claude-code - 💡(How to fix) Fix UX feedback: hard firewall between Claude Code and Claude Cowork makes cross-surface work invisible

Type: UX feedback (not a bug report, not a feature request — observation about how the surfaces feel to use together).

What happened:

I do real work in both Claude Code and Claude Cowork. Today I asked a Claude Code session: "In the AI video storytelling process, what SaaS services did we subscribe to? I'm not using them so I should cancel them."

Claude Code had no visibility into that work at all. It exhaustively searched my memory directory, my .shared/secrets/, my canonical service-access guide, every package.json, every .env file across all my repos, and source code for any video-AI SDK or API call. Came back with a confident, well-evidenced "nothing." Which was technically correct for the Code side — but completely missed the actual answer, because the subscriptions were set up inside a Cowork session.

The frustration:

From a user perspective, this is one assistant working with me on one ongoing body of work. The fact that Claude Code and Claude Cowork operate as if they're separate isolated products — no shared memory, no shared session history, no way for one to even acknowledge that the other exists or that I have a Cowork account — is really bad UX. It forces me to:

1. Remember which surface I did which work in
2. Re-explain context every time I cross over
3. Watch a capable agent confidently tell me something doesn't exist when it does — just on the other side of the firewall

The "hallucinate by omission" failure mode this creates is worse than a normal hallucination, because the answer is well-researched and internally consistent. There's no signal that the assistant is blind to half my work.

What "good" would look like, even minimally:

• At minimum, Claude Code should be aware that I have a Cowork account and surface it as a known blind spot ("I can't see Cowork history; want to check there?")
• Better: shared memory / session index across both surfaces
• Best: one assistant, one context, multiple UIs

No action requested. Just wanted this on the record. The current setup feels like Anthropic built two separate products that happen to share a brand, rather than two surfaces of one assistant.