Three releases, no headlines: a Claude Code patch with 28 fixes, and both SDKs gaining streaming token estimates for extended thinking.
The most visible change in v2.1.147 is a rename: /simplify is now /code-review, and it ships with effort levels (/code-review high) and a --comment flag that posts findings directly as inline PR comments on GitHub. Clean integration, no clipboard archaeology.
The second change worth tracking is operational: pinned background sessions (Ctrl+T in claude agents) now survive idle, restart in place when an update lands, and are shed under memory pressure only after non-pinned sessions clear first. Sessions that hold their state through an update cycle are sessions you can rely on.
The rest is a 28-item fix list covering PowerShell on Windows, MCP server pagination, auto-updater retries, diff rendering performance, and edge cases in plugin and shell snapshot handling. Nothing changes what the tool can do. Several things change how reliably it does it.
/simplify renamed to /code-review with effort-level reporting (/code-review high) and a --comment flag for GitHub PR inline comments. Pinned background sessions (Ctrl+T in claude agents) now survive idle and restart in place when updates land. 28 additional fixes including: auto-updater retry logic, diff rendering performance on large files, enterprise login enforcement for forceLoginOrgUUID and forceLoginMethod, PowerShell on Windows (winget and Microsoft Store installs), MCP server pagination (resources and prompts past page 1 were silently dropped), shell snapshot user-function handling, plugin agent multi-type declarations, hook if condition matching, and full-screen strobing on Windows Terminal.
claude update to get v2.1.147. Then use /code-review in any session, or /code-review high for a deeper pass. Add the --comment flag to post findings as inline GitHub PR review comments.Adds thinking-token-count beta support: when streaming extended thinking, the SDK now surfaces estimated token counts in thinking block deltas. Useful for cost estimation and context budget management in agentic loops that rely on extended thinking.
thinking-token-count beta header in your request. Token estimates arrive as metadata on thinking block delta events. See streaming docs for the full schema.Parity release with Python v0.104.0: same thinking-token-count beta feature for streaming extended thinking token estimates. Both SDKs shipped the same capability on the same day.
thinking-token-count beta header. Token estimates appear in thinking block delta events during streaming.No model releases. No API feature announcements. Code with Claude London wrapped its two-day run on May 20 (MIT Technology Review and Fortune published coverage today), but the features shown there, self-hosted sandboxes and MCP tunnels for Claude Managed Agents, were announced on May 19. Nothing from the claude-apps or system-prompts release channels confirmed in this window. The docs.anthropic.com release notes pages returned 403 on this sweep and could not be directly verified.
Daily digest at 9 PM ET. Weekly magazine every Friday morning. Six labs, one feed. No spam, one-click unsubscribe.