The Leak

Anthropic’s source code is out—read it while you can.

Claude Code CLI Full TypeScript Source Leaked via npm .map Misconfiguration

Claude Code CLI full TypeScript source accidentally exposed via misconfigured .map file — rare primary-source architectural visibility for agent builders.

Anthropic OSS Trio: Hermes-agent, Paperclip Heartbeat, Phone Gateway

Points to three specific Anthropic open-source tools (Hermes-agent, Paperclip, Phone gateway) that are genuinely useful for agent builders, but thin on detail.

Cognition

Explicit memory, multi-agent critique, and smarter AI design.

Karpathy: Explicit Memory Artifacts Beat Implicit AI Personalization

Karpathy (primary, credible source) elaborating on explicit memory artifacts for LLM personalization — architectural insight with real-world example (Farzapedia).

Claude Skill: Multi-Agent Critique via Karpathy’s LLM Council

References a concrete Claude skill implementing Karpathy’s LLM council as multi-agent critique sub-agents — useful pattern but needs more detail.