[BUG] Subagents return only "Done" with no output when dispatched for search/research tasks (Opus 4.8)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

When the main thread dispatches a subagent (via the Task tool) to do search/research work, the subagent consistently returns a single word — "Done" — with no actual output, summary, or findings. The work product the subagent was supposed to produce never makes it back to the main thread.

This happens reliably, not intermittently. The subagent appears to run, but the only thing surfaced to the orchestrating agent is "Done", so the result is unusable and the main thread has nothing to work with.

What Should Happen?

The subagent should return its complete work product as its final message — the research findings, the design, the summary, etc. — since that final message is the only thing the calling (main) thread can see. A bare "Done" defeats the entire point of dispatching a subagent.

Error Messages/Logs

No error messages — the subagent reports success ("Done") but produces no content. ⚠️ (If you have a transcript / session log showing the subagent run, paste it here — it strengthens the report a lot.)

Steps to Reproduce

1. Start a Claude Code session using Opus 4.8.
2. From the main thread, dispatch a subagent via the Task tool with a substantial research/design prompt (e.g. "research X and return your complete findings as your final message").
3. Explicitly instruct the subagent that its final message is the only thing the caller will see (e.g. "Return your COMPLETE result as your final message — this is the only thing the caller sees").
4. Observe the subagent's returned output.

Expected: the subagent returns its full findings/design.
Actual: the subagent returns only "Done" with no content.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

Claude 1.11187.4 (584005) 2026-06-05T20:15:04.000Z,desktop

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

The dispatched prompt was a frontend implementation-planning task (design a plan and return the complete design as the final message). Sanitized version of the prompt:

Design the frontend implementation plan for a new admin/operations console in a TanStack-based web app monorepo. Authoritative facts are provided; do not re-explore broadly. Return your COMPLETE design as your final message — this is the only thing the caller sees. [structured spec for routing/shell, a typed-client namespace, a mock-server extension, auth-token wiring, the console UI, and a test strategy follows]

The subagent returned only "Done" instead of the requested complete plan. The "return your complete output as your final message" instruction was explicit, which makes the empty return especially notable.