Source Intelligence
Follow Discord

DisclaimerUnofficial, and not affiliated with Anthropic. Nearly all of this is read straight out of what ships: npm bundles, captured prompts, published docs. Anthropic's own notes go in verbatim, marked as theirs. The rest is my reading, and every entry carries the strings behind it. If one looks wrong, vote it down and say why.

← All releases Home ←olderv2.1.20 v2.1.22newer→

Claude Code v2.1.21

13 entries read diff v2.1.20 → v2.1.21 Markdown

This release adds enterprise organization policy controls, improves the hook system for better SDK integration, and enhances shell completion support with automatic cache generation. The update also refines user-facing messages and adds experimental tool search optimization.

Find
Pick an entry · j / k steps through
11 entries

Changesopen

Organization Policy Limits (Enterprise)#

What

Enterprise organizations can now enforce access policies that restrict certain Claude Code features like remote sessions.

Details
  • New API endpoint /api/claude_code/policy_limits fetches organization-defined restrictions
  • Policies are cached locally at ~/.config/claude-code/policy-limits.json
  • Automatic refresh on authentication changes and periodic background polling
  • Currently enforces allow_remote_sessions policy for teleport and remote session creation
  • Only available for enterprise subscription users with proper API scopes
  • When blocked, users see: "Remote sessions are disabled by your organization's policy."
Evidence

BB() at line 388033 (enterprise check, uses subscriptionType !== "enterprise"), YP() at line 380468 (policy checker, uses "allow_remote_sessions"), IX2() at line 380287 (API endpoint builder, contains "/api/claude_code/policy_limits")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Shell Completion Cache Management#

What

Shell completion files are now automatically generated and cached, with auto-regeneration on updates.

Usage

Completion files are stored at:

  • ~/.claude/completion.bash for bash
  • ~/.claude/completion.zsh for zsh
  • ~/.claude/completion.fish for fish (stored in $XDG_CONFIG_HOME/fish/config.fish)
Details
  • Completion caches are automatically regenerated after Claude Code updates
  • Uses claude completion <shell-flag> --output <cache-file> under the hood
  • Logs show update: Regenerating <shell> completion cache and update: Regenerated <shell> completion cache at <path>
  • Improves shell startup performance by avoiding regenerating completions on every shell launch
Evidence

k69() at line 184272-184299 (shell detection, contains "completion.bash", "completion.fish", "completion.zsh"), wq6() at line 184307 (cache regeneration, contains "update: Regenerating" and "completion cache")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Enhanced Hook System for SDK Integration#

What

The compacting system now uses structured progress callbacks instead of text spinner messages, enabling better programmatic integration.

Details
  • New onCompactProgress callback with { type, hookType } events
  • Hook types include "pre_compact" and "session_start"
  • Event types: "hooks_start" when hooks begin execution
  • Replaces legacy setSpinnerMessage calls with structured events
  • UI still displays "Running PreCompact hooks…" and "Running SessionStart hooks…" but via event-driven rendering
Evidence

Line 449199 (hook start event, K.onCompactProgress?.({ type: "hooks_start", hookType: "pre_compact" })), line 449280 (session start event, hookType: "session_start"), line 571772 (UI rendering, contains "Running PreCompact hooks…")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Task List ID Ordering#

What

Task list documentation now recommends working on tasks in ID order (lowest first) when multiple tasks are available.

Details
  • New .highwatermark file for tracking task ID generation
  • Documentation updated to say "Prefer working on tasks in ID order (lowest ID first)"
  • Helps teams coordinate work more predictably
  • Part of the teammate workflow system
Evidence

Co3 = ".highwatermark" at line 155744 (highwatermark constant), string literal "Prefer working on tasks in ID order" in TaskList tool description

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Personalized Welcome Message#

What

The startup welcome message now shows which IDE or terminal you're using.

Details
  • Format changed from "Welcome to Claude Code" to "Welcome to Claude Code for <IDE name>"
  • Shows editor name like "VS Code", "Cursor", "Zed", or terminal name
  • Detected from IDE integration context or $TERM environment variable
  • Makes multi-IDE workflows clearer
Evidence

Line 306414 (welcome message rendering, "Welcome to Claude Code for ", $ where $ is IDE name from J0())

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Clearer Thinking Mode Warning#

The warning when changing thinking mode mid-conversation is now more explicit about performance impacts.

Old message: "Changing mid-conversation may reduce quality. For best results, set this at the start of a session."

New message: "Changing thinking mode mid-conversation will increase latency and may reduce quality. For best results, set this at the start of a session."

The message now mentions latency impact explicitly, helping users make informed decisions.

Evidence

Line 472714 (settings UI), line 559581 (toggle confirmation dialog), contains "Changing thinking mode mid-conversation will increase latency and may reduce quality"

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Simplified Auto Tool Search Logging#

Debug log messages for automatic tool search are now clearer and more concise.

Changes:

  • When enabled: "Auto tool search enabled: <token count>" (no more enabled/disabled toggle noise)
  • When disabled: "Auto tool search disabled: <token count>"
  • Removed redundant "threshold" and "percentage" details from enable/disable messages
  • Threshold context still present but only for character fallback edge cases
Evidence

Line 451497 (enabled message, "Auto tool search enabled: ${X}"), line 451529 (disabled message, "Auto tool search disabled: ${X}")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Improved Marketplace Installation for Headless Mode#

Plugin installation in headless/SDK mode now supports extra marketplaces and has better error reporting.

Details
  • Now installs plugins from all configured marketplaces, not just default
  • Skips plugins from unknown marketplaces with clear logging
  • Individual marketplace installation failures don't block other installations
  • Logs: installPluginsForHeadless: installed extra marketplace <name> on success
  • Logs: installPluginsForHeadless: failed to install extra marketplace <name> on failure
  • Logs: installPluginsForHeadless: skipping N plugins from unknown marketplaces: <list>
Evidence

Lines 580240-580263 (marketplace installation loop, contains "installPluginsForHeadless: installed extra marketplace" and "skipping ${J.length} plugins from unknown marketplaces")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Related

Other releases about the same thing. Found by shared names or similar wording; neither means one caused the other.

Enhanced PermissionRequest Hook Error Messages#

When a PermissionRequest hook denies an action, the error message now defaults to a clearer system message if the hook doesn't provide one.

Details
  • Default message: "Permission denied by PermissionRequest hook"
  • Helps debug which hook is blocking an operation
  • Part of the plugin hook system for security controls
Evidence

Line 579288 (hook denial, X.message || "Permission denied by PermissionRequest hook")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

CLI Init Only Flag Description Update#

The --init-only flag description now accurately reflects that it runs both Setup and SessionStart:startup hooks, not just Setup.

Old: "Run Setup hooks with init trigger, then exit" New: "Run Setup and SessionStart:startup hooks, then exit"

Evidence

Line 585008 (CLI option description, contains "Run Setup and SessionStart:startup hooks, then exit")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Deferred Tool Search Experiment (tengu_tst_kx7)#

An A/B test is running to evaluate enabling automatic tool search even when below the token threshold, if deferred tools are present in the context.

Details
  • Feature flag: tengu_tst_kx7 (controlled by GrowthBook)
  • Activates when: below normal tool search threshold BUT deferred tools exist
  • Hypothesis: Deferred tools signal potential tool use even at low token counts
  • Logs: "Tool search enabled via experiment (tengu_tst_kx7)" or "disabled via experiment"
  • Falls back silently if GrowthBook not ready: "tengu_tst_kx7: GrowthBook not ready, skipping"
Evidence

Line 451507 (experiment check, contains V4("tengu_tst_kx7", !1) and "via experiment (tengu_tst_kx7): below threshold, deferred tools present")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

1 entry

Bug Fixesopen

#

  • Fixed stream completion fallback trigger when message_start received but no content blocks completed ("Stream completed with message_start but no content blocks completed - triggering non-streaming fallback")
  • Fixed agent logging to properly identify agent in compact messages ("[inProcessRunner] ${K.agentId} compacting history" now includes agent ID)
1 entry

Notesopen

#

For SDK users: The onCompactProgress callback replaces setSpinnerMessage for hook events. Update your integration to handle the new structured events:

{
  type: "hooks_start",
  hookType: "pre_compact" | "session_start"
}

For enterprise users: If remote sessions stop working, check with your organization admin about the new allow_remote_sessions policy. This is controlled at the organization level via the policy limits API.

Verbatim
Official · Anthropic

Anthropic’s official release notes

Published verbatim by Anthropic for v2.1.21. Text is unmodified from the upstream changelog. Everything else on this page came out of the bundle instead, which is why the two lists don't match.

  • Added support for full-width (zenkaku) number input from Japanese IME in option selection prompts
  • Fixed shell completion cache files being truncated on exit
  • Fixed API errors when resuming sessions that were interrupted during tool execution
  • Fixed auto-compact triggering too early on models with large output token limits
  • Fixed task IDs potentially being reused after deletion
  • Fixed file search not working in VS Code extension on Windows
  • Improved read/search progress indicators to show "Reading…" while in progress and "Read" when complete
  • Improved Claude to prefer file operation tools (Read, Edit, Write) over bash equivalents (cat, sed, awk)
  • [VSCode] Added automatic Python virtual environment activation, ensuring python and pip commands use the correct interpreter (configurable via claudeCode.usePythonEnvironment setting)
  • [VSCode] Fixed message action buttons having incorrect background colors
Read it on GitHub →
System prompt

The system prompt was not captured for this release, so this page cannot say whether it moved.

Which releases were captured →