Enhanced Claude Code Provider with Codebase-Aware Task Generation

Minor Changes

• #1098 36468f3 Thanks @Crunchyman-ralph! - Enhanced Claude Code provider with codebase-aware task generation

  • Added automatic codebase analysis for Claude Code provider in parse-prd, expand-task, and analyze-complexity commands
  • When using Claude Code as the AI provider, Task Master now instructs the AI to analyze the project structure, existing implementations, and patterns before generating tasks or subtasks
  • Tasks and subtasks generated by Claude Code are now informed by actual codebase analysis, resulting in more accurate and contextual outputs

• #1105 75c514c Thanks @Crunchyman-ralph! - Add GPT-5 support with proper parameter handling

  • Added GPT-5 model to supported models configuration with SWE score of 0.749

• #1091 4bb6370 Thanks @Crunchyman-ralph! - Add Claude Code subagent support with task-orchestrator, task-executor, and task-checker

  New Claude Code Agents

  Added specialized agents for Claude Code users to enable parallel task execution, intelligent task orchestration, and quality assurance:

  task-orchestrator

  Coordinates and manages the execution of Task Master tasks with intelligent dependency analysis:

  • Analyzes task dependencies to identify parallelizable work
  • Deploys multiple task-executor agents for concurrent execution
  • Monitors task completion and updates the dependency graph
  • Automatically identifies and starts newly unblocked tasks

  task-executor

  Handles the actual implementation of individual tasks:

  • Executes specific tasks identified by the orchestrator
  • Works on concrete implementation rather than planning
  • Updates task status and logs progress
  • Can work in parallel with other executors on independent tasks

  task-checker

  Verifies that completed tasks meet their specifications:

  • Reviews tasks marked as 'review' status
  • Validates implementation against requirements
  • Runs tests and checks for best practices
  • Ensures quality before marking tasks as 'done'

  Installation

  When using the Claude profile (task-master rules add claude), the agents are automatically installed to .claude/agents/ directory.

  Usage Example

  # In Claude Code, after initializing a project with tasks:
  
  # Use task-orchestrator to analyze and coordinate work
  # The orchestrator will:
  # 1. Check task dependencies
  # 2. Identify tasks that can run in parallel
  # 3. Deploy executors for available work
  # 4. Monitor progress and deploy new executors as tasks complete
  
  # Use task-executor for specific task implementation
  # When the orchestrator identifies task 2.3 needs work:
  # The executor will implement that specific task
  

  Benefits

  • Parallel Execution: Multiple independent tasks can be worked on simultaneously
  • Intelligent Scheduling: Orchestrator understands dependencies and optimizes execution order
  • Separation of Concerns: Planning (orchestrator) is separated from execution (executor)
  • Progress Tracking: Real-time updates as tasks are completed
  • Automatic Progression: As tasks complete, newly unblocked tasks are automatically started

Patch Changes

• #1094 4357af3 Thanks @Crunchyman-ralph! - Fix expand task generating unrelated generic subtasks

  Fixed an issue where task-master expand would generate generic authentication-related subtasks regardless of the parent task context when using complexity reports. The expansion now properly includes the parent task details alongside any expansion guidance.

• #1079 e495b2b Thanks @Crunchyman-ralph! - Fix scope-up/down prompts to include all required fields for better AI model compatibility

  • Added missing priority field to scope adjustment prompts to prevent validation errors with Claude-code and other models
  • Ensures generated JSON includes all fields required by the schema

• #1079 e495b2b Thanks @Crunchyman-ralph! - Fix MCP scope-up/down tools not finding tasks

  • Fixed task ID parsing in MCP layer - now correctly converts string IDs to numbers
  • scope_up_task and scope_down_task MCP tools now work properly

• #1079 e495b2b Thanks @Crunchyman-ralph! - Improve AI provider compatibility for JSON generation

  • Fixed schema compatibility issues between Perplexity and OpenAI o3 models
  • Removed nullable/default modifiers from Zod schemas for broader compatibility
  • Added automatic JSON repair for malformed AI responses (handles cases like missing array values)
  • Perplexity now uses JSON mode for more reliable structured output
  • Post-processing handles default values separately from schema validation

Extension 0.23.1

Patch Changes

• #1090 a464e55 Thanks @Crunchyman-ralph! - Fix issues with some users not being able to connect to Taskmaster MCP server while using the extension