Claude Code · Source-level changelog

Version 1.0.94

Version 1.0.94 introduces comprehensive performance monitoring with new warnings for large agent descriptions and MCP tools, significantly improves security with sophisticated sed command validation, and enhances the permission system with dynamic update capabilities.

Package @anthropic-ai/claude-code Diff v1.0.93 → v1.0.94
10
Features & Changes
0
Bug Fixes
0
In Development
0
Env Vars / Flags

Changes

10 items

Performance Context Warnings #

What: New warning system that monitors context consumption from custom agents and MCP tools How to use:

# Warnings appear automatically when thresholds are exceeded
# Agent descriptions > 15,000 tokens will trigger a warning
# MCP tools > 25,000 tokens will trigger a warning

Details:

  • Monitors cumulative token count of custom agent whenToUse descriptions
  • Tracks total context consumed by MCP tool definitions
  • Shows top contributors by token consumption for optimization
  • Integrated with existing CLAUDE.md size warnings into unified system
  • Evidence: kY1() at line 399348, XV5() at line 399376, FV5() at line 399400, WSB() at line 399442

Advanced sed Command Security #

What: Complete replacement of sed command validation with sophisticated parsing that blocks dangerous operations How to use:

# Safe sed commands work normally
claude "Use sed to replace text in file.txt"

# Dangerous operations are now blocked:
# - File writes (s///w, w command)
# - Command execution (s///e, e command)
# - Multiple expression exploits

Details:

  • Prevents sed commands from writing to files or executing shell commands
  • Parses actual sed expressions instead of using regex patterns
  • Detects and blocks write operations (w), execute operations (e), and dangerous flags
  • Evidence: P1B() at line 379595, mS6() at line 379676, hS6() at line 379611, gS6() at line 379638

Dynamic Permission Updates #

What: Tools can now suggest and apply permission changes during execution Details:

  • Permission prompts can include tool-suggested rules
  • "Yes, and apply suggestions" option for streamlined workflows
  • Automatic directory access rule creation when accessing folders
  • Context-aware permission suggestions based on access patterns
  • Evidence: qW1() at line 430572, Cz1() at line 348277, CR5() at line 421430

Enhanced Error Messages #

What: Improved error handling for PDF files Details:

  • Clear message when PDF exceeds page limits
  • Guidance to edit message and retry with smaller PDF
  • Evidence: bb6 variable at line 391166 in error handler B_1()

Checkpointing Status Display #

What: Better visibility into checkpointing errors and status Details:

  • Shows checkpointing errors in status display
  • Indicates when checkpointing is disabled
  • Displays save failures with error messages
  • Evidence: KV5() at line 399769

Memory File Selection UI #

What: Redesigned memory file selector interface Details:

  • Improved organization of memory file options
  • Better visual hierarchy for nested imports
  • Clearer descriptions of file locations
  • Evidence: sb1() at line 400461 replacing hb1()

Git Command Flag Handling #

What: Fixed handling of numeric flags in git commands Details:

  • Git commands with numeric flags like git log -5 now work correctly
  • Improved flag parsing to distinguish between numeric flags and invalid options
  • Evidence: lS6() at line 380053, added check for /^-\d+$/ pattern at line 1404

Model Deprecation Updates #

What: Added deprecation dates for additional Claude models Details:

  • Added claude-3-opus-20240229 with deprecation date January 5th, 2026
  • Added claude-3-5-sonnet-20241022 with deprecation date October 22, 2025
  • Added claude-3-5-sonnet-20240620 with deprecation date October 22, 2025
  • Evidence: My2 object at line 366970, jy2 object at line 367622

Settings Processing #

What: Enhanced --settings flag processing during initialization Details:

  • Settings are now processed earlier in the startup sequence
  • Supports both JSON strings and file paths
  • Better error messages for invalid settings
  • Evidence: wS5() at line 433801, initialization check at lines 1537-1541

Local Memory Type Selection #

What: Simplified memory file selection by removing separate "Local" memory type Details:

  • CLAUDE.local.md functionality appears to be consolidated
  • Memory selection UI simplified from three types to unified approach
  • Evidence: Removed functions cv1(), fb1(), VyB variable, hb1(), rz5(), AD5()