lore

Layered Orchestration for Rules and Extensions

A small tool for managing the skills and behaviors your AI agents use. Clone a repo, run one command, and the skills are live. There's no package registry behind it and nothing to lock you in — under the hood it's just symlinks.

Why it exists

If you've tried npx skills add on a heavy repo, you know it tends to choke. And once you're running more than one agent, your config ends up scattered: a bit in ~/.claude/, a bit in ~/.cursor/, more somewhere else. You either install everything twice or quietly let the setups drift apart. There's no agreed-on home for agent rules, and no clean way to mix behaviors coming from different repos.

lore's answer is boring on purpose: one directory, ~/.agents/, that every tool points at.

Install

Grab the latest stable binary:

curl -fsSL https://raw.githubusercontent.com/rjfranc0/lore/main/install.sh | bash

Or clone and let the installer build from source (requires cargo):

git clone https://github.com/rjfranc0/lore
cd lore && ./install.sh

To install the latest pre-release from the dev channel:

curl -fsSL https://raw.githubusercontent.com/rjfranc0/lore/main/install.sh | bash -s -- --dev

Either way, bootstrap once when you're done:

lore init

A quick taste

Say you've cloned a repo full of skills and behaviors. You install them from wherever they live:

git clone https://github.com/someone/ai-skills ~/repos/ai-skills

cd ~/repos/ai-skills/skills
lore install coding-style type-conventions

cd ~/repos/ai-skills/behaviors
lore behavior add strict-review

lore list

That's the whole loop: cd into a repo, install what you want, check it with list.

How it actually works

After lore init, your layout looks like this:

~/.agents/
├── AGENTS.md          ← universal agent config (managed by lore)
├── skills/            ← symlinks → skill dirs in your repos
└── behaviors/         ← symlinks → behavior dirs in your repos

~/.claude/
├── LORE.md            ← @~/.agents/AGENTS.md  (written by lore init)
├── CLAUDE.md          ← @~/.claude/LORE.md  (lore only ensures this one line)
└── skills/            ← symlink → ~/.agents/skills/

The important part: your skills and behaviors never leave their source repos. lore just creates symlinks that point at them. A skill is "installed" when its symlink exists and "uninstalled" when you remove it — the original files are never moved or modified.

Behaviors work the same way, with one extra step. When you run lore behavior add, lore finds the behavior's entry .md file and appends it to ~/.agents/AGENTS.md as a Claude @import. From then on, Claude loads it at the start of every session.

Commands

Command	What it does
lore init [--account <name>]	Bootstrap ~/.agents/ and wire up a Claude account
lore install <skill> [...] [--account <name>]	Install skill(s) from the current directory
lore remove <skill> [...] [--account <name>]	Uninstall skill(s)
lore behavior add <name> [...] [--account <name>]	Install behavior(s) and update AGENTS.md/LORE.md
lore behavior remove <name> [...] [--account <name>]	Remove behavior(s)
lore accounts list	Show every registered Claude account
lore accounts remove <name>	Forget an account (registry only, nothing on disk)
lore accounts sync	Re-wire any account that's broken on disk
lore sync	Reconcile AGENTS.md from disk
lore update <name> [--path <path>] / --all	Re-link a skill/behavior whose source repo moved
lore list	Show everything that's installed
lore version	Print the version
lore help	Full manual

The full write-up of every command and file format lives in docs/index.md.

Already have a Claude setup?

You don't have to clean anything up first. lore init notices existing content in ~/.claude/CLAUDE.md and migrates it for you, surgically:

• A copy of your old instructions moves to ~/.agents/behaviors/from-claude/RULES.md
• Any skills sitting in ~/.claude/skills/ move into ~/.agents/skills/
• ~/.claude/CLAUDE.md keeps its original content exactly as it was — lore only appends a @~/.claude/LORE.md import line underneath it

Nothing gets thrown away or overwritten, and lore prints exactly where each piece landed, plus any @import lines left over from other tools so you know they're still there, untouched. If a skill name would collide with one you already have, lore stops before changing anything and tells you what to resolve — better a clear halt than a half-migrated mess. And if you ever delete AGENTS.md by accident, re-running lore init rebuilds it from the behaviors still on disk.

More than one Claude account

~/.agents/ is shared, but you can wire it into more than one Claude config directory — say, a personal ~/.claude/ and a work ~/.claude-work/:

lore init --account work
lore accounts list
lore accounts sync   # re-wire anything that's drifted or broken on disk

Every account gets its own LORE.md, but they all import the same shared AGENTS.md — install a skill or behavior once, scope it to an account with --account <name> if you don't want it everywhere.

Config file

lore's own settings live in ~/.config/lore/lore.toml — just agents_dir and the accounts registry. It's created automatically the first time you register something; you won't normally touch it by hand.

Point lore at a different config entirely with LORE_CONF — handy for testing without going near your real setup:

LORE_CONF=/tmp/test-lore.toml lore init

Tests

The test suite uses cargo test — no extra tooling required:

cargo test

Tests run against throwaway temp directories, so they never touch your real setup.

Roadmap

• Integrations for other tools (Cursor, Windsurf, Zed)

License

MIT