Changelog

All notable changes to Chronicle are documented here.
Format follows Keep a Changelog.
Versioning follows Semantic Versioning.

Unreleased

Phase 4: Tool adapters (Cursor .cursorrules, Aider --read, Gemini GEMINI.md, Copilot instructions)
Phase 5: Mermaid diagram auto-generation (architecture, dependency graph, evolution timeline)
Phase 6: Evolution records and system milestone timeline
Phase 7 (v2): Semantic clustering extraction strategy (group commits by file overlap + time proximity)
Phase 8 (v3): Two-pass extraction (cheap LLM filter → quality model for complex decisions)

Initial release of Chronicle — AI-native development memory.

Core (@chronicle/core)

store.ts — file-based markdown store with findLoreRoot (walks up like git), readStore, writeStore, appendToStore, writeDeepDecision, initStore
scanner.ts — git history scanner with noise filtering (chore/style/docs/test prefixes), diff size threshold (≥20 changed lines), diff capping (4000 chars), tag detection
extractor.ts — LLM extraction engine with pluggable strategy pattern; v1 simple strategy (batches of 6, ≤5000 chars); clustered and two-pass slots reserved for v2/v3
cache.ts — SHA-keyed JSON file cache; prevents reprocessing commits across runs
ExtractionCache interface for swappable cache backends (in-memory, file, future SQLite)

CLI (chronicle-dev)

chronicle init [--depth] — bootstraps .lore/ from git history; progressive scan defaults to 6 months
chronicle inject [--files] [--full] [--format] — outputs compressed context to stdout; scopes to relevant files; supports markdown, xml, plain formats
chronicle deepen [--depth] — extends scan further back without reprocessing cached commits
chronicle hooks install/remove — installs post-commit (async decision capture) and prepare-commit-msg (risk annotation) git hooks
Internal chronicle capture and chronicle enrich-commit commands (invoked by hooks)

MCP Server (@chronicle/mcp)

chronicle_get_context — injects compressed project context; scopes to files if specified
chronicle_log_decision — AI logs architectural choices mid-session
chronicle_log_rejection — AI logs abandoned approaches (crown jewel: prevents future repetition)
chronicle_get_risks — returns blast-radius info before touching a file
chronicle_save_session — summarizes session to .lore/sessions/YYYY-MM-DD.md

Python wrapper (chronicle-dev on PyPI)

Project

.lore/ store structure: index.md, decisions.md, decisions/, rejected.md, risks.md, evolution.md, diagrams/, sessions/
Hub-and-spoke ADR model: shallow decisions inline, complex decisions in decisions/<slug>.md
ARCHITECTURE.md — full system design documentation
GitHub Actions CI (test on every push/PR) and Release (publish npm + PyPI on tag)
Vitest test suite for core package (store, extractor, scanner)
pytest suite for Python wrapper

Markdown-only store: no vector DB, no embeddings — plain files readable by any LLM
Commit SHA cache ensures bootstrap is idempotent and cost-efficient
Strategy pattern in extractor: v1 ships today, v2/v3 are drop-in replacements
Python package delegates to Node — single source of truth, no duplication