Claude Code v1.0.81

What: Complete plugin framework for extending Claude Code with custom commands, agents, and hooks How to use:

# Enable plugin system (currently requires environment variable)
export ENABLE_PLUGINS=1
claude

# Plugins are loaded from ~/.claude/plugins/repos/
# Structure: ~/.claude/plugins/repos/owner/repo/

Details:

• Supports three types of extensions: commands, agents, and hooks
• Git-based distribution with automatic updates via git pull --ff-only
• Plugin manifest format (plugin.json) with name, version, description, and author metadata
• Commands and agents are namespaced (e.g., pluginName:commandName)
• Individual plugins can be enabled/disabled per repository
• Supports ${CLAUDE_PLUGIN_ROOT} variable for referencing plugin files
• Currently disabled by default - requires ENABLE_PLUGINS environment variable

What: Infrastructure for caching API prompts to local JSON files Details:

• Saves prompts to .claude-cache/prompts/{sessionId}/{timestamp}.json
• Currently disabled via hardcoded flag - likely for future release
• Captures complete prompt data including tools, system prompts, and messages
• Silent error handling to prevent interruption of normal operations

What changed: Concurrent queries are now blocked with user feedback instead of causing undefined behavior Impact: Prevents race conditions and provides clear messaging when users submit queries too quickly. Shows "Previous query still processing. Please try again." warning message.

What changed: Automatic cache refresh when output styles are created or modified Impact: New output styles are immediately available without requiring a restart. The output-style-setup agent now includes a callback that clears the style cache after successful operations.

What changed: Added categorization for agent and output style usage tracking Impact: Better insights into feature usage patterns with categories like:

• agent:default, agent:custom, agent:{agentType} for agent usage
• outputStyle:default, outputStyle:custom, outputStyle:{styleName} for output styles

What changed: Agent discovery now supports plugin-provided agents in addition to built-in and custom agents Impact: Plugin agents appear seamlessly integrated in the agent list with "Plugin: {pluginName}" attribution

• Issue: Users could submit multiple queries simultaneously, causing unpredictable behavior
• Cause: Previous implementation only logged telemetry without preventing concurrent execution
• Resolution: Now blocks concurrent queries with proper mutex protection and user feedback

• Issue: Newly created output styles weren't immediately available
• Cause: Output style cache wasn't cleared after modifications
• Resolution: Added callback mechanism to clear cache after output-style-setup agent completes