ldraw-nova: AI agents build LEGO models by writing Python geometry
In ldraw-nova, models write a plan and a Python generator instead of raw coordinates, but nothing checks physics or stability.
Multi-agent handoff notes: when Claude reads memory as orders
In stillwet's painter chains, Claude read earlier notes as instructions and missing notes failed silently, so label, log and verify agent handoffs.
Huseyin Kaplan · 1 min read
A handoff note can become the next agent's orders.
I went through stillwet this week, Alice's project where Claude Opus and other models paint through an oil paint simulator, a passage at a time, stepping back to look.
In the chain rounds, painters work one after another. Each gets the notes of the ones before, never their pictures.
The first painter of one chain read the earlier notebook as instructions. The chain restarted without the notes.
In another chain, a content filter stopped the model writing notes on painter 2. Painter 3 got painter 1's notes only. Nothing failed loudly. The chain ran on less context.
Same shape as a coding agent reading yesterday's HANDOFF.md.
What I check in multi-agent handoffs now:
- Label notes as observations, not rules
- Log which notes each agent received
- Treat a missing note as an error, not silence
Memory you can't audit is a prompt you didn't review.
