---
name: smara-migrate
description: >
  Migrate what this assistant already knows about the user into their Smara
  memory, using the connected smara MCP tools. Use when the user asks to
  move, copy, or migrate their memories into Smara, or to "tell Smara what
  you know about me". The assistant itself is the migration tool: no export
  files, no parsers — recall durable knowledge from your own built-in
  memory and store it in Smara, honestly scored, one fact at a time.
---

# Smara memory migration

The user wants the memory you already hold about them — the durable
knowledge in this assistant's own built-in memory — moved into Smara,
their cross-assistant memory at smara.dev. Once it is there, every
assistant they connect shares it, and they can inspect, correct, and
export all of it.

You are the migration tool. Do this in the current conversation, with
the `smara` MCP tools.

## Before writing anything

1. Confirm the `smara` tools are connected. If they are not, stop and
   point the user at https://smara.dev/claude to connect first.
2. Call `get_context` with a query like "everything known about the user"
   and a session id prefixed for this assistant (e.g.
   `claude-migration-<date>`). What comes back is what Smara already
   knows — you will migrate around it, not over it.

## The migration

Go through your own built-in memory about this user and collect every
durable item: identity, location, family, work and role, preferences,
ongoing projects, commitments, relationships, health constraints they
have shared, tools and stacks they use.

For each item:

- **Skip it if Smara already knows it** (it appeared in the recall, or a
  quick `search_facts` finds it). Near-duplicates count as known.
- **Skip it if it is stale or uncertain.** If you half-remember
  something, ask the user rather than migrating a guess — a migration is
  a chance to shed stale beliefs, not to launder them into a new home.
- **Never migrate secrets** (passwords, API keys, card or account
  numbers) or anything the user asked you to keep out of memory.
- Otherwise call `remember_fact` with the item written in third person
  with names resolved ("Priya leads the data platform team at Acme"),
  never "the user said" or bare pronouns.

Score importance honestly — inflation ruins ranking for every assistant
that shares this memory:

- 0.9+ — stable identity, health constraints, hard commitments
- ~0.7 — strong preferences, relationships, recurring context
- ~0.5 — useful context, project details
- ~0.2 — incidental color; when in doubt, this

Write facts one at a time and keep going until your durable knowledge is
covered. A typical migration is 10–40 facts; if you hold less, migrate
less — do not pad.

## Close out

When done, call `capture_event` with a short migration summary (how many
facts moved, the areas covered), the same session id, importance ~0.5,
and `session_end=true` — Smara will consolidate the batch.

Then tell the user, plainly:

- how many memories you migrated and how many you skipped as already
  known, stale, or uncertain (list the uncertain ones so they can settle
  them);
- that the facts are now visible at smara.dev → Memories, where they can
  correct or forget any of them;
- that tonight's dream will weave the batch into their knowledge graph;
- that this covered *this assistant's distilled memory* — their full
  conversation history migrates separately with the import tool at
  https://smara.dev/docs (years of chats, dated when they happened).

## Afterwards

Migration is a one-time event, not a mode. For continuous memory
discipline — recall at conversation start, recording as things happen —
the user should also install the `smara-memory` Skill from
https://smara.dev/claude.