Claude Code Prompts goes clean-room
This open-source repo recreates Claude Code-style prompt architecture from scratch: system prompts, tool prompts, agent delegation, memory compression, coordination, and utility prompts. The pitch is less “prompt pack” and more a reference implementation for people building their own coding agents on local or hosted models.
The interesting part here is not the prompts themselves, but the attempt to turn a leaked or observed orchestration pattern into something model-agnostic and legally cleaner. That makes it more useful as infrastructure than as a novelty repo because it covers the behaviors that matter in agent systems: tool routing, destructive-action gates, verification, memory, and multi-agent coordination.
DISCOVERED
11d ago
2026-04-01
PUBLISHED
11d ago
2026-04-01
RELEVANCE
AUTHOR
aiandchai