April 4, 2026
Anthropic just cut off your agents.
Your agents still work.
Yesterday Anthropic blocked OpenClaw and all third-party agent harnesses from using Claude subscriptions. If you built agents on OpenClaw, Paperclip, or any external runtime — your agent's persistent memory, identity, and behavior just lost its home on Claude.
The ban targets how these tools authenticate — OAuth is now locked to Claude Code and claude.ai only. But your agent's identity was never in the runtime. It was in the files. The SOUL.md. The memory folder. The session transcripts. Those are yours.
Agent-Mask doesn't need OAuth. It doesn't need a runtime. It doesn't need permission from anyone.
It's just a bunch of .md files being read and written to a folder path you specify on your machine or cloud drive.
What happened
Anthropic cut off OAuth access for all third-party agent tools, effective April 4 at 12pm PT. OpenClaw was named first, but the policy expands to all external harnesses in the coming weeks.
The reason cited: third-party runtimes don't optimize prompt cache hit rates the way Claude Code does, creating “unsustainable strain” on infrastructure. A single OpenClaw agent could burn $1,000–$5,000/day in compute that Anthropic was absorbing on flat-rate subscriptions.
The alternatives Anthropic offers: pay-as-you-go metered billing, or standard API keys at API pricing. Both are significantly more expensive than the subscription you're already paying for.
Why Agent-Mask isn't affected
No runtime, no ban
Agent-Mask is a set of skill files that run inside Claude Code, ChatGPT, Codex, or any tool that supports markdown-based skills. It doesn't wrap, proxy, or replace the host tool. There's no OAuth, no external auth, no separate process. Nothing to ban.
Your agent lives in a folder
SOUL.md, IDENTITY.md, MEMORY.md, daily memory files, session transcripts — all plain markdown in a directory you control. On your machine. On a cloud drive. Wherever you want. The agent's identity isn't locked inside any vendor's runtime.
Works everywhere, always
Hydrate your agent in Claude Code today. Shut down, switch to Codex tomorrow. The session transcripts are source-tagged so the agent knows where each conversation happened. Same identity, same memory, any tool.
Three-tier memory that survives anything
Full session transcripts for replay. Daily working memory for carry-forward context. Curated durable memory for long-term facts. Even if a platform disappears overnight, your agent's knowledge is in files you own.
If you're coming from OpenClaw
If you already have an agent workspace with SOUL.md, IDENTITY.md, MEMORY.md, and a memory folder — Agent-Mask can hydrate it as-is. Point it at your existing workspace and run /hydrate. Your agent wakes up with its full memory and identity intact.
If you don't have a workspace yet, run /create-workspace and describe your agent in a sentence. You'll get a full workspace with identity files, memory system, and session tracking in seconds.
When you're done, run /shutdown and every conversation turn, decision, and handoff note gets flushed back to your folder. Pick it up again in any tool, any time.
> /hydrate Agent hydrated: jensen Workspace: ~/agents/jensen/workspace Memory: warm (loaded 2 daily files, 23 durable entries) Source: Claude Skill Ready. What are we working on?
Get Agent-Mask
Three skill files. No runtime. No auth. No vendor lock-in.
Your agents keep working.