Agent offers to run sub-agent but cant see results in 'user-visible summary', so re-runs sub-agent functionality (wasting tokens)

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

I was happy to see an instance of Opus 4.7 Extra High agent chat offer to run a ‘form compliance auditor’ sub-agent I recently created.

Relevant sub-agent config is:

  • model: inherit
  • readonly: true
  • is_background: (not defined, so defaults to false)

I accepted, but then saw the agent responding with:

The subagent is delivering its checklist via the user-visible summary which I can’t see directly. Let me just audit the files myself — I know the conventions well.

So I stopped the response because I was thinking:

…hang on, the sub-agent has already done this work (and used the necessary tokens to achieve it), so I dont want the agent performing the same work again (and using additional tokens)

I get the best results from using this model at this setting (Opus 4.7 Extra High), but it hoovers up the tokens, so i really dont want agents re-running actions that a sub-agent has already performed.

When calling the sub-agent directly with the forward slash, it seems to be able to reference the sub-agent results as expected.

Steps to Reproduce

  • Try and get the agent to offer to use a sub-agent
  • Accept the offer
  • See if the agent can read the results from the sub-agent or not

Expected Behavior

I expected the agent to be able to see the results of the sub-agent.

Operating System

Windows 10/11

Version Information

Version: 3.9.16 (user setup)
VS Code Extension API: 1.105.1
Commit: 042b3c1a4c53f2c3808067f519fbfc67b72cad80
Date: 2026-06-27T06:41:01.941Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Windows_NT x64 10.0.26200

For AI issues: which model did you use?

Opus 4.7 Extra High

Does this stop you from using Cursor

No - Cursor works, but with this issue

Edit:

In case it is not clear from the UI screenshot, this occurred in Cursor Editor View, not Cursor Agent View.

Hey @litecode!

Good question — the parent model should be able to read both the user-visible summary and the subagent response. So this looks like a model hallucination rather than a real limitation.

Could you try telling the agent explicitly that it can read both the summary and the response, and see if that changes its behavior?

For context: Opus 4.7 is famously quite literal about prompting, and the agent is told:

“The beginning of the above subagent result is already visible to the user. … DO NOT regurgitate or reiterate its result unless asked.”

My guess is it’s interpreting that instruction more strictly than intended.