A runtime governance layer for AI coding agents.
Engineers choose their preferred editor. They also choose their preferred AI. But today, each AI tool lives in its own silo — separate instruction files, separate context, separate memory. Meta Agent sits above all of them as a unified layer.
Every AI coding tool has invented its own instruction format: CLAUDE.md, AGENTS.md, .cursorrules, and more. Existing solutions (rulesync, ai-rulez, etc.) attempt to bridge these by generating static files. They are file converters. They do not solve the deeper problems:
- State dies between sessions. Switch tools or close a terminal, and all context is gone.
- Rules are static. No tool applies different rules based on what you're actually doing.
- No agent coordination. When multiple AI agents work on the same project, there is no governance.
State Manager — Persistent, three-tier memory (session / project / knowledge) that survives across sessions, tools, and time.
Context Compiler — Assembles the right context for the right agent at the right moment, fitted to its context window.
Instruction Hub — One entry point for all AI instructions. Non-destructive: works alongside existing files or replaces them. Your choice.
Rule Engine — Runtime rule resolution with types (constraint > preference > context > skill), scoping, and priority.
Consensus Engine — Multi-agent decision patterns: dictator, validator, quorum, pipeline, specialist, auction.
Optimizer — User-controlled tradeoff between accuracy, cost, and speed.
Existing tools generate files. Meta Agent governs execution.
Existing tools: Static file generators
Meta Agent: Runtime governance layer
It does not compete with AGENTS.md or CLAUDE.md. It sits on top of them.
- PRODUCT.md — Architecture, philosophy, and competitive analysis
- ROADMAP.md — Phased delivery plan
AGPL-3.0