eyaltoledano/claude-task-master v0.13.0

on GitHub

Minor Changes

• #240 ef782ff Thanks @eyaltoledano! - feat(expand): Enhance expand and expand-all commands

  • Integrate task-complexity-report.json to automatically determine the number of subtasks and use tailored prompts for expansion based on prior analysis. You no longer need to try copy-pasting the recommended prompt. If it exists, it will use it for you. You can just run task-master update --id=[id of task] --research and it will use that prompt automatically. No extra prompt needed.
  • Change default behavior to append new subtasks to existing ones. Use the --force flag to clear existing subtasks before expanding. This is helpful if you need to add more subtasks to a task but you want to do it by the batch from a given prompt. Use force if you want to start fresh with a task's subtasks.

• #240 87d97bb Thanks @eyaltoledano! - Adds support for the OpenRouter AI provider. Users can now configure models available through OpenRouter (requiring an OPENROUTER_API_KEY) via the task-master models command, granting access to a wide range of additional LLMs. - IMPORTANT FYI ABOUT OPENROUTER: Taskmaster relies on AI SDK, which itself relies on tool use. It looks like free models sometimes do not include tool use. For example, Gemini 2.5 pro (free) failed via OpenRouter (no tool use) but worked fine on the paid version of the model. Custom model support for Open Router is considered experimental and likely will not be further improved for some time.

• #240 1ab836f Thanks @eyaltoledano! - Adds model management and new configuration file .taskmasterconfig which houses the models used for main, research and fallback. Adds models command and setter flags. Adds a --setup flag with an interactive setup. We should be calling this during init. Shows a table of active and available models when models is called without flags. Includes SWE scores and token costs, which are manually entered into the supported_models.json, the new place where models are defined for support. Config-manager.js is the core module responsible for managing the new config."

• #240 c8722b0 Thanks @eyaltoledano! - Adds custom model ID support for Ollama and OpenRouter providers.

  • Adds the --ollama and --openrouter flags to task-master models --set-<role> command to set models for those providers outside of the support models list.
  • Updated task-master models --setup interactive mode with options to explicitly enter custom Ollama or OpenRouter model IDs.
  • Implemented live validation against OpenRouter API (/api/v1/models) when setting a custom OpenRouter model ID (via flag or setup).
  • Refined logic to prioritize explicit provider flags/choices over internal model list lookups in case of ID conflicts.
  • Added warnings when setting custom/unvalidated models.
  • We obviously don't recommend going with a custom, unproven model. If you do and find performance is good, please let us know so we can add it to the list of supported models.

• #240 2517bc1 Thanks @eyaltoledano! - Integrate OpenAI as a new AI provider. - Enhance models command/tool to display API key status. - Implement model-specific maxTokens override based on supported-models.json to save you if you use an incorrect max token value.

• #240 9a48278 Thanks @eyaltoledano! - Tweaks Perplexity AI calls for research mode to max out input tokens and get day-fresh information - Forces temp at 0.1 for highly deterministic output, no variations - Adds a system prompt to further improve the output - Correctly uses the maximum input tokens (8,719, used 8,700) for perplexity - Specificies to use a high degree of research across the web - Specifies to use information that is as fresh as today; this support stuff like capturing brand new announcements like new GPT models and being able to query for those in research. 🔥

Patch Changes

• #240 842eaf7 Thanks @eyaltoledano! - - Add support for Google Gemini models via Vercel AI SDK integration.

• #240 ed79d4f Thanks @eyaltoledano! - Add xAI provider and Grok models support

• #378 ad89253 Thanks @eyaltoledano! - Better support for file paths on Windows, Linux & WSL.

  • Standardizes handling of different path formats (URI encoded, Windows, Linux, WSL).
  • Ensures tools receive a clean, absolute path suitable for the server OS.
  • Simplifies tool implementation by centralizing normalization logic.

• #285 2acba94 Thanks @neno-is-ooo! - Add integration for Roo Code

• #378 d63964a Thanks @eyaltoledano! - Improved update-subtask - Now it has context about the parent task details - It also has context about the subtask before it and the subtask after it (if they exist) - Not passing all subtasks to stay token efficient

• #240 5f504fa Thanks @eyaltoledano! - Improve and adjust init command for robustness and updated dependencies.

  • Update Initialization Dependencies: Ensure newly initialized projects (task-master init) include all required AI SDK dependencies (@ai-sdk/*, ai, provider wrappers) in their package.json for out-of-the-box AI feature compatibility. Remove unnecessary dependencies (e.g., uuid) from the init template.
  • Silence npm install during init: Prevent npm install output from interfering with non-interactive/MCP initialization by suppressing its stdio in silent mode.
  • Improve Conditional Model Setup: Reliably skip interactive models --setup during non-interactive init runs (e.g., init -y or MCP) by checking isSilentMode() instead of passing flags.
  • Refactor init.js: Remove internal isInteractive flag logic.
  • Update init Instructions: Tweak the "Getting Started" text displayed after init.
  • Fix MCP Server Launch: Update .cursor/mcp.json template to use node ./mcp-server/server.js instead of npx task-master-mcp.
  • Update Default Model: Change the default main model in the .taskmasterconfig template.

• #240 96aeeff Thanks @eyaltoledano! - Fixes an issue with add-task which did not use the manually defined properties and still needlessly hit the AI endpoint.

• #240 5aea93d Thanks @eyaltoledano! - Fixes an issue that prevented remove-subtask with comma separated tasks/subtasks from being deleted (only the first ID was being deleted). Closes #140

• #240 66ac9ab Thanks @eyaltoledano! - Improves next command to be subtask-aware - The logic for determining the "next task" (findNextTask function, used by task-master next and the next_task MCP tool) has been significantly improved. Previously, it only considered top-level tasks, making its recommendation less useful when a parent task containing subtasks was already marked 'in-progress'. - The updated logic now prioritizes finding the next available subtask within any 'in-progress' parent task, considering subtask dependencies and priority. - If no suitable subtask is found within active parent tasks, it falls back to recommending the next eligible top-level task based on the original criteria (status, dependencies, priority).

  This change makes the next command much more relevant and helpful during the implementation phase of complex tasks.

• #240 ca7b045 Thanks @eyaltoledano! - Add --status flag to show command to filter displayed subtasks.

• #328 5a2371b Thanks @knoxgraeme! - Fix --task to --num-tasks in ui + related tests - issue #324

• #240 6cb213e Thanks @eyaltoledano! - Adds a 'models' CLI and MCP command to get the current model configuration, available models, and gives the ability to set main/research/fallback models." - In the CLI, task-master models shows the current models config. Using the --setup flag launches an interactive set up that allows you to easily select the models you want to use for each of the three roles. Use q during the interactive setup to cancel the setup. - In the MCP, responses are simplified in RESTful format (instead of the full CLI output). The agent can use the models tool with different arguments, including listAvailableModels to get available models. Run without arguments, it will return the current configuration. Arguments are available to set the model for each of the three roles. This allows you to manage Taskmaster AI providers and models directly from either the CLI or MCP or both. - Updated the CLI help menu when you run task-master to include missing commands and .taskmasterconfig information. - Adds --research flag to add-task so you can hit up Perplexity right from the add-task flow, rather than having to add a task and then update it.