Weekly Claws #6: Claude Code Leaked, Anthropic Cuts Off OpenClaw

512,000 lines of Claude Code just leaked.

Anthropic accidentally published their entire CLI source code via a source map file in npm. The leak revealed unreleased features — and sparked a debate about AI-generated code quality.

The same week, Anthropic announced Claude subscriptions will no longer cover third-party tools like OpenClaw. Starting April 4th, users must pay separately via “usage bundles” or API keys.

Coincidence? You decide.

Let’s dive in. 🐾

🚀 AI News

🔓 Claude Code Source Code Leaked — Claude Code’s source code was accidentally published via a source map file in their NPM registry. The leak revealed unreleased features including “assistant mode” (codename Kairos), a Tamagotchi-style “Buddy System,” and “Undercover mode” for stripping Anthropic internal info from commits. A 3,167-line function with 486 branch points and 12 levels of nesting sparked debate about AI-generated code quality.

📰 Anthropic Bans OpenClaw from Claude Subscriptions — Starting April 4th, Claude subscriptions no longer cover third-party access from tools like OpenClaw. Users must now pay separately via “usage bundles” or API keys. Anthropic says their subscriptions “weren’t built for the usage patterns of these third-party tools” (@bcherny). OpenClaw founder Peter Steinberger says he “tried to talk sense into Anthropic, best we managed was delaying this for a week.”

🐦 GitHub Becomes Official OpenClaw Sponsor — GitHub is now an official sponsor of OpenClaw, providing Copilot Pro+, funding for security through the Secure Open Source Fund, and infrastructure to help them “keep building, stay secure, and scale.” GitHub’s COO Kyle Daigle: “Fastest growing open source project in history.”

OpenClaw News

OpenClaw 2026.4.7 — openclaw infer (headless inference hub), music + video editing, session branch/restore, webhook-driven TaskFlows, Arcee/Gemma 4/Ollama vision support, and memory-wiki for persistent knowledge with contradiction tracking and compiled digests. “Because ‘trust me bro’ is not a knowledge system.”

OpenClaw 2026.4.5 — Built-in video + music generation, /dreaming is now real, structured task progress, better prompt-cache reuse, and Control UI + Docs now speak 12 more languages. “Anthropic cut us off. GPT-5.4 got better. We moved on.”

OpenClaw 2026.4.2 — Durable Task Flow orchestration, better native exec defaults + approvals, Copilot + Kimi + provider hardening, tighter plugin activation boundaries, and hardened provider transport + routing.

OpenClaw 2026.4.1 — GLM 5.1 + failover that doesn’t loop, AWS Bedrock Guardrails, /tasks for agent receipts, cron per-job tool allowlists, and 40+ stability & exec fixes.

📖 OpenClaw Dreaming — Experimental, opt-in memory consolidation system that promotes meaningful short-term signals into durable memory through explainable Light, Deep, and REM-style phases. Background processing with human-readable Dream Diary output.

🎬 Give Your OpenClaw a Personality Upgrade — Peter Steinberger shares the ultimate SOUL.md hack: make your AI have opinions, drop the corporate speak, embrace brevity, and “be the assistant you’d actually want to talk to at 2am.” Swearing optional but encouraged when it lands.

🔌 Around the Web

Tools & Resources

🛠️ Hermes Agent — Manim Skill — Nous Research adds Manim support to Hermes Agent. Create precise programmatic animations for mathematical and technical explainers, made famous by the @3blue1brown channel.
🛠️ Hermes Agent — LLM-Wiki Integration — Hermes Agent now comes packaged with Karpathy’s LLM-Wiki for creating knowledgebases and research vaults with Obsidian. Type /llm-wiki <research x> to have Hermes study the web, code, and papers to build a knowledge base around any topic.

😂 Fun

Mission Accomplished meme

Where were you when OpenClaw got banned?

That’s your weekly claws! See you next Monday. 👋