This week in Claude Code · May 24, 2026

Claude Code & SDK Weekly: /code-review lands, usage breakdowns, and streaming thinking tokens

Get the next issue in your inbox. Weekly · Free

TLDR;

Claude Code v2.1.147-v2.1.150 shipped a run of meaningful developer improvements: /simplify is now /code-review with configurable effort levels and inline GitHub PR comment support, pinned background sessions survive idle and auto-update in place, and /usage now shows per-category limit breakdowns covering skills, subagents, plugins, and MCP servers. The Bash tool exit-code-127 regression introduced in v2.1.147 was patched immediately in v2.1.148. On the SDK side, the TypeScript SDK v0.98.0 adds thinking-token-count beta support for streaming thinking block deltas. Update all SDKs to their latest versions now to avoid the Bash regression.

Claude Code & CLI

Claude Code v2.1.149: /usage per-category breakdown and scrollable /diff

The /usage command now shows a line-item breakdown of what is consuming your limits across skills, subagents, plugins, and individual MCP servers.
The /diff detail view gains full keyboard scrollability (arrows, j/k, PgUp/PgDn, Space, Home/End), and Markdown output now renders GFM task list checkboxes.

Claude Code v2.1.148: Bash exit-code-127 regression hotfix

A regression introduced in v2.1.147 caused the Bash tool to return exit code 127 on every command for some users.
v2.1.148 patches this immediately with no other user-facing changes.

Claude Code v2.1.147: /code-review command, pinned background sessions, and improved auto-updater

/simplify has been renamed to /code-review and now reports correctness bugs at a chosen effort level (e.g., /code-review high); pass --comment to post findings as inline GitHub PR comments.
Pinned background sessions (Ctrl+T in claude agents) now stay alive when idle, restart in place to apply updates, and are the last to be shed under memory pressure.

Claude Code v2.1.146: /code-review debut, PowerShell winget fix, MCP pagination fix

This release first introduced the /code-review rename with optional effort levels, fixed the Windows PowerShell tool failing with 'command line is invalid' when pwsh is installed via winget or the Microsoft Store (regression from v2.1.124), and patched MCP resources/list and prompts/list dropping items beyond page 1 on paginating servers.

Claude Code v2.1.150: Infrastructure-only update

v2.1.150 ships internal infrastructure improvements with no user-facing changes, and is the version bundled with claude-agent-sdk-python v0.2.87 and claude-agent-sdk-typescript v0.3.150.

Agent SDK & Managed Agents

claude-agent-sdk-python v0.2.87: Bundles CLI v2.1.150 and switches CI to Workload Identity Federation

This release updates the bundled CLI to v2.1.150 and migrates CI authentication from static long-lived API keys to Workload Identity Federation using short-lived tokens, improving supply-chain security.

claude-agent-sdk-typescript v0.3.149: Fix for options.env dropping CLAUDE_AGENT_SDK_VERSION

Supplying a custom environment via options.env was silently dropping CLAUDE_AGENT_SDK_VERSION, breaking User-Agent headers and telemetry.
The fix is shipped in v0.3.149, along with corrected docs clarifying that options.env replaces the subprocess environment rather than merging with process.env.

claude-agent-sdk-typescript v0.3.150: Parity with Claude Code v2.1.150

Routine parity release tracking the CLI infrastructure update in v2.1.150.
Available via npm, yarn, pnpm, and bun.

claude-agent-sdk-typescript v0.3.147-v0.3.148: Parity releases for CLI v2.1.147 and v2.1.148

These two back-to-back parity releases keep the TypeScript SDK in lockstep with the /code-review launch and the Bash exit-code-127 hotfix in the underlying CLI.

API & Model Updates

anthropic-sdk-typescript v0.98.0: Thinking-token-count beta for streaming thinking block deltas

Released 2026-05-21, v0.98.0 adds API-level support for the thinking-token-count beta, exposing estimated token counts inside thinking block deltas when streaming extended thinking responses.

anthropic-sdk-typescript v0.97.0: Self-hosted sandbox helpers and Node 26 tsc-multi fix

Released 2026-05-19, v0.97.0 adds sandbox helper support for self-hosted CMA sandboxes and upgrades tsc-multi to restore compatibility with Node 26.

anthropic-sdk-typescript v0.97.1: SessionToolRunner skip fix

Released 2026-05-19, this patch corrects the runner to skip tool calls that SessionToolRunner does not own, preventing incorrect dispatch in multi-tool sessions.

Bedrock SDK v0.29.2 and Vertex SDK v0.16.1: @types/node alignment for CI

Both cloud-provider SDK sub-packages released 2026-05-19 to align @types/node versions across sub-packages, fixing a CI build break introduced by the monorepo structure.

MCP Ecosystem

Claude Code v2.1.149 Enterprise: allowAllClaudeAiMcps managed setting

Enterprise deployments can now set allowAllClaudeAiMcps in their managed configuration to load claude.ai cloud MCP connectors alongside the local managed-mcp.json, reducing manual connector management for org admins.

Claude Code v2.1.146: MCP pagination bug fixed for resources and prompts lists

MCP servers that paginate resources/list, resources/templates/list, and prompts/list were silently dropping all items beyond the first page.
v2.1.146 patches this so all pages are fetched correctly.

Action Items

Immediate

Update Claude Code to v2.1.150 (or at minimum v2.1.148) to avoid the Bash exit-code-127 regression shipped in v2.1.147.
Update claude-agent-sdk-python to v0.2.87 (pip install claude-agent-sdk==0.2.87) to pick up CLI v2.1.150 and the CI Workload Identity Federation change.
Update claude-agent-sdk-typescript to v0.3.150 (npm install @anthropic-ai/claude-agent-sdk@0.3.150) to stay at CLI parity and receive the options.env telemetry fix from v0.3.149.
If you use the Anthropic TypeScript SDK and stream extended thinking, update to v0.98.0 to access the thinking-token-count beta for estimated token counts in thinking block deltas.
If you use SessionToolRunner in multi-tool workflows, patch to anthropic-sdk-typescript v0.97.1 to fix incorrect tool call dispatch.
Replace any /simplify calls in scripts or documentation with /code-review; update CI comment steps to use the --comment flag if you want inline PR annotations.
Enterprise admins: evaluate the new allowAllClaudeAiMcps managed setting in v2.1.149 if your org relies on claude.ai cloud MCP connectors alongside managed-mcp.json.
If you run MCP servers that paginate large resource or prompt lists, verify your integration against the v2.1.146 pagination fix to ensure all items are now surfaced.

All Resources

Every Monday morning

This week in Claude Code

Five minutes, every Monday — the tools, releases and tactics for shipping solo.

unsubscribe anytime.