Claude Code · Source-level changelog

Version 2.0.73

This release introduces a new /skills command for viewing available skills, adds the /theme command as a standalone slash command, expands LSP tool support with call hierarchy features, and includes a "year-end 2025" promotional message about 2x rate limits. The SlashCommand tool has been removed, and prompt suggestion filtering has been significantly improved.

Package @anthropic-ai/claude-code Diff v2.0.72 → v2.0.73
13
Features & Changes
2
Bug Fixes
0
In Development
0
Env Vars / Flags

Changes

13 items

/skills Command #

What

New command to view all available skills and their sources.

Usage
/skills
Details
  • Displays skills grouped by source (policy settings, user settings, project settings, plugins)
  • Shows skill names, estimated token counts, and source locations
  • Skill locations displayed: ~/.claude/skills/, .claude/skills/, or plugin sources
  • Skills with user-invocable: false in their frontmatter are now hidden from user invocation
Evidence

YX9() at line 503132 (skills dialog, contains "Skills", "skill")

/theme Command (Standalone) #

What

The theme picker is now available as a standalone /theme command.

Usage
/theme
Details
  • Previously only accessible via /settings, now directly invocable
  • Allows quick theme changes without navigating through settings
Evidence

FY7() at line 503626 (theme command, contains "theme", "Theme picker")

LSP Call Hierarchy Support #

What

New LSP operations for exploring function call relationships.

Usage

The LSP tool now supports three new operations:

  • prepareCallHierarchy: Get call hierarchy item at a position
  • incomingCalls: Find all functions/methods that call the function at a position
  • outgoingCalls: Find all functions/methods called by the function at a position
Details
  • Results are grouped by file for easier navigation
  • Shows call sites with line numbers
  • Useful for understanding code flow and dependencies
Evidence

WL0 at line 481308 (LSP tool prompt, contains "prepareCallHierarchy", "incomingCalls", "outgoingCalls")

Year-End 2025 Rate Limit Promotion #

What

A promotional message announcing 2x higher rate limits through December 31, 2025.

Details
  • Displayed via the tengu_year_end_2025_campaign_promo feature flag
  • Message: "A gift for you - Your rate limits are 2x higher through 12/31"
Evidence

_q0() at line 464135 (promo message, contains "2x higher", "12/31")

Kill Ring Yank Cycling (Meta-y) #

What

Added Emacs-style yank cycling with Meta-y to rotate through killed text history.

Usage

After yanking with Ctrl-y, press Meta-y (Alt-y on most terminals) to cycle through previously killed text.

Details
  • Stores up to 10 killed text entries
  • Cycles through the kill ring each time Meta-y is pressed
  • Follows standard Emacs keybinding conventions
Evidence

BA() function at line 521106 (yank cycling handler, key mapping ["y", BA] in Meta handlers)

SlashCommand Tool Removed #

The SlashCommand tool has been removed. Slash commands are now invoked directly by the user or through the Skill tool. This simplifies the tool surface and removes redundant functionality.

Evidence

Removed $S = "SlashCommand" at line 424048, removed ir = { name: $S, ...} tool definition

Prompt Suggestion Filtering Enhanced #

Significantly improved filtering of unsuitable prompt suggestions with new rejection criteria:

  • Word count limits: Suggestions must be 2-8 words
  • No evaluative phrases: Filters out "thanks", "looks good", "sounds good", etc.
  • No Claude voice: Rejects suggestions starting with "let me", "I'll", "here's", etc.
  • No multiple sentences: Suggestions with sentence boundaries are rejected
  • Added early exit conditions for pending permissions, active elicitations, and plan mode
Evidence

km5() at line 404234 (prompt suggestion filter, contains "claude_voice", "evaluative", "too_few_words")

Claude 3.5 Haiku Deprecation Notice #

Added deprecation warning for Claude 3.5 Haiku with retirement date of February 19, 2026 (first-party API only).

Evidence

d97 at line 476533 (deprecation config, contains "claude-3-5-haiku", "February 19, 2026")

Image Cache with Session Cleanup #

Images pasted during a session are now cached to disk for potential later reference, with automatic cleanup of old session image caches.

Evidence

ys2() at line 454701 (image cache cleanup), Rs2 = "image-cache" at line 454721

MCP Tool Search Feature Flag #

Tool search for MCP can now be enabled via the tengu_mcp_tool_search feature flag, expanding the tool search functionality to MCP servers.

Evidence

zIA() at line 259974 (tool mode selector, contains 'if (aU("tengu_mcp_tool_search", !1)) return "tst"')

Skills Frontmatter: user-invocable #

Skills now support a user-invocable frontmatter field. When set to false, the skill is hidden from user-facing command lists but can still be invoked by the model.

Evidence

BX9() at line 502839 (skill builder, contains "userInvocable", "user-invocable")

Improved Team Task Handling on Teammate Termination #

When a teammate is terminated or shuts down, their open tasks are now properly unassigned and a notification is sent listing the affected tasks for reassignment.

Evidence

a11() at line 207479 (task unassignment handler, contains "was terminated", "unassigned")

Deep Search Removed from Resume Dialog #

The "deep search" (: prefix) feature for searching within message content has been removed from the session resume dialog, simplifying the search interface.

Evidence

Removed tengu_session_deep_search_toggled telemetry event at line 8130

Bug Fixes

2 items
  • Fixed hooks from plugins being incorrectly loaded when hooks are restricted by policy (aI7() at line 523289, checks pluginName before including hooks)
  • Fixed update command to support installing specific versions via --version flag (LDA() at line 524409, rZA() at line 146572)

Notes

The removal of the SlashCommand tool means that models can no longer programmatically invoke slash commands on behalf of users. Skills and custom commands should be invoked directly by users or through the Skill tool for model-initiated invocation.