Claude Code · Source-level changelog

Version 1.0.112

Version 1.0.112 introduces skill.md file support for organizing commands hierarchically, adds enhanced network sandboxing with allow-lists, and removes unused internal network sandboxing code. Several improvements enhance command parsing, error tracking, and user configuration options.

Package @anthropic-ai/claude-code Diff v1.0.111 → v1.0.112
9
Features & Changes
0
Bug Fixes
0
In Development
0
Env Vars / Flags

Changes

9 items

skill.md File Support #

What: Organize multiple commands in a directory under a single skill.md file How to use:

# Create a directory with related commands
mkdir my-commands/
echo "# My Skill" > my-commands/skill.md
echo "command content" > my-commands/command1.md

# Claude Code will now treat all .md files in this directory as part of the skill

Details:

  • When multiple command files exist in a directory with a skill.md file, Claude Code prioritizes the skill.md
  • Case-insensitive matching (skill.md, Skill.md, SKILL.MD all work)
  • Helps organize complex command structures hierarchically
  • Evidence: uM0() at line 424817 and pZ5() at line 424820 in v1.0.112

Enhanced Network Sandboxing with Allow-Lists #

What: Fine-grained network access control for sandboxed bash commands How to use:

# Network sandboxing now supports allow-lists for specific hosts
# Configured through internal settings: sandbox.network.allow

Details:

  • NetworkManager filters connections based on configured allow-list
  • Blocks all network access except explicitly allowed hosts
  • Logs allowed and blocked connections for transparency
  • Works with HTTP proxy for controlled network access
  • Evidence: NetworkManager implementation at line 373982-374129 in v1.0.112

Automatic Feature Flag Updates #

What: Statsig feature flags now refresh automatically every 6 hours Details:

  • Keeps feature flags up-to-date without restarting the CLI
  • Runs in background, transparent to users
  • Automatically disabled when using Bedrock/Vertex or when telemetry is disabled
  • Evidence: xqB() at line 428347 in v1.0.112

Configurable Spinner Tips #

What changed: Added ability to disable spinner tips through configuration How it works now:

# Configure in settings to disable tips during long operations
# Set spinnerTipsEnabled: false in configuration

Evidence: NrB() at line 437820

Enhanced Command Parsing and Validation #

What changed: Better error handling and validation before command execution Benefits:

  • Clearer error message: "Command cannot be parsed" for malformed commands
  • Improved detection of input redirection patterns (e.g., < file.txt)
  • More robust tokenization of complex commands with special characters
  • Evidence: lAB() at line 384563, qz0() at line 373516, ut2() at line 373524

Better Error Tracking #

What changed: Error reports now include unique error IDs Benefits:

  • Easier to track and debug specific errors when reporting issues
  • Better correlation between user reports and backend logs
  • Evidence: Rh1() at line 428069 and U1() at line 428104

Improved Native Installation Detection #

What changed: Prevents auto-migration prompts for users already on native installations Benefits:

  • No more unnecessary migration notifications
  • Cleaner user experience for native install users
  • Evidence: Lm1() at line 435204

Streamlined Terminal Output #

What changed: More consistent terminal output rendering Benefits:

  • Fewer visual glitches in terminal display
  • Better structured output handling
  • Evidence: IC9 variable at line 361107

Internal Network Sandboxing (macOS sandbox-exec) #

What was removed: Unused internal network sandboxing implementation Details:

  • Removed ST6() (sandbox-exec profile generator) at line 373486 in v1.0.111
  • Removed ct2() (macOS command wrapper) at line 373499 in v1.0.111
  • Removed at2() (platform-agnostic sandbox orchestrator) at line 373547 in v1.0.111
  • This was internal-only functionality not exposed to users
  • Replaced by the enhanced NetworkManager implementation