ob-migrate

name	ob-migrate
description	use when: bootstrapping or extending open-brain memory by importing facts from prior conversation context, a JSONL export, or Obsidian/Markdown files. NOT for: ingesting URLs (use ingest-content), or saving a single observation (use `ob save` directly). boundary: ob-migrate is for batch + interactive memory writes; ingest-content is for one URL at a time; ob-search is read-only.
version	0.2.0
requires_standards	["open-brain/cli-routing","open-brain/memory-write-patterns"]
requires	["standard:english-only","mcp:open-brain"]

open-brain Memory Migration

Bootstrap or extend open-brain memory by migrating knowledge from prior context or external files. Idempotent (see write-patterns standard).

Quick Start

/ob-migrate                          # Interactive: extract facts from prior context
/ob-migrate /path/to/export.jsonl    # Batch: import JSONL file
/ob-migrate /path/to/notes.md        # Batch: import Obsidian/Markdown file
/ob-migrate /path/to/notes.md project:my-project

Mode 1: Interactive Mode

Trigger: /ob-migrate with no file argument.

1. Review current conversation history and loaded context (CLAUDE.md, standards, memory context). Identify concrete facts, decisions, patterns, learnings, or observations worth long-term preservation.

2. Present the extraction plan to the user as a numbered list with [type] text (project), then ask Proceed with migration? (y/n/edit) — wait for the user's answer before saving.

3. For each approved fact, call:

   save_memory(
     text=..., type=..., project=..., title=..., narrative=...,
     provenance={"producer":"ob-migrate","source_ref":"agent-session:<harness>:<session-id>"}
   )
   

   The capture router runs automatically — pass through whatever fields apply.

4. Track and print the summary per the write-patterns standard.

Focus on: technical decisions and rationale, conventions, project-specific knowledge (architecture, deployment, tools), preferences, session discoveries.

Mode 2: Batch — JSONL

Each line is a JSON object. Required: text. Optional: type, project, title, narrative, metadata.

{"text": "Use asyncpg for all DB access.", "type": "learning", "project": "open-brain"}
{"text": "Voyage-4 chosen for embeddings.", "type": "decision"}

Parsing rules:

• Skip blank lines silently
• Lines that are not valid JSON → count as error, continue
• Lines missing text → count as error, continue
• Unknown extra fields pass through into metadata

Mode 2: Batch — Obsidian/Markdown

Single file → one memory; entire content becomes text, filename (without extension) becomes title.

Multi-section file (sections separated by ---): each section is one memory. First heading (# ...) in a section becomes title.

Defaults for Markdown:

• type → "observation" (override via type:learning)
• project → null (override via project:my-project)

Batch Workflow

1. Read the file via the Read tool.
2. Parse all items per the format rules above. Count malformed entries as errors immediately.
3. Preview to the user with the count and proposed action — wait for y/n.
4. For each valid item, call save_memory with pass-through fields and one canonical source reference for the conversation or imported file.
5. Process the response per the write-patterns standard: duplicate_of → skipped, missing id or exception → error, otherwise → migrated.
6. For imports >10 items, print progress every 10.
7. Print the summary.

Arguments

Argument	Description
(none)	Interactive mode
<file-path>	Batch mode — JSONL or Markdown
project:<name>	Default project for all items
type:<type>	Default type for all items (Markdown)
limit:<n>	Import first N only
dry-run	Parse + preview without saving

Rules

• NEVER save without user confirmation in interactive mode
• Batch mode continues on error — malformed lines are counted, not fatal
• dry-run skips all save_memory calls (see write-patterns standard §5)
• All other write semantics (duplicate handling, summary, capture router, metadata encoding) inherit from open-brain/memory-write-patterns

Type Reference

Type	Use for
learning	Technical lessons, best practices, conventions
observation	Facts about a project, system, environment
decision	Architectural/design decisions with rationale
discovery	New findings, surprising behaviors
bugfix	Bug found and fixed (document root cause)
feature	Feature implemented or planned
session_summary	Summary of a work session