[BUG] Claude Code stalls in "Thinking/Doodling/Imagining" phase after reading multiple files in agentic chains

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?

Claude Code consistently stalls during the planning/thinking phase after completing a sequence of Read/Search tool calls in agentic tasks. The session appears frozen — no token output, no tool calls — for 10–15+ minutes before timing out or requiring manual interruption.

This happened repeatedly across multiple separate sessions today on the same task. Related to #26224 but specifically triggered by the file-reading → planning transition under high context load.

What Should Happen?

After completing a batch of file reads, Claude Code should proceed to the implementation phase and begin producing output or tool calls within a reasonable time. If extended thinking is required, it should eventually complete rather than stall indefinitely.

Error Messages/Logs

No error message produced. The thinking indicator (e.g. "✽ Imagining…", "✽ Doodling…", "✽ Envisioning…") remains active indefinitely. Token count does not increase during the stall. Session must be manually interrupted with Ctrl+C.

Steps to Reproduce

1. Give Claude Code an implementation task that requires reading multiple existing files (e.g. "create a Python package and implement a module based on these existing tests")
2. Observe Claude Code begin a batch of Read/Search tool calls (typically 5–10 files)
3. Wait for it to enter the thinking phase after the final read
4. Observe the thinking indicator running indefinitely with no further output

Example sequence observed:

• Search(pattern: "packages/backend/syntek-graphql-crypto/**/*") ✓
• Read(conftest.py) ✓
• Read(test_directives.py) ✓
• Read(test_read_path.py) ✓
• Read(test_write_path.py) ✓
• Read(test_error_handling.py) ✓
• Read(test_auth_guard.py) ✓
• ✽ Imagining… (14m 37s · ↓ 419 tokens · thinking) ← stalled, never continued

Workaround: Ctrl+C to interrupt, then re-prompt with explicit step-by-step instructions and a prohibition on further file reading. This reliably unblocks the session.

Claude Model

Sonnet (default)

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Unknown — issue began appearing today (2026-03-10). Correlates with the hanging bug reported in #26224.

Claude Code Version

2.1.72

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Other

Additional Information

Zed IDE integrated terminal, Oh My Zsh (zsh)

• Token count does not increase during the stall — this appears to be an SSE hang on the Anthropic side, not local processing
• Not using a VPN
• Not related to the March 8 DST infinite loop incident
• claude update confirms 2.1.72 is the latest version
• Stall occurs specifically at the transition from information-gathering to planning/implementation when accumulated context is large
• Sending a follow-up prompt (any content) sometimes kicks it back into action, consistent with #26224