@mattpocockuk Give domain.md the same split you already use for research: respect accepted ADRs only, and open superseded ones only when reopening the decision.
Building a tower defense game is hard.
After about 6~7 times iterations, this tower defense game (ironpath) reaches this playful status. It's a simple demo, but contains some essential elements:
- fullscreen canvas
- background music
- hit sound effects
- music control and mix
- lives, gold balance, waves
- economy system
- random algorithm
- 4 kinds of enemies
- 4 kinds of weapons
- enemy sprite sheet and animation
- battle numeric system
- bullet textures
- ultra weapon: Meteor strike
- enemy detection...
All of them are done in 6 iteration of prompts. https://t.co/j08ErGlCSe makes the iteration go stable and predictable. This is super important for any serious project.
https://t.co/AOvTwPIZjY
Try it on https://t.co/j08ErGlCSe.
> Mods take direct inspiration from Pi's extension system, where agents programmatically modify the harness they run inside.
https://t.co/4h0UuusgRs
Build a plugin once and use it across compatible agent clients.
Introducing Agent Plugins, an open standard developed with @awsdevelopers, @cursor_ai, @github, @code, and @vercel that packages Agent Skills and supports MCP server configurations in a shared format.
Claude Code 2.1.221 has been released.
39 CLI changes
Highlights:
• Focus view hides tool activity behind per-turn summaries with a live running-tool indicator, reducing clutter
• Write tool documented to create or modify local files, clarifying outputs will alter disk state
• Grep note recommends ripgrep-backed tool, documents output modes and line/offset limits for stable searches
Full details available in thread ↓
Asking the models to not preserve backward compatibility can be dangerous.
A few weeks ago one of my agents nuked an entire production table when doing a migration, and I almost lost $2,000 worth of data.
Thankfully I had backups I could restore.
Having the right systems in place is more important than ever.