Lore Plugin — Development skill for Claude Code

lore-plugin

A Lore plugin bundle for both Claude Code and Codex that provides persistent, compounding long-term memory via [Lore](https://lore.dev).

On Claude Code, every session ends with an automatic transcript ingest. On Codex, the plugin offers the same setup and memory skills, plus an optional hook install that preloads the Lore index at session start and ingests each completed user/assistant turn.

**Configuration is per-project.** Every repo that wants Lore memory has its own `.lore.env` (committed, contains app + namespace) and `.lore.env.local` (gitignored, contains your API key). There is no global config. Teammates cloning a configured repo only need to run `/lore-setup` once to add their own API key — the app/namespace wiring is inherited from git.

What you get

• lore-memory skill — Claude auto-invokes this when the conversation references your long-term memory, asks a question prior context might answer, or states a durable fact worth keeping. It reads the wiki's _index, follows [[wikilinks]] between files, and calls remember to ingest new content.
• lore-setup skill — three-phase interactive connect flow: (1) credentials, (2) namespace schema addendum tailored to what you want remembered, (3) optional seeding from the last N GitHub PRs in the current workspace. Also available as the /lore-setup slash command. Handles three scenarios automatically: fresh project, teammate onboarding an existing repo, and reconfiguring a project that's already set up.
• SessionStart hook — once the project is configured, pre-loads your namespace _index into Claude's session context on every session start so the wiki's table of contents is available from turn 1 without a lazy round-trip. If the current project isn't configured yet (or your .lore.env.local is missing an API key), it instead nudges Claude to proactively offer /lore-setup on your first substantive message. Capped at 8 KB of index content by default (tunable); fails si