Thinking summaries deprecated in Claude Code v2.1.90+

[virtualian]

Date: 2026-04-14 · Affected: Claude Code ≥ v2.1.90

Problem

During long operations, the spinner showed only Thinking a bit longer… still working on it… instead of the live ∴ Thinking… narration that previous versions streamed. Extended thinking still ran (and was still billed), but the content was no longer visible.

Root cause

Claude Code v2.1.90 (April 1, 2026) changelog:

"Changed thinking summaries to no longer be generated by default in interactive sessions — set showThinkingSummaries: true in settings.json to restore"

The v2.1.107 entry "Show thinking hints sooner during long operations" only improves placeholder timing, not content visibility — upgrading further does not fix it.

Fix

One key in the user-level settings file, then restart Claude Code:

// ~/.claude/settings.json
{
  "showThinkingSummaries": true
}

Verification

• jq -e '.showThinkingSummaries' ~/.claude/settings.json → true + valid JSON
• After restart, confirmed empirically — the ∴ Thinking… narration streams live in the spinner area again. Ian pasted back verbatim content from a prior thinking block to prove it.

Token cost

• Dollars: zero. Thinking tokens are billed either way; the summarizer itself is free (adaptive thinking docs: "No charge: Tokens used to generate the summary."). Hiding the summary didn't save money — it just hid content you'd already paid for.
• Latency: small penalty. display: "omitted" (the v2.1.90 default) skips thinking-delta streaming for faster time-to-first-text; enabling summaries reintroduces the summarizer into the stream. Noticeable on short tasks, lost in the noise on long ones.
• Context & caching: unaffected. Thinking blocks aren't preserved across turns in normal conversations; prompt cache survives as long as the setting stays consistent through the session.

Net: latency for visibility, not dollars.

Not the fix (dead ends during investigation)

• Explanatory output style / its plugin — orthogonal. Controls narrative text between tool calls, not thinking visibility.
• CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING=1, /effort, MAX_THINKING_TOKENS — control thinking depth and cost, not visibility.
• Any settings.local.json outputStyle rename — unrelated; that was a separate length-limit rule issue that got conflated during investigation.

Implication

With the setting on, the thinking stream is visible to the user in real time. A user memory at ~/.claude/projects/-Users-ianmarr-projects-pai/memory/user_thinking_visible.md records this so future sessions write thinking knowing it's part of the conversation surface, not a private scratchpad.

Reference

• Claude Code Changelog — v2.1.90 entry

[danielmiessler]

Hey @virtualian, thanks for raising this, and sorry it sat for a while.

We're changing how LifeOS ships. Instead of cloning a full ~/.claude directory and running it as a complete system, LifeOS is becoming a skill you install through an agentic installer. The installer hands integration to your own AI, which reads your actual machine (your OS, your paths, your harness) and wires the hooks and system prompt in where they belong.

That's aimed right at what you hit here. The old "one directory, one layout, hope it matches your setup" approach is exactly what broke for so many people, and the new model should handle it far better because your AI does the integration per machine instead of us guessing.

So we're closing this in prep for that release. If it still bites you once the skill-based version is out, reopen or file a fresh one and we'll jump on it. Appreciate you taking the time.