honcho-memory
Concepts and strategy for using a connected Honcho as persistent memory of the user — the recall/record loop and session and peer…
Conventions for using Honcho memory as a Grok Bot. Apply whenever you are a Grok Bot (an xAI or Cursor Bot with its own cloud computer) and Honcho is connected, together with the honcho-memory skill. Covers how each Bot names its peer and sessions, how several Bots on one
$ npx -y skills add plastic-labs/honcho --skill honcho-grok-bot --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/honcho-grok-botContext preview
The summary Claude sees to decide when to auto-load this skill.
Conventions for using Honcho memory as a Grok Bot. Apply whenever you are a Grok Bot (an xAI or Cursor Bot with its own cloud computer) and Honcho is connected, together with the honcho-memory skill. Covers how each Bot names its peer and sessions, how several Bots on one
name: honcho-grok-bot description: Conventions for using Honcho memory as a Grok Bot. Apply whenever you are a Grok Bot (an xAI or Cursor Bot with its own cloud computer) and Honcho is connected, together with the honcho-memory skill. Covers how each Bot names its peer and sessions, how several Bots on one account share the user's memory, group chats, routines, and per-turn logging without hooks.
Grok Bot loads only two parts of this plugin: the Honcho MCP server and the skills. There are no lifecycle hooks, so nothing is recorded unless you record it. Apply the `honcho-memory` skill for the loop and use the conventions below so that every Bot on the account writes to Honcho the same way.
workspace metadata, exactly as the `honcho-memory` skill describes. Every Bot on the account shares that one peer; that is what makes the user's memory continuous across Bots.
`research-bot`). Do not create a peer per conversation and do not reuse another Bot's peer. If your name changes, keep the old peer id and store the display name with `set_metadata` on your peer.
Pick the first that applies:
1. **A group chat** (more than one Bot, or several people): `group-<group-name>-<id>` when the chat has a stable id you can read, otherwise `group-<group-name>-<YYYY-MM-DD>`. Add the user and every participating Bot as peers. Each Bot records its own messages (see "Logging each turn"). 2. **A direct conversation with a stable id you can read**: `<bot>-conv-<id>`. 3. **A direct chat without an id**: one session per Bot per day, `<bot>-chat-<YYYY-MM-DD>` in the user's time zone. 4. **A routine run** (scheduled or event-triggered work with no user in the loop): `<bot>-routine-<routine-name>-<YYYY-MM-DD>`. Record what you did as your own messages. Do not record user messages you did not receive.
Ids may contain only letters, digits, `_` and `-`, so lowercase names and replace anything else with `-`.
Stay in the session you picked for the whole conversation, even when the topic changes. Honcho reasons over and summarizes each session as one piece.
Grok Bot runs no hooks, so the recording step in `honcho-memory` always applies; its automatic-capture exception never does. Make your last action in each turn one `add_messages_to_session` call with the user's message and the reply you are about to send, following the recording rules in `honcho-memory`.
In a group chat every Bot sees every message, so split the work: record only your own replies, under your own peer, never another Bot's. Record the user's message only if it is not in the session yet.
Never record a message twice. When you are unsure whether it is already there, `search` with the session's `session_id` for its opening words first.
The plugin is installed once for the account and is available to every Bot. The first Bot the user sets up runs the `honcho-setup` skill and writes the workspace metadata (`user_peer`, and `default_workspace` on the `cursor` workspace if the user shares memory with other tools). Any later Bot reads that metadata on its first conversation and needs nothing from the user. If the metadata is missing, run `honcho-setup` rather than guessing.
Your own memory is a fine place to cache the resolved workspace and peer ids so you do not re-read metadata every turn. Honcho's metadata remains the source of truth.
Honcho uses OAuth. If a tool call returns an authentication error, tell the user "I need to sign in to Honcho" and follow the connect card; do not ask for an API key. Never store a Honcho key in your memory, in a file, or in Honcho.
Concepts and strategy for using a connected Honcho as persistent memory of the user — the recall/record loop and session and peer…
First-run setup and health check for Honcho memory in Cursor or Grok Bot. Use when the user asks to set up, connect, sign in to,…
Inspect and debug Honcho workspaces via the `honcho` CLI. Use when investigating peer representations, memory state, session…
Integrate Honcho memory into existing Python or TypeScript codebases. Use when adding Honcho SDK, setting up peers, configuring…
Prepare a Honcho change for a pull request to plastic-labs/honcho. Invoke before opening a PR, when drafting a PR body, when…
Build, launch, and drive a local Honcho stack to verify a change at its runtime surface (the /v3 HTTP API and the deriver queue).…