git-notes-memory CLAUDE.md

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

git-notes-memory is a Python library and Claude Code plugin that provides git-native, semantically-searchable memory storage. Memories are stored as git notes with YAML front matter and indexed in SQLite with sqlite-vec for vector similarity search.

Development Commands

# Install with dev dependencies
uv sync

# Run all tests
make test

# Run tests with coverage (80% minimum required)
make coverage

# Run a specific test file
uv run pytest tests/test_capture.py -v

# Run a single test
uv run pytest tests/test_capture.py::TestCaptureService::test_capture_basic -v

# Skip slow tests
uv run pytest -m "not slow"

# Run all quality checks (format, lint, typecheck, security, tests)
make quality

# Individual quality checks
make format     # Auto-fix formatting
make lint       # Ruff linting
make typecheck  # mypy strict mode
make security   # bandit security scan

# Version management
make version    # Show current version
make bump       # Bump patch version (0.3.1 → 0.3.2)
make bump-minor # Bump minor version (0.3.1 → 0.4.0)
make bump-major # Bump major version (0.3.1 → 1.0.0)
make bump-dry   # Preview version bump changes

Architecture

Service Layer Pattern

The codebase uses a singleton service factory pattern with lazy initialization:

__init__.py          # Lazy imports via __getattr__ to avoid loading embedding model at import
    └── get_capture_service()  → CaptureService
    └── get_recall_service()   → RecallService
    └── get_sync_service()     → SyncService

Services are exposed through factory functions (get_*_service()) that return singleton instances. Internal modules use get_default_service() naming.

Core Data Flow

Capture:
  CaptureService.capture()
      → validate (namespace, summary ≤100 chars, content ≤100KB)
      → serialize_note() (YAML front matter + body)
      → GitOps.append_note() (atomic append to refs/notes/mem/{namespace})
      → EmbeddingService.embed() (sentence-transformers, graceful degradation)
      → IndexService.insert() (SQLite + sqlite-vec)

Recall:
  RecallService.search()
      → EmbeddingService.embed(query)
      → IndexService.search_vector() (KNN via sqlite-vec)
      → Memory objects with distance scores