Skip to main content
CHAPTER I — IN WHICH YOUR AGENTS LEARN TO READ

Tome.

One bookshelf for everything your coding agents know — read everywhere, loaded one skill at a time.

THE SPRAWL

Five agents. Five dialects. One you.

Your prompts, rules, and skills are pasted into .claude/, .cursor/rules, AGENTS.md, GEMINI.md — each copy drifting from the last. And the ones you do load sit in the context window all day, billing you for knowledge the agent isn’t using.

Claude Code · .claude/skills/
└─ review-checklist/SKILL.md
v1
Cursor · .cursor/rules/
└─ review-checklist/SKILL.md
v3
Codex · AGENTS.md
└─ review-checklist/SKILL.md
v2 (edited)
Gemini CLI · GEMINI.md
└─ review-checklist/SKILL.md
v2 !!STALE!!
OpenCode · AGENTS.md
└─ review-checklist/SKILL.md
v1 (conflicted copy)
CHAPTER II — THE METHOD

Shelve it once. Read it everywhere.

ACQUIRE
tome catalog add <repo>
BIND
tome harness use cursor
CONSULT
tome query "…"

Inside a harness the same search runs over MCP — the agent pulls exactly the skill it needs, mid-task, by itself.

COLOPHON — SPECIFICATIONS OF THIS VOLUME
Local, entirely

index, embeddings & reranker on your disk — no telemetry, no cloud

One binary, in Rust

a single static binary; semantic search built in

Honest exits

every failure has its own exit code — scriptable, CI-friendly

Five harnesses

native config for each — rules, MCP, agents, hooks

CHAPTER III — WRITE YOUR OWN

Already have plugins? Convert them.

Bring a Claude Code marketplace, a Cursor rules folder, or a loose pile of SKILL.md files — tome convert turns them into a catalog. create scaffolds new ones; lint keeps you publishable.

Read the conversion guide →

BEGIN

Add it to your bookshelf.

brew install aaronbassett/homebrew-tap/tome
or cargo install tome-mcp
Quickstart → tome init