🎻 Symphony

Symphony is my personal, AI-native delegation system. It's based on my working model of Understand, then accelerate - optimized for keeping me in the loop on important decisions and ensuring I maintain a solid mental model of what I'm building, while allowing me to deploy AI to work faster when the main bottleneck becomes writing code.

Core values

• No AI slop: Succinct docs. Clean code. Output must meet my quality bar always.
• Human-in-the-loop: I review, decide, and course-correct. Can easily move up or down the stack depending on what we're working on.
• Self-improving: The system learns and adapts to my taste over time.
• Fast feedback loops: I know quickly if something's working or needs intervention, and making improvements to the system is low-friction.

Installation

1. Set up the Symphony home directory:

   ./scripts/setup

   This creates ~/.symphony/ with:

   • memory/diary/ — session diary entries
   • memory/reflections/ — synthesized insights
   • AGENTS.md — agent-agnostic preferences file

2. Install commands to your agent(s):

   ./scripts/sync

   Currently supports Claude Code and Cursor.

Dependencies

Currently relies on using Linear via MCP for issue tracking. I've been trying beads but I don't like having issue-tracking coupled with version control.

Commands

Command	Description
/plan	Turn a seed idea into a PRD in your preferred format
/work	Execute a plan in small steps, producing clean PRs
/review	Surface high-signal questions and risks for code review
/compound	Self-improvement—collect and propose system updates
/diary	Create a structured diary entry from the current session
/reflect	Analyze diary entries to identify patterns

Sources & Inspirations

I've done my best to credit any prompts, snippets, or techniques that are not mine, though inspiration comes from many places and this list is not perfect.

Inspired by:

• Compounding Engineering Plugin by Every — the original inspiration for Symphony's core workflow
• claude-diary by Lance Martin — diary + reflect system for session memory and pattern synthesis
• ai-dev-tasks by snarktank — PRD → task list → implementation workflow
• Geoffrey Litt's "tutorial doc" workflow