A Claude Code plugin that teaches you to use and learn Lore — Epic's content-addressed VCS — directly inside your AI assistant.
Community project. Not affiliated with or endorsed by Epic Games.
Lore (GitHub) is Epic's open-source, content-addressed version control system designed for both code and large binary assets. Every payload, tree node, and revision is keyed by its BLAKE3 hash, forming a Merkle DAG where a single 32-byte root hash verifies the entire tree. Workspaces are sparse by default and hydrate content on demand.
Once installed, Claude activates this skill whenever you are working with Lore. It covers:
- Setup & install — installing the Lore CLI, creating or cloning repositories, configuring authentication, running a local service, and building language bindings from source.
- Daily CLI VCS operations — status, dirty/stage/unstage, commit, branch, push, sync, diff, history, locks, and links/layers.
- Core concepts — content-addressing, the BLAKE3 Merkle DAG, offline vs. online mode, sparse workspaces,
.lore/view,.loreignore, and the Lore glossary. - Language SDKs — Python (
lore-vcs), JavaScript (lore-js), C# (lore-dotnet), and Go (lore-go). All four bind to the same C ABI; the skill covers the shared lifecycle (log_configure→ operations →shutdown) and language-specific quickstarts. - Failure diagnosis — CLI and SDK error patterns with actionable remediation steps.
In Claude Code, run:
/plugin marketplace add ABostrom/lore-skill
/plugin install lore@lore-tools
The marketplace package name is lore-tools and the plugin name is lore.
Skills activate automatically: once installed, Claude loads this skill whenever your request matches Lore work (detected from the skill's description).
skills/lore/
SKILL.md Core skill — quick-reference tables, iron rules,
detection gate, anti-patterns. Grounded against
Lore CLI 0.8.3+201.
reference/
concepts.md Architecture: content-addressing, Merkle DAG,
offline/online modes, hydration, sparse workspaces,
links vs. layers, glossary.
setup.md Installing the CLI, init/clone, local vs. server
mode, auth, building from source.
cli.md Full CLI command and flag reference for daily
VCS work.
troubleshooting.md CLI and SDK failure patterns with remediation.
sdk/
model.md Universal SDK model: globals/op-args, the
op(globals, args).callback().wait() lifecycle,
event system, logging, shutdown.
python.md Python SDK (lore-vcs) quickstart + worked example.
javascript.md JavaScript SDK (lore-js) quickstart + worked example.
dotnet.md C# SDK (lore-dotnet) quickstart + worked example.
go.md Go SDK (lore-go) quickstart + worked example.
scripts/
validate_plugin.py Validation harness (see Development below).
README.md Script documentation.
.claude-plugin/
plugin.json Plugin manifest (name: lore).
marketplace.json Marketplace manifest (name: lore-tools).
The skill content is grounded against a real Lore installation:
Lore CLI 0.8.3+201 (verified 2026-06-18). SDK commits: lore-python @ 4f90b7f, lore-js @ 396ad34, lore-dotnet @ a56df73, lore-go @ 91ea1c5.
Any claim that could not be confirmed directly from the installed CLI or SDK source is marked <!-- UNVERIFIED --> inline. The validation script counts and surfaces these markers as warnings (see below).
Because Lore is pre-1.0 software, flag shapes and SDK method names can change between minor versions. The skill's iron rule: always run lore <cmd> --help before acting on flags, and never assume CLI verb names match SDK operation names.
A stdlib-only validation script checks the repo for structural correctness:
python scripts/validate_plugin.pyIt verifies:
- Plugin and marketplace manifests are well-formed.
- Every
SKILL.mdhas valid frontmatter (name:+description:). - No author-only material has leaked into shipped files.
- All
reference/...mdlinks in skill docs resolve to real files on disk. UNVERIFIEDmarkers are reported as warnings (never a failure).
CI runs the same script via .github/workflows/validate.yml.