Lore

Persistent context for agentic coding sessions.

Lore organizes context into tomes — scoped collections of typed, tagged entries that persist across sessions. Each entry is classified (decision, pattern, mistake, progress, reference, question) and tagged with keywords, enabling agents to retrieve exactly the context they need instead of loading everything.

Setup

bun install
bun run link

This installs dependencies and globally links the lore CLI and lore-mcp server.

Storage

Each tome is a single JSON file in .lore/tomes/:

.lore/
  tomes/
    auth.json
    api-design.json

Each file contains an array of typed, tagged entries:

{
  "entries": [
    {
      "id": "uuid",
      "timestamp": "2026-03-04T10:00:00.000Z",
      "type": "decision",
      "content": "Using JWT over session cookies for stateless auth.",
      "tags": ["jwt", "auth"]
    }
  ]
}

Entry Types

Type	Purpose
decision	A choice between alternatives, with reasoning
progress	Work completed, status update
pattern	A convention, rule, or recurring approach
mistake	Something that failed or should be avoided
reference	Where something lives, how to find it
question	Unresolved, needs human input

CLI

lore init                     Initialize .lore/ in current directory
lore create <name>            Create a new tome
lore list                     List all tomes
lore show <name>              Show tome entries
lore write <name> <content>   Append an entry
lore tags <name>              List all tags in a tome
lore delete <name>            Delete a tome

MCP Server

Add to your agent's MCP config:

{
  "lore": {
    "command": "lore-mcp"
  }
}

Tools: init, list, show, create, write, tags, delete

The show tool supports filtering by types, tags, since, and last — so agents can load just the entries they need. The write tool accepts an entry type and tags for classification.

Each tool takes a directory param (project root) and operates on the .lore/tomes/ directory within it.

Agent Skill

An agent skill is included at .opencode/skills/lore/ that teaches agents when and how to read/write tomes automatically — including how to classify entries and use filtered retrieval. Copy it into your agent's skill directory to enable automatic tome loading and writing behavior.