synapse

name	synapse
description	Explains and operates the SYNAPSE context engine, including domains, layers, manifest management, and context bracket adaptation
version	1.0.0
agent	synapse
user-invocable	true
maxTurns	25

SYNAPSE Context Engine

Overview

SYNAPSE (MegaBrain Adaptive Processing & State Engine) is the unified context engine for Mega Brain. It injects contextual rules into every prompt via an 8-layer processing pipeline, adapting to context window usage through bracket-aware filtering.

What it does:

• Injects rules per-prompt via Claude Code's UserPromptSubmit hook
• Processes 8 layers (L0 Constitution through L7 Star-Commands) sequentially
• Adapts injection volume based on context brackets (FRESH/MODERATE/DEPLETED/CRITICAL)
• Integrates with agent state (active agent, workflow, task, squad)
• Outputs <synapse-rules> XML block appended to each prompt

What it replaces: SYNAPSE replaces the legacy CARL system with full feature parity plus 8 new capabilities including agent-scoped domains, workflow activation, and CRUD management commands.

Architecture model: Open Core — the 8-layer engine lives in mega-brain-core (open source), memory integration is feature-gated in mega-brain-pro.

Quick Start

Verify SYNAPSE is Active

SYNAPSE runs automatically via the Claude Code hook. To check status:

*synapse status

This shows: active domains, current bracket, session info, and loaded layers.

Basic Commands

Command	What it does
*synapse status	Show current engine state
*synapse domains	List all registered domains
*synapse debug	Show detailed debug info (manifest parse, load times, rule counts)
*synapse help	Show all available synapse commands
*brief	Switch to brief response mode
*dev	Switch to developer mode (code-focused)
*review	Switch to code review mode

Create a Custom Domain

*synapse create

This walks you through creating a new domain file + manifest entry. See references/domains.md for the full domain guide.

Architecture

SYNAPSE operates as a 4-layer architecture:

.claude/hooks/synapse-engine.js          # Layer 1: Hook Entry (~50 lines)
        |
        v imports
mega-brain-core/core/synapse/                 # Layer 2: Engine Modules
|-- engine.js                            #   SynapseEngine class
|-- layers/                              #   8 layer processors (L0-L7)
|-- session/session-manager.js           #   Session state (JSON v2.0)
|-- domain/domain-loader.js              #   Manifest + domain parser
|-- context/context-tracker.js           #   Bracket calculation
|-- memory/memory-bridge.js              #   Pro-gated MIS consumer
|-- output/formatter.js                  #   <synapse-rules> XML
        |
        v reads/writes
.synapse/                                # Layer 3: Runtime Data
|-- manifest                             #   Central domain registry (KEY=VALUE)
|-- constitution, global, context        #   Core domains (L0, L1)
|-- agent-*, workflow-*                  #   Scoped domains (L2, L3)
|-- commands                             #   Star-command definitions (L7)
|-- sessions/, cache/                    #   Session state (gitignored)
        |
        v user-invoked
.claude/skills/synapse/                  # Layer 4: CRUD Skills + Skill Docs
|-- SKILL.md                             #   Router/dispatcher
|-- references/ + assets/                #   Domain docs, templates, commands

Key principle: SYNAPSE is a consumer of existing systems (UAP for session state, MIS for memories). It never rewrites code from other epics.

References

Reference Guides

Guide	Description
domains.md	Domain types (L0-L7), KEY=VALUE format, creation guide
commands.md	Star-commands, *synapse sub-commands, CRUD operations
manifest.md	Manifest format specification, all valid keys
brackets.md	Context bracket system, token budgets, layer activation
layers.md	8-layer processor architecture, priority, conflict resolution

Assets (Templates)

Templates for creating custom domains and manifest entries are maintained at:

• Domain template: .claude/skills/synapse/assets/domain-template
• Manifest entry template: .claude/skills/synapse/assets/manifest-entry-template

See assets/README.md for details.

CRUD Commands

For domain management operations, use the SYNAPSE manager:

Command	Purpose
*synapse create	Create new domain + manifest entry
*synapse add	Add rule to existing domain
*synapse edit	Edit or remove rule by index
*synapse toggle	Toggle domain active/inactive
*synapse command	Create new L7 routing definition
*synapse suggest	Suggest best domain for a rule

Full details: references/commands.md

Key Files

File	Purpose
.claude/hooks/synapse-engine.js	Hook entry point (UserPromptSubmit)
mega-brain-core/core/synapse/engine.js	SynapseEngine orchestrator
.synapse/manifest	Domain registry (KEY=VALUE)
.synapse/commands	L7 explicit mode/routing definitions
.claude/skills/synapse/SKILL.md	CRUD skill router