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 ←olderv2.1.12 v2.1.15newer→

Claude Code v2.1.14

16 entries read diff v2.1.12 → v2.1.14 Markdown

This release renames MCPSearch to ToolSearch to better reflect its broader tool discovery capabilities, adds support for dynamically discovered skills from .claude/skills directories, introduces a new CLAUDE_CODE_GLOB_TIMEOUT_SECONDS environment variable for controlling glob timeouts, and simplifies the ExitPlanMode tool documentation. Additionally, Claude.ai MCP server integration has been expanded with improved telemetry and eligibility checking.

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

Changesopen

ToolSearch (renamed from MCPSearch)#

What

The MCPSearch tool has been renamed to ToolSearch to better reflect its ability to search for and load both MCP tools and deferred built-in tools.

Usage: The tool works the same way but with the new name:

ToolSearch with query: "select:NotebookEdit"
ToolSearch with query: "slack"
Details
  • Direct selection mode: Use select:<tool_name> for exact tool loading
  • Keyword search: Use natural language to find relevant tools
  • Now supports loading deferred built-in tools in addition to MCP tools
  • Documentation updated throughout to reference the new name
Evidence

aj at line 280183 (tool name constant, contains "ToolSearch")

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.

Configurable Glob Timeout#

What

A new environment variable CLAUDE_CODE_GLOB_TIMEOUT_SECONDS allows customizing the timeout for glob/ripgrep operations.

Usage
export CLAUDE_CODE_GLOB_TIMEOUT_SECONDS=30
claude
Details
  • Default timeout remains 20 seconds (60 seconds on WSL)
  • Set a custom value in seconds to override the default
  • Useful for large codebases where glob operations may take longer
Evidence

r0Q() at line 30721 (ripgrep executor, contains "CLAUDE_CODE_GLOB_TIMEOUT_SECONDS")

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.

Dynamic Skill Discovery#

What

Skills can now be discovered dynamically from .claude/skills directories at any level of the project hierarchy.

Details
  • Scans for SKILL.md files in subdirectories of .claude/skills
  • Skills are discovered based on directory triggers during operation
  • Enables projects to define reusable skills that Claude can invoke
  • Skills defined in parent directories are also discovered
Evidence

vl5() at line 373896 (dynamic skill discovery, references "SKILL.md" and "dynamic_skill")

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

Skills with Required MCP Server Dependencies#

What

Skills can now specify required MCP servers and will only be available when those servers are connected.

Details
  • Skills can declare requiredMcpServers as dependencies
  • Skill availability is filtered based on connected MCP servers
  • Enables context-aware skill loading
Evidence

UV0() at line 313441 (MCP server requirement check, contains "requiredMcpServers")

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

WebFetch GitHub URL Guidance#

What

WebFetch tool documentation now recommends using the gh CLI for GitHub URLs.

Details
  • New guidance: "For GitHub URLs, prefer using the gh CLI via Bash instead (e.g., gh pr view, gh issue view, gh api)"
  • Reduces reliance on web scraping for GitHub content
  • Leverages the more reliable gh CLI for GitHub operations
Evidence

AwQ at line 96805 (WebFetch tool description, contains "gh pr view, gh issue view, gh api")

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

Simplified ExitPlanMode Tool Description#

What

The ExitPlanMode tool documentation has been simplified by removing the allowedPrompts permission request section.

Details
  • Removed detailed instructions about requesting Bash permissions during plan approval
  • Streamlines the plan mode workflow
  • Documentation now focuses on core functionality
Evidence

Zn2 at line 416019 (ExitPlanMode description, no longer contains "allowedPrompts")

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

Claude.ai MCP Server Integration Improvements#

What

Enhanced Claude.ai MCP server integration with better eligibility checking and telemetry.

Details
  • New eligibility states tracked: disabled_gate, disabled_env_var, no_oauth_token, missing_scope, eligible
  • Requires user:mcp_servers OAuth scope
  • Can be disabled via ENABLE_CLAUDEAI_MCP_SERVERS environment variable
  • Improved logging for debugging MCP server connectivity
Evidence

Q_0 at line 458270 (Claude.ai MCP initialization, contains "tengu_claudeai_mcp_eligibility")

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

Prompt Suggestion Mode Improvements#

Feature flag
tengu_plank_river_frost Not enough to say

Nothing here resolved what this flag was doing on this version, so nothing here should be read as on or off.

This account: no value returned · anonymous baseline: no value returned · compiled default in v2.1.14: no gate table built for this version

These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.

Read once, for one account on one subscription tier, against v2.1.14. It isn't a statement about your account. What a flag value here can and cannot tell you

What

The prompt suggestion system now supports a new "stated intent" mode in addition to the existing "user intent" mode.

Details
  • New stated_intent mode focuses on finding explicitly stated next steps from user messages
  • Looks for multi-part requests ("do X and Y"), stated intent ("then I'll Z"), and answers to Claude's questions
  • Controlled by tengu_plank_river_frost feature flag
  • Provides more conservative suggestions based on user's explicit plans
Evidence

ls5 at line 395184 (stated intent prompt, contains "stated_intent" and "SEARCH FOR")

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

Git Clone with SHA Support#

What

Git clone operations now support checking out specific commit SHAs in addition to branches.

Details
  • Uses --no-checkout flag followed by SHA fetch and checkout
  • Falls back to --unshallow fetch if shallow fetch of SHA fails
  • Enables working with specific commit references
Evidence

FU7() at line 497957 (git clone function, contains "--no-checkout" and "checkout")

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.

Debug Log Cleanup#

What

Added automatic cleanup of old debug log files.

Details
  • Cleans up .txt files from the debug directory
  • Removes files older than a threshold date
  • Helps manage disk space from accumulated debug logs
Evidence

Tj7() at line 530506 (debug cleanup function, accesses "debug" directory)

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

Tool Usage Tracking for Deferred Tools#

What

Added tracking infrastructure for deferred tool usage patterns.

Details
  • Tracks usageCount and lastUsedAt for tools
  • Enables smarter tool deferral decisions based on usage frequency
  • Similar tracking added for skill usage with time-decay weighting
Evidence

R22() at line 280135 (tool usage tracker, contains "toolUsage", "usageCount", "lastUsedAt")

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

Proxy Configuration Improvements#

What

Added support for no_proxy and NO_PROXY environment variables.

Details
  • New function to read proxy bypass configuration
  • Improves compatibility with corporate network environments
Evidence

kx3() at line 155303 (proxy config reader, contains "no_proxy" and "NO_PROXY")

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

1 entry

Internal Changesopen

#

  • Refactored ripgrep execution to use spawn with better timeout handling
  • Added REPL tool type constant for future use
  • Improved border style handling in UI components
  • Various code organization and cleanup changes
Verbatim
Official · Anthropic

Anthropic’s official release notes

Published verbatim by Anthropic for v2.1.14. 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 history-based autocomplete in bash mode (!) - type a partial command and press Tab to complete from your bash command history
  • Added search to installed plugins list - type to filter by name or description
  • Added support for pinning plugins to specific git commit SHAs, allowing marketplace entries to install exact versions
  • Fixed a regression where the context window blocking limit was calculated too aggressively, blocking users at ~65% context usage instead of the intended ~98%
  • Fixed memory issues that could cause crashes when running parallel subagents
  • Fixed memory leak in long-running sessions where stream resources were not cleaned up after shell commands completed
  • Fixed @ symbol incorrectly triggering file autocomplete suggestions in bash mode
  • Fixed @-mention menu folder click behavior to navigate into directories instead of selecting them
  • Fixed /feedback command generating invalid GitHub issue URLs when description is very long
  • Fixed /context command to show the same token count and percentage as the status line in verbose mode
  • Fixed an issue where /config, /context, /model, and /todos command overlays could close unexpectedly
  • Fixed slash command autocomplete selecting wrong command when typing similar commands (e.g., /context vs /compact)
  • Fixed inconsistent back navigation in plugin marketplace when only one marketplace is configured
  • Fixed iTerm2 progress bar not clearing properly on exit, preventing lingering indicators and bell sounds
  • Improved backspace to delete pasted text as a single token instead of one character at a time
  • [VSCode] Added /usage command to display current plan usage
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 →