Agent team lead (@main) stalls after teammates report back — requires keypress to resume

Environment

• Claude Code: 2.1.101
• Model: claude-opus-4-6 (1M context)
• OS: macOS 26.3.1 (Darwin 25.3.0, arm64 / Apple M2)
• Node.js: v25.8.1
• Terminal: Warp v0.2026.03.04.08.20.stable
• Shell: zsh
• Teammate mode: tmux
• CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS: "1"

Relevant settings

{
  "teammateMode": "tmux",
  "permissions": { "defaultMode": "auto" },
  "outputStyle": "default",
  "effortLevel": "high"
}

Description

When running an agent team (4 teammates: researcher, analyst, architect, devils-advocate) with the @main session as team lead, the main agent repeatedly stalls after all dispatched teammates have completed their tasks and sent their reports back via SendMessage.

The teammates finish their work, send their reports, and go idle. The main agent has received all the messages (they appear in the conversation) but does not proceed with its next action (processing the results, updating state, dispatching the next round). The main agent appears frozen — no tool calls, no text output, no progress.

Any keystroke in the @main tmux pane (Enter, arrow key, any character) immediately unblocks the agent. It then processes all the queued teammate reports and continues as expected.

Reproduction steps

This happened consistently across all 4 iterations of a research workflow:

1. Team lead creates a team via TeamCreate and spawns 4 teammates via Agent tool with team_name parameter
2. Team lead sends research questions to 3-4 teammates via SendMessage
3. Teammates work independently (web search, code analysis, file reads, etc.)
4. Teammates complete work and send reports back via SendMessage
5. Teammate idle notifications arrive in main session
6. BUG: Main agent stalls here. Does not process reports or take next action.
7. User presses any key in the main tmux pane
8. Main agent immediately resumes — processes all reports, continues workflow

This occurred at every iteration boundary (after iteration 1, 2, 3, and 4 dispatches). The stall duration varied — the agent would not self-recover without human input.

Expected behavior

The @main agent should automatically resume processing when teammate messages are delivered to its inbox. No human keypress should be required.

Actual behavior

The @main agent freezes after receiving teammate messages and idle notifications. It requires a physical keypress in its tmux pane to resume execution.

Impact

This defeats the purpose of autonomous agent teams. The team lead should be able to orchestrate multi-iteration research loops without human intervention between iterations. Currently, the human must monitor the tmux panes and manually prod the main agent each time teammates complete their work.

Workaround

Press any key in the @main tmux pane when it appears stuck. This immediately unblocks processing.