Migrating from ClawDeck to KittyClaw
ClawDeck exports to JSON and CSV. KittyClaw imports both. This guide walks you through the move: no data loss, no big-bang migration. Agents are optional; you can run KittyClaw as a plain kanban forever and add automation later.
Step 1: Export from ClawDeck
- Open your ClawDeck workspace and go to Settings → Export data.
- Choose JSON (full export): this carries lists, cards, labels, and due dates. Use CSV if you only need the card titles and statuses.
- Save the file locally, e.g.
clawdeck-export.json.
Step 2: Install KittyClaw
If you haven't set it up yet, clone and start the server:
$cd KittyClaw
$./run.sh # macOS / Linux
$run.bat # Windows
Open http://localhost:5230, create a new project, and point it at any local folder.
KittyClaw scaffolds .agents/ and initialises git if needed.
Step 3: Import your board
- In KittyClaw, open your project and go to Settings → Import.
- Select your
clawdeck-export.json(or.csv). - KittyClaw maps ClawDeck lists → columns and cards → tickets. Labels and due dates carry over automatically.
- Review the imported tickets. Drag, rename, or split them before doing anything else: nothing is permanent yet.
The whole import takes about two minutes on a typical board. Larger boards (500+ cards) may take a few minutes more.
Step 4: Wire up agents (optional)
This is the part ClawDeck doesn't have. Each board member can be a Claude Code agent with its own role, skill file, and memory. You don't need to use it: skip this step if you want to migrate the kanban first and add agents later.
- Open Automations from the sidebar.
- Click Add automation and pick a trigger, e.g.
ticketInColumnwith columnInProgress. - Add a condition if you want to scope it:
assignedTo = programmer. - Add a
runAgentaction pointing at theprogrammeragent. - Assign any ticket to
programmerand move it to InProgress. The automation fires, Claude Code picks it up, and output streams into the ticket drawer.
What KittyClaw adds over ClawDeck
| ClawDeck | KittyClaw |
|---|---|
| Single workspace | Multi-project, isolated SQLite per project |
| Manual card assignment | Agent dispatch via trigger/condition/action automations |
| No context across sessions | Per-agent memory.md persists knowledge across runs |
| Cloud-hosted, subscription | Self-hosted, AGPL-3.0-or-later: your data stays on your machine |
The kanban muscle memory is identical: drag cards, add labels, comment, assign. The difference is that an assignee can be an LLM agent that reads the ticket, writes the code, and moves the card itself.