Claude Lore

Two artifacts that keep Claude Code grounded in your multi-project codebase: a lean index Claude loads each session, and a rich HTML report humans actually read.

What it does

Claude Lore scans every project in your codebase, maps the relationships between them, and generates two things:

• CLAUDE.md — a lean index that Claude Code loads automatically. Projects table, external services, a small cross-project diagram, key locations. Optimized for fast loading, not human browsing.
• architecture.html — a single self-contained HTML file with the full architecture documentation: system overview, diagrams, data flows, per-project deep dives, external services, patterns. Open it in a browser. This is what humans read.

Tickets created by /lore-ticket re-explore the live code each time, so they're always grounded in current reality.

Quick Start

git clone https://github.com/tinzeljar/claude-lore.git architecture
cd architecture && claude
/lore-init

What you get

my-company/
├── architecture/
│   ├── CLAUDE.md          # Lean index, auto-loaded by Claude
│   ├── architecture.html  # Rich human-facing architecture report
│   └── tickets/           # Flat folder, one ticket per file
│       └── add-rate-limiting.md
├── api/
├── frontend/
└── ...

The HTML report holds the rich documentation; the lean CLAUDE.md is for Claude.

What you can do after init

/lore-ticket rate limiting for the API

Re-explores the live code across affected services and writes an architectural ticket optimized for plan mode. A senior-architect agent then reviews it against the live code.

Open architecture.html

Browse the full architecture: diagrams, data flows, per-project deep dives, external services.

"How does authentication work?"

Claude has the lean index and re-explores the actual code to answer.

"Add user notifications to the platform"

Claude knows which projects to look at and traces the impact across them.

The intended ticket workflow

1. Run /lore-ticket "the feature description" inside the architecture folder.
2. The skill re-explores the live code in every affected project and writes one shared architectural ticket at tickets/[feature-name].md. The ticket-validator agent reviews it.
3. Open Claude Code in plan mode inside each affected service repo and paste the whole ticket. Each repo's plan mode produces the per-service implementation plan with full cross-service context plus its own local context.

That's why tickets aren't split per service — every service benefits from seeing the whole shared brief.

Skills

Skill	Description
/lore-init	Scan all projects, write the lean CLAUDE.md and the rich architecture.html
/lore-update	Re-scan and regenerate both artifacts
/lore-ticket [description]	Create an architectural implementation ticket

Supported tech stacks

Marker	Stack
package.json	Node.js / TypeScript
*.csproj / *.sln	.NET / C#
pyproject.toml / requirements.txt	Python
pom.xml	Java / Maven
build.gradle	Java / Kotlin / Gradle
go.mod	Go
Cargo.toml	Rust
Gemfile	Ruby
composer.json	PHP
Dockerfile	Containerized services
nx.json / turbo.json / lerna.json	Monorepos

License

MIT