Releases: eyaltoledano/claude-task-master
task-master-ai@0.30.1-rc.0
685a8cc
Crunchyman-ralph
Ralph Khreish
task-master-ai@0.30.0
f98e0db
Crunchyman-ralph
Ralph Khreish
Minor Changes
-
#1181
a69d8c9Thanks @karol-f! - Add configurable MCP tool loading to optimize LLM context usageYou can now control which Task Master MCP tools are loaded by setting the
TASK_MASTER_TOOLSenvironment variable in your MCP configuration. This helps reduce context usage for LLMs by only loading the tools you need.Configuration Options:
all(default): Load all 36 toolscoreorlean: Load only 7 essential tools for daily development- Includes:
get_tasks,next_task,get_task,set_task_status,update_subtask,parse_prd,expand_task
- Includes:
standard: Load 15 commonly used tools (all core tools plus 8 more)- Additional tools:
initialize_project,analyze_project_complexity,expand_all,add_subtask,remove_task,generate,add_task,complexity_report
- Additional tools:
- Custom list: Comma-separated tool names (e.g.,
get_tasks,next_task,set_task_status)
Example .mcp.json configuration:
{ "mcpServers": { "task-master-ai": { "command": "npx", "args": ["-y", "task-master-ai"], "env": { "TASK_MASTER_TOOLS": "standard", "ANTHROPIC_API_KEY": "your_key_here" } } } }For complete details on all available tools, configuration examples, and usage guidelines, see the MCP Tools documentation.
-
#1312
d7fca18Thanks @Crunchyman-ralph! - Improve next command to work with remote -
#1317
548beb4Thanks @Crunchyman-ralph! - Add 4.5 haiku and sonnet to supported models for claude-code and anthropic ai providers -
#1309
ccb87a5Thanks @Crunchyman-ralph! - Add autonomous TDD workflow automation system with newtm autopilotcommands and MCP tools for AI-driven test-driven development.New CLI Commands:
tm autopilot start <taskId>- Initialize TDD workflowtm autopilot next- Get next action in workflowtm autopilot status- Check workflow progresstm autopilot complete- Advance phase with test resultstm autopilot commit- Save progress with metadatatm autopilot resume- Continue from checkpointtm autopilot abort- Cancel workflow
New MCP Tools:
Seven new autopilot tools for programmatic control:autopilot_start,autopilot_next,autopilot_status,autopilot_complete_phase,autopilot_commit,autopilot_resume,autopilot_abortFeatures:
- Complete RED → GREEN → COMMIT cycle enforcement
- Intelligent commit message generation with metadata
- Activity logging and state persistence
- Configurable workflow settings via
.taskmaster/config.json - Comprehensive AI agent integration documentation
Documentation:
- AI Agent Integration Guide (2,800+ lines)
- TDD Quick Start Guide
- Example prompts and integration patterns
Learn more: TDD Workflow Quickstart Guide
This release enables AI agents to autonomously execute test-driven development workflows with full state management and recovery capabilities.
Patch Changes
-
#1314
6bc75c0Thanks @Crunchyman-ralph! - Improve auth token refresh flow -
#1302
3283506Thanks @bjcoombs! - Enable Task Master commands to traverse parent directories to find project root from nested pathsFixes #1301
-
#1323
dc6652cThanks @Crunchyman-ralph! - Fix MCP server compatibility with Draft-07 clients (Augment IDE, gemini-cli, gemini code assist)- Resolves #1284
Problem:
- MCP tools were using Zod v4, which outputs JSON Schema Draft 2020-12
- MCP clients only support Draft-07
- Tools were not discoverable in gemini-cli and other clients
Solution:
- Updated all MCP tools to import from
zod/v3instead ofzod - Zod v3 schemas convert to Draft-07 via FastMCP's zod-to-json-schema
- Fixed logger to use stderr instead of stdout (MCP protocol requirement)
This is a temporary workaround until FastMCP adds JSON Schema version configuration.
task-master-ai@0.30.0-rc.1
Minor Changes
-
#1317
548beb4Thanks @Crunchyman-ralph! - Add 4.5 haiku and sonnet to supported models for claude-code and anthropic ai providers -
#1309
ccb87a5Thanks @Crunchyman-ralph! - Add autonomous TDD workflow automation system with newtm autopilotcommands and MCP tools for AI-driven test-driven development.New CLI Commands:
tm autopilot start <taskId>- Initialize TDD workflowtm autopilot next- Get next action in workflowtm autopilot status- Check workflow progresstm autopilot complete- Advance phase with test resultstm autopilot commit- Save progress with metadatatm autopilot resume- Continue from checkpointtm autopilot abort- Cancel workflow
New MCP Tools:
Seven new autopilot tools for programmatic control:autopilot_start,autopilot_next,autopilot_status,autopilot_complete_phase,autopilot_commit,autopilot_resume,autopilot_abortFeatures:
- Complete RED → GREEN → COMMIT cycle enforcement
- Intelligent commit message generation with metadata
- Activity logging and state persistence
- Configurable workflow settings via
.taskmaster/config.json - Comprehensive AI agent integration documentation
Documentation:
- AI Agent Integration Guide (2,800+ lines)
- TDD Quick Start Guide
- Example prompts and integration patterns
Learn more: TDD Workflow Quickstart Guide
This release enables AI agents to autonomously execute test-driven development workflows with full state management and recovery capabilities.
Patch Changes
-
#1323
dc6652cThanks @Crunchyman-ralph! - Fix MCP server compatibility with Draft-07 clients (Augment IDE, gemini-cli, gemini code assist)- Resolves #1284
Problem:
- MCP tools were using Zod v4, which outputs JSON Schema Draft 2020-12
- MCP clients only support Draft-07
- Tools were not discoverable in gemini-cli and other clients
Solution:
- Updated all MCP tools to import from
zod/v3instead ofzod - Zod v3 schemas convert to Draft-07 via FastMCP's zod-to-json-schema
- Fixed logger to use stderr instead of stdout (MCP protocol requirement)
This is a temporary workaround until FastMCP adds JSON Schema version configuration.
task-master-ai@0.30.0-rc.0
bed63fa
Crunchyman-ralph
Ralph Khreish
Minor Changes
-
#1181
a69d8c9Thanks @karol-f! - Add configurable MCP tool loading to optimize LLM context usageYou can now control which Task Master MCP tools are loaded by setting the
TASK_MASTER_TOOLSenvironment variable in your MCP configuration. This helps reduce context usage for LLMs by only loading the tools you need.Configuration Options:
all(default): Load all 36 toolscoreorlean: Load only 7 essential tools for daily development- Includes:
get_tasks,next_task,get_task,set_task_status,update_subtask,parse_prd,expand_task
- Includes:
standard: Load 15 commonly used tools (all core tools plus 8 more)- Additional tools:
initialize_project,analyze_project_complexity,expand_all,add_subtask,remove_task,generate,add_task,complexity_report
- Additional tools:
- Custom list: Comma-separated tool names (e.g.,
get_tasks,next_task,set_task_status)
Example .mcp.json configuration:
{ "mcpServers": { "task-master-ai": { "command": "npx", "args": ["-y", "task-master-ai"], "env": { "TASK_MASTER_TOOLS": "standard", "ANTHROPIC_API_KEY": "your_key_here" } } } }For complete details on all available tools, configuration examples, and usage guidelines, see the MCP Tools documentation.
-
#1312
d7fca18Thanks @Crunchyman-ralph! - Improve next command to work with remote
Patch Changes
task-master-ai@0.29.0
Minor Changes
-
#1286
f12a16dThanks @Crunchyman-ralph! - Add changelog highlights to auto-update notificationsWhen the CLI auto-updates to a new version, it now displays a "What's New" section.
-
#1293
3010b90Thanks @Crunchyman-ralph! - Add Claude Code plugin with marketplace distributionThis release introduces official Claude Code plugin support, marking the evolution from legacy
.claudedirectory copying to a modern plugin-based architecture.🎉 New: Claude Code Plugin
Task Master AI commands and agents are now distributed as a proper Claude Code plugin:
- 49 slash commands with clean naming (
/taskmaster:command-name) - 3 specialized AI agents (task-orchestrator, task-executor, task-checker)
- MCP server integration for deep Claude Code integration
Installation:
/plugin marketplace add eyaltoledano/claude-task-master /plugin install taskmaster@taskmaster
The
rules add claudecommand no longer copies commands and agents to.claude/commands/and.claude/agents/. Instead, it now- Shows plugin installation instructions
- Only manages CLAUDE.md imports for agent instructions
- Directs users to install the official plugin
Migration for Existing Users:
If you previously used
rules add claude:- The old commands in
.claude/commands/will continue to work but won't receive updates - Install the plugin for the latest features:
/plugin install taskmaster@taskmaster - remove old
.claude/commands/and.claude/agents/directories
Why This Change?
Claude Code plugins provide:
- ✅ Automatic updates when we release new features
- ✅ Better command organization and naming
- ✅ Seamless integration with Claude Code
- ✅ No manual file copying or management
The plugin system is the future of Task Master AI integration with Claude Code!
- 49 slash commands with clean naming (
-
#1285
2a910a4Thanks @Crunchyman-ralph! - Add RPG (Repository Planning Graph) method template for structured PRD creation. The newexample_prd_rpg.txttemplate teaches AI agents and developers the RPG methodology through embedded instructions, inline good/bad examples, and XML-style tags for structure. This template enables creation of dependency-aware PRDs that automatically generate topologically-ordered task graphs when parsed with Task Master.Key features:
- Method-as-template: teaches RPG principles (dual-semantics, explicit dependencies, topological order) while being used
- Inline instructions at decision points guide AI through each section
- Good/bad examples for immediate pattern matching
- Flexible plain-text format with XML-style tags for parseability
- Critical dependency-graph section ensures correct task ordering
- Automatic inclusion during
task-master init - Comprehensive documentation at docs.task-master.dev/capabilities/rpg-method
- Tool recommendations for code-context-aware PRD creation (Claude Code, Cursor, Gemini CLI, Codex/Grok)
The RPG template complements the existing
example_prd.txtand provides a more structured approach for complex projects requiring clear module boundaries and dependency chains. -
#1287
90e6bdcThanks @Crunchyman-ralph! - Enhanceexpand_allto intelligently use complexity analysis recommendations when expanding tasks.The expand-all operation now automatically leverages recommendations from
analyze-complexityto determine optimal subtask counts for each task, resulting in more accurate and context-aware task breakdowns.Key improvements:
- Automatic integration with complexity analysis reports
- Tag-aware complexity report path resolution
- Intelligent subtask count determination based on task complexity
- Falls back to defaults when complexity analysis is unavailable
- Enhanced logging for better visibility into expansion decisions
When you run
task-master expand --allaftertask-master analyze-complexity, Task Master now uses the recommended subtask counts from the complexity analysis instead of applying uniform defaults, ensuring each task is broken down according to its actual complexity.
Patch Changes
-
#1191
aaf903fThanks @Crunchyman-ralph! - Fix cross-level task dependencies not being savedFixes an issue where adding dependencies between subtasks and top-level tasks (e.g.,
task-master add-dependency --id=2.2 --depends-on=11) would report success but fail to persist the changes. Dependencies can now be created in both directions between any task levels. -
#1299
4c1ef2cThanks @Crunchyman-ralph! - Improve refresh token when authenticating
task-master-ai@0.29.0-rc.1
e2e3e6f
Crunchyman-ralph
Ralph Khreish
Patch Changes
- #1299
a6c5152Thanks @Crunchyman-ralph! - Improve refresh token when authenticating
task-master-ai@0.29.0-rc.0
Minor Changes
-
#1286
f12a16dThanks @Crunchyman-ralph! - Add changelog highlights to auto-update notificationsWhen the CLI auto-updates to a new version, it now displays a "What's New" section.
-
#1293
3010b90Thanks @Crunchyman-ralph! - Add Claude Code plugin with marketplace distributionThis release introduces official Claude Code plugin support, marking the evolution from legacy
.claudedirectory copying to a modern plugin-based architecture.🎉 New: Claude Code Plugin
Task Master AI commands and agents are now distributed as a proper Claude Code plugin:
- 49 slash commands with clean naming (
/task-master-ai:command-name) - 3 specialized AI agents (task-orchestrator, task-executor, task-checker)
- MCP server integration for deep Claude Code integration
Installation:
/plugin marketplace add eyaltoledano/claude-task-master /plugin install taskmaster@taskmaster
The
rules add claudecommand no longer copies commands and agents to.claude/commands/and.claude/agents/. Instead, it now- Shows plugin installation instructions
- Only manages CLAUDE.md imports for agent instructions
- Directs users to install the official plugin
Migration for Existing Users:
If you previously used
rules add claude:- The old commands in
.claude/commands/will continue to work but won't receive updates - Install the plugin for the latest features:
/plugin install taskmaster@taskmaster - remove old
.claude/commands/and.claude/agents/directories
Why This Change?
Claude Code plugins provide:
- ✅ Automatic updates when we release new features
- ✅ Better command organization and naming
- ✅ Seamless integration with Claude Code
- ✅ No manual file copying or management
The plugin system is the future of Task Master AI integration with Claude Code!
- 49 slash commands with clean naming (
-
#1285
2a910a4Thanks @Crunchyman-ralph! - Add RPG (Repository Planning Graph) method template for structured PRD creation. The newexample_prd_rpg.txttemplate teaches AI agents and developers the RPG methodology through embedded instructions, inline good/bad examples, and XML-style tags for structure. This template enables creation of dependency-aware PRDs that automatically generate topologically-ordered task graphs when parsed with Task Master.Key features:
- Method-as-template: teaches RPG principles (dual-semantics, explicit dependencies, topological order) while being used
- Inline instructions at decision points guide AI through each section
- Good/bad examples for immediate pattern matching
- Flexible plain-text format with XML-style tags for parseability
- Critical dependency-graph section ensures correct task ordering
- Automatic inclusion during
task-master init - Comprehensive documentation at docs.task-master.dev/capabilities/rpg-method
- Tool recommendations for code-context-aware PRD creation (Claude Code, Cursor, Gemini CLI, Codex/Grok)
The RPG template complements the existing
example_prd.txtand provides a more structured approach for complex projects requiring clear module boundaries and dependency chains. -
#1287
90e6bdcThanks @Crunchyman-ralph! - Enhanceexpand_allto intelligently use complexity analysis recommendations when expanding tasks.The expand-all operation now automatically leverages recommendations from
analyze-complexityto determine optimal subtask counts for each task, resulting in more accurate and context-aware task breakdowns.Key improvements:
- Automatic integration with complexity analysis reports
- Tag-aware complexity report path resolution
- Intelligent subtask count determination based on task complexity
- Falls back to defaults when complexity analysis is unavailable
- Enhanced logging for better visibility into expansion decisions
When you run
task-master expand --allaftertask-master analyze-complexity, Task Master now uses the recommended subtask counts from the complexity analysis instead of applying uniform defaults, ensuring each task is broken down according to its actual complexity.
Patch Changes
-
#1191
aaf903fThanks @Crunchyman-ralph! - Fix cross-level task dependencies not being savedFixes an issue where adding dependencies between subtasks and top-level tasks (e.g.,
task-master add-dependency --id=2.2 --depends-on=11) would report success but fail to persist the changes. Dependencies can now be created in both directions between any task levels.
task-master-ai@0.28.0
Minor Changes
-
#1273
b43b7ceThanks @ben-vargas! - Add Codex CLI provider with OAuth authentication- Added codex-cli provider for GPT-5 and GPT-5-Codex models (272K input / 128K output)
- OAuth-first authentication via
codex login- no API key required - Optional OPENAI_CODEX_API_KEY support
- Codebase analysis capabilities automatically enabled
- Command-specific settings and approval/sandbox modes
-
#1215
0079b7dThanks @joedanz! - Add Cursor IDE custom slash command supportExpose Task Master commands as Cursor slash commands by copying assets/claude/commands to .cursor/commands on profile add and cleaning up on remove.
-
#1246
18aa416Thanks @Crunchyman-ralph! - Added api keys page on docs website: docs.task-master.dev/getting-started/api-keys -
#1246
18aa416Thanks @Crunchyman-ralph! - Move to AI SDK v5: -
#1262
738ec51Thanks @Crunchyman-ralph! - Migrate AI services to use generateObject for structured data generationThis update migrates all AI service calls from generateText to generateObject, ensuring more reliable and structured responses across all commands.
Key Changes:
- Unified AI Service: Replaced separate generateText implementations with a single generateObjectService that handles structured data generation
- JSON Mode Support: Added proper JSON mode configuration for providers that support it (OpenAI, Anthropic, Google, Groq)
- Schema Validation: Integrated Zod schemas for all AI-generated content with automatic validation
- Provider Compatibility: Maintained compatibility with all existing providers while leveraging their native structured output capabilities
- Improved Reliability: Structured output generation reduces parsing errors and ensures consistent data formats
Technical Improvements:
- Centralized provider configuration in
ai-providers-unified.js - Added
generateObjectsupport detection for each provider - Implemented proper error handling for schema validation failures
- Maintained backward compatibility with existing prompt structures
Bug Fixes:
- Fixed subtask ID numbering issue where AI was generating inconsistent IDs (101-105, 601-603) instead of sequential numbering (1, 2, 3...)
- Enhanced prompt instructions to enforce proper ID generation patterns
- Ensured subtasks display correctly as X.1, X.2, X.3 format
This migration improves the reliability and consistency of AI-generated content throughout the Task Master application.
-
#1112
d67b81dThanks @olssonsten! - Enhanced Roo Code profile with MCP timeout configuration for improved reliability during long-running AI operations. The Roo profile now automatically configures a 300-second timeout for MCP server operations, preventing timeouts during complex tasks likeparse-prd,expand-all,analyze-complexity, andresearchoperations. This change also replaces static MCP configuration files with programmatic generation for better maintainability.What's New:
- 300-second timeout for MCP operations (up from default 60 seconds)
- Programmatic MCP configuration generation (replaces static asset files)
- Enhanced reliability for AI-powered operations
- Consistent with other AI coding assistant profiles
Migration: No user action required - existing Roo Code installations will automatically receive the enhanced MCP configuration on next initialization.
-
#1246
986ac11Thanks @Crunchyman-ralph! - Upgrade grok-cli ai provider to ai sdk v5
Patch Changes
-
#1235
aaacc3dThanks @Crunchyman-ralph! - Improveanalyze-complexitycli docs and--researchflag documentation -
#1251
0b2c696Thanks @Crunchyman-ralph! - Change parent task back to "pending" when all subtasks are in "pending" state -
#1274
4f984f8Thanks @Crunchyman-ralph! - Do a quick fix on build -
#1277
7b5a7c4Thanks @Crunchyman-ralph! - Fix MCP connection errors caused by deprecated generateTaskFiles calls. Resolves "Cannot read properties of null (reading 'toString')" errors when using MCP tools for task management operations. -
#1276
caee040Thanks @Crunchyman-ralph! - Fix MCP server error when file parameter not provided - now properly constructs default tasks.json path instead of failing with 'tasksJsonPath is required' error. -
#1172
b5fe723Thanks @jujax! - Fix Claude Code settings validation for pathToClaudeCodeExecutable -
#1192
2b69936Thanks @nukunga! - Fix sonar deep research model failing, should be calledsonar-deep-research -
#1270
20004a3Thanks @Crunchyman-ralph! - Fix complexity score not showing fortask-master showandtask-master list- Added complexity score on "next task" when running
task-master list - Added colors to complexity to reflect complexity (easy, medium, hard)
- Added complexity score on "next task" when running
task-master-ai@0.28.0-rc.2
Minor Changes
- #1273
b43b7ceThanks @ben-vargas! - Add Codex CLI provider with OAuth authentication- Added codex-cli provider for GPT-5 and GPT-5-Codex models (272K input / 128K output)
- OAuth-first authentication via
codex login- no API key required - Optional OPENAI_CODEX_API_KEY support
- Codebase analysis capabilities automatically enabled
- Command-specific settings and approval/sandbox modes
Patch Changes
-
#1277
7b5a7c4Thanks @Crunchyman-ralph! - Fix MCP connection errors caused by deprecated generateTaskFiles calls. Resolves "Cannot read properties of null (reading 'toString')" errors when using MCP tools for task management operations. -
#1276
caee040Thanks @Crunchyman-ralph! - Fix MCP server error when file parameter not provided - now properly constructs default tasks.json path instead of failing with 'tasksJsonPath is required' error.
task-master-ai@0.28.0-rc.1
Patch Changes
- #1274
4f984f8Thanks @Crunchyman-ralph! - Do a quick fix on build