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 ←olderv2.1.8 v2.1.10newer→

Claude Code v2.1.9

10 entries read diff v2.1.8 → v2.1.9 Markdown

This release adds Opus 4.5 upgrade nudges for users on older Opus models, introduces a new plansDirectory setting to customize where plan files are stored, enhances MCP tool search with better scoring for server and action names, and includes new telemetry for SSL/TLS certificate configurations. Internal improvements include a new HybridTransport for remote sessions with retry logic.

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

Changesopen

Plans Directory Configuration#

What

You can now specify a custom directory for plan files via the plansDirectory setting.

Usage
{
  "plansDirectory": "./plans"
}
Details
  • The directory must be within the project root
  • If not specified, plans are stored in the default session directory
  • Useful for keeping plan files in version control alongside your project
Evidence

jN() at line 300782 (plans directory resolution, contains "plansDirectory must be within project root")

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.

Opus 4.5 Upgrade Nudge#

What

Users on older Opus 4.x models (Opus 4.0, 4.1) now see a notification suggesting they upgrade to Opus 4.5.

Details
  • Only shows for first-party users (claude.ai)
  • Displays: "Opus 4.5 is available and is our best model for coding"
  • Includes a hint to use /model command to upgrade
  • Nudge appears for ~10 seconds then dismisses
Evidence

vN9() at line 535826 (opus upgrade nudge, contains "opus-45-upgrade-nudge", "Opus 4.5 is available and is our best model for coding")

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

ENABLE_TOOL_SEARCH Percentage Control#

What

The ENABLE_TOOL_SEARCH environment variable now supports percentage-based control with the auto:N syntax.

Usage
ENABLE_TOOL_SEARCH=auto:50  # 50% tool search probability
ENABLE_TOOL_SEARCH=auto:0   # Equivalent to "tst" (always tool search)
ENABLE_TOOL_SEARCH=auto:100 # Equivalent to standard/mcp-cli modes
Details
  • Values are clamped between 0-100
  • Invalid values log a warning and fall back to defaults
  • Provides fine-grained control over tool search behavior
Evidence

DX0() at line 292496 (auto percentage parser, contains "Invalid ENABLE_TOOL_SEARCH value", "expected auto:N where N is a number")

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

Enhanced MCP Tool Search Scoring#

Tool search now provides better relevance scoring for MCP tools by parsing server and action names from the tool name format (mcp__server__action_name).

Details
  • Server name exact matches: +12 points
  • Server name partial matches: +6 points
  • Action name exact matches: +8 points
  • Action name partial matches: +4 points
  • Full name matches: +3 points
  • Description word boundary matches: +2 points
Evidence

EQ7() at line 428976 (MCP tool name parser, uses regex mcp__)

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

HybridTransport for Remote Sessions#

Remote sessions now use a hybrid transport that sends messages via HTTP POST with WebSocket fallback, improving reliability.

Details
  • Implements retry logic with exponential backoff (up to 10 attempts)
  • Converts WebSocket URLs to HTTP POST endpoints for events API
  • Client errors (4xx except 429) are not retried
  • Maximum backoff delay of 8 seconds
Evidence

Yv0 class at line 542866 (HybridTransport, contains "HybridTransport: POST URL", "cli_hybrid_transport_initialized")

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.

Staging Environment Detection#

Sessions can now properly detect and connect to staging environments (staging.claude.ai) for internal testing.

Evidence

Q65() at line 298312 (staging detection, contains "_staging_", "https://staging.claude.ai")

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

SSL/TLS Certificate Telemetry#

New telemetry tracks SSL/TLS certificate configuration for debugging connectivity issues.

Details
  • Detects NODE_EXTRA_CA_CERTS environment variable
  • Detects CLAUDE_CODE_CLIENT_CERT for mTLS
  • Detects --use-system-ca and --use-openssl-ca Node.js options
Evidence

wT7() at line 545895 (certificate telemetry, contains "has_node_extra_ca_certs", "has_use_system_ca")

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.

Session Phase Status Tracking#

Internal support for tracking session phases: planning, implementing, reviewing, verifying.

Evidence

Vd5() at line 373496 (phase status mapper, contains "planning", "implementing", "reviewing", "verifying")

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

Improved Screen Buffer Reuse#

The terminal rendering system now reuses screen buffers for better memory efficiency, reducing allocations during rendering.

Evidence

C3B() at line 187260 (resetScreen function, screen buffer management)

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

2 entries

Bug Fixesopen

#

  • Fixed MCP client cleanup to properly handle timeouts during disconnect (bd() at line 296688, 5-second timeout for cleanup)
  • Improved screen array handling to prevent resizing issues in edge cases

Related

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

/btw Side Questions [In Development]#

What

Infrastructure for asking quick side questions without interrupting the main conversation.

Status

Stubbed - isEnabled: () => !1 (always returns false)

Details
  • Command exists: /btw <question>
  • Full UI component exists with loading spinner
  • Error handling for failed responses
  • Dismiss with Enter or Escape
  • The command type changed from "local" to "local-jsx" (JSX component rendering)
Evidence

eB7 at line 437798 (btw command, isEnabled: () => !1, contains "/btw", "Answering...")

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.

Verbatim
Official · Anthropic

Anthropic’s official release notes

Published verbatim by Anthropic for v2.1.9. 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 auto:N syntax for configuring the MCP tool search auto-enable threshold, where N is the context window percentage (0-100)
  • Added plansDirectory setting to customize where plan files are stored
  • Added external editor support (Ctrl+G) in AskUserQuestion "Other" input field
  • Added session URL attribution to commits and PRs created from web sessions
  • Added support for PreToolUse hooks to return additionalContext to the model
  • Added ${CLAUDE_SESSION_ID} string substitution for skills to access the current session ID
  • Fixed long sessions with parallel tool calls failing with an API error about orphan tool_result blocks
  • Fixed MCP server reconnection hanging when cached connection promise never resolves
  • Fixed Ctrl+Z suspend not working in terminals using Kitty keyboard protocol (Ghostty, iTerm2, kitty, WezTerm)
Read it on GitHub →
System prompt

2 added and 2 removed, of 144 lines, about 131 words, in the prompt all 27 arms receive.

Agent SDK, print mode

Read the prompts for this release →