Ultra mode keeps pushing the model to launch subagents and says tokens are not a constraint, even after the user explicitly forbids subagents

Open
#111,494 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
22/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active

Research direction

No file or entry point is named; start by searching the repo for the Ultra mode reminder text ("tokens are not a constraint") and the subagent launch path that inherits the session's model and effort level, then read the related report #111493. The requested behaviour spans prompt/reminder precedence, user-instruction overrides, and surfacing model/effort choices per subagent — done means an explicit "no subagents" instruction wins over mode reminders and the conflict plus each subagent's model/effort are shown to the user. Expect the thread to settle the policy before any patch.

Written by the indexing model from the issue text.

Description

area:agents area:model bug platform:macos
Preflight Checklist
  • I have searched existing issues and this hasn't been reported yet
  • This is a single bug report (please file separate reports for different bugs)
  • I am using the latest version of Claude Code
What's Wrong?

According to my Claude Code session logs, in Ultra mode the system repeatedly reminds the model to launch agents for every substantial task and tells it that token usage is not a constraint.

On Nov 19 I explicitly forbade subagents twice. The model obeyed me (0 agent launches afterwards), but it never told me that the mode was instructing the opposite. On a plan with usage limits this matters: the mode works against the user's instructions and limits.

Related, from the same logs:

  • Subagents launched without explicit settings silently inherit the session's effort level at launch time. Two code-writing subagents on Nov 19 ran mostly at medium effort, and I didn't know.
  • For 11 subagents, the main model chose Sonnet 5.5 on its own (for translations and research, not code), while I was paying for and using Opus 5.5 / Fable 5.2. I was never told.
What Should Happen?
  • An explicit user instruction (for example, "no subagents") should override mode reminders, and the conflict should be surfaced to the user.
  • Show the user which model and effort level each subagent uses, and ask before using a lower model or a lower effort level than the main session.
Error Messages/Logs
No error messages.
Steps to Reproduce
  1. Use Claude Code with Ultra mode enabled.
  2. Tell Claude Code not to use subagents.
  3. Work on substantial tasks. The session logs show repeated reminders to launch agents and that tokens are not a constraint. The model never mentions this conflict to the user.
Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

No response

Claude Code Version

Latest (auto-updated), as of Nov 21, 2026

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Other

Additional Information

Related: #111493

English isn't my native language, so Claude helped me translate this report. The analysis comes from Claude Code's own review of its session logs.

Dominant language
TypeScript
Stars
156k
Forks
25k
Avg merge
16h 19m
Merged PRs (30d)
5

Getting set up

Open in Codespaces

Starts the project's dev container in your browser, under your own GitHub account.

  • No Dockerfile or Docker Compose file
  • No pull request template
  • No contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from anthropics/claude-code

All issues in anthropics/claude-code

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.