honcho-integration — Claude Code skill: what it does, how to install

What honcho-integration does

The skill gets an agent to wire Honcho, an open-source memory library for stateful agents, into a codebase that already exists in Python or TypeScript. It reads the repo first: finds LLM calls, user models, chat handlers. Then it runs a short interview with AskUserQuestion about which entities become peers, how conversations map to sessions, and how the model should read context.

Reference files open in step order: references/core-patterns.md for the client, peers and message storage, references/agent-patterns.md for tool call, pre-fetch and context() wiring, references/bot-frameworks.md for agent loops such as nanobot and openclaw. Allowed tools stay narrow — read, search, edit files, plus uv, bun and npm for installs.

How the memory behaves: your app sends messages, background reasoning builds a representation of each peer, and the agent pulls it back with session.context() or asks peer.chat(query). chat() spends a few seconds reasoning live, so keep it for questions that need a reasoned answer and take context() per turn.

Get a Honcho API key from app.honcho.dev and put it in the environment before anything runs. The SDK is honcho-ai on Python, @honcho-ai/sdk on TypeScript. Bots with deterministic behaviour should get observe_me=False, otherwise Honcho models a peer that gains nothing from it. The project ships under AGPL-3.0.