skillmind 0.1.0

Active Skill Listener & Trainer — structured memory layer for AI coding assistants

SkillMind

Active Skill Listener & Trainer — the structured memory layer for AI coding assistants.

Replace flat markdown memory files with a vector-database-backed system that listens, learns, and surfaces only relevant context per conversation.

🌐 Visit skill-mind.com — Documentation, setup guides, and live demos.

---

The Problem

Current state	SkillMind
15+ flat markdown memory files	One vector DB with semantic search
Everything loaded into context	Only relevant memories surfaced
Manual save/forget	Auto-listen, auto-classify, auto-expire
No deduplication	Automatic dedup + consolidation
No protection	Sanitizer redacts API keys, PII before storing

Quick Start (3 Commands)

# 1. Install
pip install skillmind[pinecone]

# 2. Initialize
skillmind init --backend pinecone

# 3. Import your existing Claude Code memories
skillmind import ~/.claude/projects/*/memory/

That's it. Your memories are now in a vector DB with semantic search.

How It Works

┌────────────┐    ┌────────────┐    ┌─────────────────┐
│  LISTENER   │───▶│  TRAINER    │───▶│  VECTOR STORE    │
│             │    │             │    │                  │
│ • Git hooks │    │ • Classify  │    │ • Chroma (local) │
│ • File watch│    │ • Sanitize  │    │ • Pinecone       │
│ • Convos    │    │ • Dedup     │    │ • Supabase       │
│ • YouTube   │    │ • Merge     │    │ • Qdrant         │
│ • Screen    │    │ • Expire    │    │ • FAISS          │
└────────────┘    └────────────┘    └─────────────────┘
                                            │
                                            ▼
                                    ┌─────────────────┐
                                    │  CLAUDE CODE     │
                                    │  (MCP Server)    │
                                    │  14 tools        │
                                    └─────────────────┘

5 Vector Store Backends

Backend	Best for	Requires
ChromaDB	Solo dev, local, default	pip install skillmind[chroma]
Pinecone	Multi-device cloud sync	API key
Supabase	SQL + vectors, team sharing	Supabase project
Qdrant	Self-hosted or cloud, great filtering	Qdrant server
FAISS	Offline, air-gapped, fastest	pip install skillmind[faiss]

All backends implement the same interface — switch anytime with zero data loss.

14 MCP Tools for Claude Code

Once installed, Claude Code gets these tools:

Tool	What it does
remember	Store memory (auto-classified, sanitized, deduped)
recall	Semantic search across all memories
forget	Delete a memory
update_memory	Edit existing memory
context	Generate focused context for current file/topic
consolidate	Merge duplicates, expire stale, cleanup
memory_stats	Counts by type, backend info
list_memories	Filter by type/topic
import_markdown_memories	Bulk import from Claude Code markdown files
learn_youtube	Extract knowledge from YouTube video
learn_youtube_channel	Learn from channel's latest videos
learn_video	Learn from local video/screen recording
record_screen	Record screen to MP4
screenshot	Capture screenshot

MCP Setup (Claude Code settings.json)

{
  "mcpServers": {
    "skillmind": {
      "command": "python",
      "args": ["-m", "skillmind.mcp.server"],
      "env": {
        "PINECONE_API_KEY": "your-key",
        "SKILLMIND_BACKEND": "pinecone"
      }
    }
  }
}

Memory Types

Type	What	Auto-expires
user	Role, preferences, expertise	No
feedback	Corrections, confirmed approaches	No
project	Deadlines, client context, status	90 days
reference	External URLs, dashboards, wikis	No
skill	Patterns, workflows, how-tos	No

Learn From Video

# YouTube
skillmind learn-youtube "https://youtube.com/watch?v=..."

# YouTube channel (latest 5 videos)
skillmind learn-youtube-channel "UCxxxxx" --limit 5

# Local video file
skillmind learn-video recording.mp4

# Screen recording
skillmind record-screen --duration 60 --fps 15

Built-in Sanitizer

API keys, emails, phone numbers, IBANs, and PII are automatically redacted before storing:

Input:  "My Pinecone key is pcsk_2mrRyA_9BSbdn5i..."
Stored: "My Pinecone key is [REDACTED:PINECONE_KEY]"

Configurable allowlists, custom patterns, and name anonymization.

CLI Reference

skillmind init --backend chroma     # Initialize with backend
skillmind remember "content"        # Store a memory
skillmind recall "query"            # Semantic search
skillmind list                      # List all memories
skillmind list --type feedback      # Filter by type
skillmind forget <id>               # Delete
skillmind import                    # Import Claude Code memories
skillmind consolidate               # Cleanup & deduplicate
skillmind context                   # Generate context for current dir
skillmind stats                     # Show statistics
skillmind serve                     # Start MCP server

Installation

# Core + one backend
pip install skillmind[chroma]       # Local (default)
pip install skillmind[pinecone]     # Cloud sync
pip install skillmind[supabase]     # SQL + vectors
pip install skillmind[qdrant]       # Self-hosted
pip install skillmind[faiss]        # Offline

# Add-ons
pip install skillmind[youtube]      # YouTube learning
pip install skillmind[video]        # Screen recording + OCR
pip install skillmind[mcp]          # MCP server for Claude Code

# Everything
pip install skillmind[all]

Architecture

src/skillmind/
├── models.py          # Memory, QueryFilter, QueryResult (Pydantic)
├── config.py          # SkillMindConfig (YAML-based)
├── embeddings.py      # EmbeddingEngine (sentence-transformers | OpenAI)
├── trainer.py         # Auto-classify, dedup, merge, consolidate
├── sanitizer.py       # Redact API keys, PII before storage
├── listener.py        # GitListener, FileListener, ConversationListener
├── context.py         # ContextGenerator — dynamic context injection
├── migration.py       # Import existing Claude Code markdown memories
├── store/
│   ├── base.py            # Abstract MemoryStore interface
│   ├── chroma_store.py    # ChromaDB backend
│   ├── pinecone_store.py  # Pinecone backend
│   ├── supabase_store.py  # Supabase/pgvector backend
│   ├── qdrant_store.py    # Qdrant backend
│   └── faiss_store.py     # FAISS + JSON backend
├── video/
│   ├── youtube_learner.py     # YouTube transcript extraction
│   ├── video_learner.py       # Local video learning
│   └── screen_recorder.py     # Screen capture
├── mcp/
│   └── server.py      # FastMCP server (14 tools)
└── cli/
    └── main.py        # CLI (click-based)

Contributing

PRs welcome! See CONTRIBUTING.md.

License

MIT — see LICENSE.

---

Built by Antonio Blago | skill-mind.com