Skip to content

qubicDB/skills

Repository files navigation

QubicDB Agent Skills

Brain-like organic memory for AI agents and IDE assistants — powered by QubicDB.

Includes ready-to-use configurations for Windsurf, Cursor, VS Code (GitHub Copilot), and Claude Code. Also available as a Claude Code Plugin via marketplace.


Quick Start

1. Start QubicDB

Option A: Non-vector Docker Compose

docker compose -f docker-compose.qubicdb.yml up -d

Option B: Vector Docker Compose

Clone qubicdb and skills side-by-side:

your-folder/
├── qubicdb/
└── skills/

Then run:

docker compose -f docker-compose.qubicdb.vector.yml up -d --build

docker-compose.qubicdb.vector.yml expects the vector model at ../qubicdb/dist/MiniLM-L6-v2.Q8_0.gguf.

Important: Download the GGUF model first. Full download commands are documented in SETUP.md.

Helper instructions and local patch scripts are also available in the main qubicdb repo:

Option C: Bundled Vector Docker Compose

docker compose -f docker-compose.qubicdb.bundled.yml up -d

This lane uses qubicdb/qubicdb-bundled:latest, which includes the example MIT-licensed MiniLM-L6-v2.Q8_0.gguf model (about 25 MB). No local build or dist mount is required.

Option D: Docker Run (non-vector)

docker pull qubicdb/qubicdb:latest
docker pull qubicdb/qubicdb-ui:latest

docker network create qubicdb-net

docker run -d \
  --name qubicdb \
  --network qubicdb-net \
  -p 6060:6060 \
  -v qubicdb_data:/app/data \
  -e QUBICDB_HTTP_ADDR=:6060 \
  -e QUBICDB_DATA_PATH=/app/data \
  -e QUBICDB_ADMIN_ENABLED=true \
  -e QUBICDB_ADMIN_USER=admin \
  -e QUBICDB_ADMIN_PASSWORD=changeme \
  -e QUBICDB_MCP_ENABLED=true \
  -e QUBICDB_MCP_PATH=/mcp \
  -e QUBICDB_MCP_STATELESS=true \
  -e QUBICDB_MCP_RATE_LIMIT_RPS=30 \
  -e QUBICDB_MCP_RATE_LIMIT_BURST=60 \
  -e QUBICDB_MCP_ENABLE_PROMPTS=true \
  -e QUBICDB_MCP_API_KEY=qubicdb-mcp-secret-key \
  -e QUBICDB_ALLOWED_ORIGINS=http://localhost:6060,http://localhost:8080,http://127.0.0.1:8080 \
  -e QUBICDB_REGISTRY_ENABLED=false \
  qubicdb/qubicdb:latest

docker run -d \
  --name qubicdb-ui \
  --network qubicdb-net \
  -p 8080:80 \
  qubicdb/qubicdb-ui:latest

📖 For full non-vector setup, vector setup, vector parameters, and troubleshooting, see SETUP.md.

Verify:

curl http://localhost:6060/health

Admin UI: http://localhost:8080 — login with admin / changeme.

Note: The QUBICDB_MCP_API_KEY must match the X-API-Key header in your IDE's MCP config. Change both if you use a custom key.

2. Install skills

Option A: Claude Code Plugin (recommended for Claude Code)

/plugin marketplace add qubicDB/skills
/plugin install qubicdb-skills@qubicdb-agent-skills

Option B: Copy IDE folder — pick your IDE below, copy the relevant folder into your project root.

3. Start using

Start a conversation — the AI will use QubicDB as persistent memory automatically.


Windsurf

Copy .windsurf/ into your project root. Add MCP server in Windsurf settings (or use mcp_config.json as reference):

your-project/
└── .windsurf/
    ├── rules/
    │   └── qubic.md              ← always-on rules (191 lines)
    ├── skills/
    │   └── qubic/
    │       └── SKILL.md          ← skill definition (140 lines)
    └── workflows/
        ├── qubic-init.md         ← /qubic-init workflow
        ├── qubic-search.md       ← /qubic-search workflow
        └── qubic-write.md        ← /qubic-write workflow

MCP config for Windsurf settings:

{
  "mcpServers": {
    "qubicdb": {
      "serverUrl": "http://localhost:6060/mcp",
      "headers": {
        "X-API-Key": "qubicdb-mcp-secret-key"
      }
    }
  }
}

Cursor

Copy .cursor/ into your project root:

your-project/
└── .cursor/
    ├── mcp.json                  ← MCP server config
    └── rules/
        ├── qubic-rules.mdc      ← always-on rules (full port of Windsurf rules)
        ├── qubic-skill.mdc      ← skill definition (full port of Windsurf skill)
        ├── qubic-init.mdc       ← init workflow
        ├── qubic-search.mdc     ← search workflow
        └── qubic-write.mdc      ← write workflow

VS Code (GitHub Copilot)

Copy .vscode/ and .github/ into your project root:

your-project/
├── .vscode/
│   └── mcp.json                  ← MCP server config
└── .github/
    └── copilot-instructions.md   ← all rules + skills + workflows combined

Claude Code

Copy .claude/ into your project root:

your-project/
└── .claude/
    ├── CLAUDE.md                     ← project rules (always loaded)
    ├── settings.local.json           ← MCP server config
    └── skills/
        ├── qubic/
        │   └── SKILL.md              ← main skill definition
        ├── qubic-init/
        │   └── SKILL.md              ← /qubic-init skill
        ├── qubic-search/
        │   └── SKILL.md              ← /qubic-search skill
        └── qubic-write/
            └── SKILL.md              ← /qubic-write skill

What's Included

Per-IDE Files

IDE MCP Config Rules/Skills
Windsurf mcp_config.json (add in settings) .windsurf/rules/ + .windsurf/skills/ + .windsurf/workflows/
Cursor .cursor/mcp.json .cursor/rules/*.mdc (5 files)
VS Code .vscode/mcp.json .github/copilot-instructions.md
Claude Code .claude/settings.local.json .claude/CLAUDE.md + .claude/skills/ (4 skills)

Content (identical across all IDEs)

Content Description
Rules Always-on behavior: session lifecycle, forced behaviors (no fabrication, auto-write triggers), persona, naming conventions
Skill Architecture (index/thread/neuron model), MCP tool reference, metadata types, usage examples
Init workflow Register project brain index, load existing memories
Search workflow Spreading activation search, metadata filters, parameters
Write workflow Store decisions, preferences, todos, facts with metadata

How It Works

QubicDB gives your AI assistant a persistent, organic memory across conversations:

  • One brain per projectbrain-{project} index, created once, reused forever
  • One thread per conversationconv-{uuid} groups neurons within a session
  • Hebbian learning — memories that fire together, wire together
  • Spreading activation search — finds related memories through synapse connections
  • Lifecycle states — Active → Idle → Sleeping → Dormant with automatic consolidation

Claude Code Plugin (Marketplace)

This repo is a Claude Code Plugin Marketplace. Install directly:

# Add the marketplace
/plugin marketplace add qubicDB/skills

# Install core skills (MCP memory: write, search, recall, context)
/plugin install qubicdb-skills@qubicdb-agent-skills

# Install experimental skills (multi-index, agent-to-agent, sentiment, REST API, RAG)
/plugin install qubicdb-experimental@qubicdb-agent-skills

Plugin structure:

plugins/
├── qubicdb-skills/              ← core plugin
│   ├── .claude-plugin/plugin.json
│   ├── .mcp.json                ← auto-configures MCP server
│   └── skills/
│       ├── qubic/SKILL.md
│       ├── qubic-init/SKILL.md
│       ├── qubic-search/SKILL.md
│       └── qubic-write/SKILL.md
└── qubicdb-experimental/        ← experimental plugin
    ├── .claude-plugin/plugin.json
    └── skills/
        ├── multi-index-research/SKILL.md
        ├── agent-to-agent/SKILL.md
        ├── conversation-chains/SKILL.md
        ├── sentiment-journal/SKILL.md
        ├── knowledge-base-server/SKILL.md
        └── rag-context-assembly/SKILL.md

New in v1.1.0: Cross-Index MCP Tools

QubicDB now supports global and multi-index operations via MCP:

Tool Description
qubicdb_list_indexes List all registered indexes with stats
qubicdb_global_search Search across ALL active indexes
qubicdb_multi_search Search across a specific list of indexes
qubicdb_recent_indexes Get most recently active indexes

Use case: Agents can now discover relationships across hundreds of repositories/indexes — perfect for monorepo insights and cross-project memory.


Experimental Skills

The qubicdb-experimental plugin showcases diverse use cases beyond IDE memory:

Skill Description
multi-index-research Separate brains per research domain, cross-reference between them
agent-to-agent Multiple agents sharing memory through a common brain with role metadata
conversation-chains Track multi-session conversations with thread_id chaining and parent links
sentiment-journal Emotion-aware journaling leveraging built-in VADER sentiment analysis
knowledge-base Build a persistent knowledge base with rich metadata and hybrid search
rag-context-assembly Token-budgeted context assembly using spreading activation (not just vector similarity)

Each skill includes full Prerequisites with Docker setup and MCP config for all 4 IDEs.


Links

About

Skills for qubicdb

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors