metadata and controls

149 lines (118 loc) · 5.22 KB
Raw

Enterprise Orchestration Framework

Multi-agent orchestration with state management, team heirarchy, coordination & observability.

Uses claude-code agents, commands & hooks with mcp & system tools.

Components

Agents

  • Task-specific agent definitions for efficient and effective collaboration
  • Agents have specific system prompts, tool access, and recieve context at runtime from orchestrator and shared state management system.

Teams

  • Teams are composed of multiple agents. Some are responsible for research and documentation tasks, others on feature implementation, others on review/testing/reporting.
  • Every team has one Orchestrator agent which has the ability to initialize sprints, assign tasks, kick off sub-agents and workflows.

Hooks (State Management System)

  • Shared state & context using JSON or MCP
    • Active tasks, Task Queue, Dependencies
    • Bugs Intake, handoff & observe
    • Cross-agent/team Communication: Questions, Findings, Coordination, Handoffs
    • Locked files, Worktrees
  • Logs

Output Styles

  • JSON / Yaml / XML
  • HTML / React (bun create)
  • Python Script (uv run)
  • Markdown Readme / CC Template

Status Line

  • Team
  • Current Epic
  • Current Sprint
  • Active / Queued / Completed Tasks in Sprint
  • Test Coverage / Failing / Passing
  • Open Issues
  • Active Agents (Color Dots ?)

Teams

Each team runs in it's own claude-code instance in a terminal. A team is initialized with a slash command that specifies the team, epic and sprint to begin work on.

Example: .claude/commands/team/start.md

  • /teams:start <team> [options]
  • /team:start product --epic=2 --sprint=0

Uppon running the command, the root claude-code instance assumes the role of team orchestrator. If no options are passed, the orchestrator checks the state for active or next pending sprint & spawns specified agents / tasks.

Product

  • CPO / Product Director (Orchestrator)
  • Team Strategy & Coordination manager (Track Epic / Sprint Process & State, Cross-team Collaboration & Observability)
  • Product Manager (Feature Requirements / PRD, Feature Review & Signoff)
  • Business Analyst (Research & Design Business Logic, Analyze & Review Implementation)
  • Data Scientist (Research & Design Data Models, Analyze & Review Implementation)
  • Deep Research / Investigation (Parallel Web Search & Summary)
  • Team Analytics & Reporting (Observe Logs, Analytics, Reports)

Marketing

  • Director of Marketing (Orchestrator)
  • Content Strategist (Research, Content Inventory)
  • SEO Researcher (SEO Research & Strategy)
  • SEO Engineer (Implement SEO)
  • SEO Analyst (Track Performance & Write Reports)
  • Copywriter (Write Specific Copy)

Creative

  • Creative Director (Orchestrator)
  • UX Lead (Design System -- Colors, Fonts, Spacing)
  • Wireframe designer (Image gen)
  • Photographer (Image gen)
  • Illustrator (Image gen)
  • Logo designer (Image gen)

Engineering

  • CTO / Director of Engineering (Orchestrator)
  • Engineering Manager (Track Epic / Sprint Process & State, Team Observability, Sign off on sprints)
  • Documentation Research Specialist (Web Search / Scrape / Summarize)
  • Tech Lead / Architect (Write Specs & Data Models)
  • Fullstack Engineer (Business Logic E2E Implementation)
  • UX Engineer (Components, Layouts, Design System, Desktop/Mobile)
  • API Engineer (Backend Business Logic from Spec)
  • Test Engineer (Write & Maintain Valid tests from Spec)
  • Documentation Writer (Write & Maintain Project Docs)
  • Tech Lead (Code review, Consistency Check, Test Passing)

QA

  • QA Director (Orchestrator)
  • QA Engineer (e2e testing)
  • QA Engineer (Write test scripts, data, etc)
  • QA Analyst (Reports, Issue tracking)

DevOps

  • DevOps Manager (Orchestrator)
  • CI/CD Engineer (Github Actions, Scripts)
  • Infrastructure Engineer (Containerization, Infra as Code)
  • Cleanup Engineer (Delete extra tests, organize files)
  • Release Manager (Release Notes, Changelog)

Project Flow & Task Delegation

The purpose of this system is to mirror that of a real software company, composed of teams with internal heirarchy and specialized contributors. Additionally, this system must implement each aspect of the SDLC and have specific commands & workflows for each phase of development.

SLDC

  • Discovery / Planning / Requirements
  • Specs / Data Model / Designs / User Stories
  • Epics / Sprints / Tasks / Deliverables & Dependencies
  • Implementation Loop (Epics, Sprints, Bugs)
  • Monitoring, Reports
  • Releases, Issue tracking, CI/CD

Workflows

  • Brainstorm Loop (Leadership)
  • Project spec > Project SDLC
  • Feature spec > Feature SDLC
  • Epic (Plan, Start, Status)
  • Sprint (Plan, Start, Status, Continue, Next)
  • Task (Start, Status, Fix, Dependencies)
  • Bugs (Review, Status, Fix)
  • Release (New, Status, Schedule)
  • Code Review & Consistency Check
  • End-to-end Test Suite
  • Security Audit, Compliance, Accessibility
  • Update Docs, User Stories
  • Generate Test Data, Test Scripts
  • Infinite Agent loop 🤔 How ?

Questions & Ideas

  • Cross agent coordination, memory (Hooks)
    • Epics, Sprint, Tasks
    • State, Logs, Queue, Dependencies & Handoff
    • Create specific Mcp tools?
  • Team > Orchestrator > Agent
  • How can we implement a real schedule ? Burndown ? Reporting ?
  • When does it make most sense to use worktrees?
    • For each feature? Per team? Per Sprint?