Adds a new CLI command and MCP tool to reorganize tasks and subtasks within the hierarchy. Features include:
- Moving tasks between different positions in the task list
- Converting tasks to subtasks and vice versa
- Moving subtasks between parents
- Moving multiple tasks at once with comma-separated IDs
- Creating placeholder tasks when moving to new IDs
- Validation to prevent accidental data loss
This is particularly useful for resolving merge conflicts when multiple team members create tasks on different branches.
Enhance analyze-complexity to support analyzing specific tasks by ID or range:
- Add --id option for comma-separated task IDs
- Add --from/--to options for analyzing tasks within a range
- Implement intelligent merging with existing reports
- Update CLI, MCP tools, and direct functions for consistent support
- Add changeset documenting the feature
This commit implements automatic tasks.json file creation when it doesn't exist:
- When tasks.json is missing or invalid, create a new one with { tasks: [] }
- Allows adding tasks immediately after initializing a project without parsing a PRD
- Replaces error with informative feedback about file creation
- Enables smoother workflow for new projects or directories
This change improves user experience by removing the requirement to parse a PRD
before adding the first task to a newly initialized project. Closes#494
This commit significantly improves the functionality by implementing
fuzzy semantic search to find contextually relevant dependencies:
- Add Fuse.js for powerful fuzzy search capability with weighted multi-field matching
- Implement score-based relevance ranking with high/medium relevance tiers
- Enhance context generation to include detailed information about similar tasks
- Fix context shadowing issue that prevented detailed task information from
reaching the AI model
- Add informative CLI output showing semantic search results and dependency patterns
- Improve formatting of dependency information in prompts with task titles
The result is that newly created tasks are automatically placed within the correct
dependency structure without manual intervention, with the AI having much better
context about which tasks are most relevant to the new one being created.
This significantly improves the user experience by reducing the need to manually
update task dependencies after creation, all without increasing token usage or costs.
This commit introduces several improvements to AI interactions and
task management functionalities:
- AI Provider Enhancements (for Telemetry & Robustness):
- :
- Added a check in to ensure
is a string, throwing an error if not. This prevents downstream
errors (e.g., in ).
- , , :
- Standardized return structures for their respective
and functions to consistently include /
and fields. This aligns them with other providers (like
Anthropic, Google, Perplexity) for consistent telemetry data
collection, as part of implementing subtask 77.14 and similar work.
- Task Expansion ():
- Updated to be more explicit
about using an empty array for empty to
better guide AI output.
- Implemented a pre-emptive cleanup step in
to replace malformed with
before JSON parsing. This improves resilience to AI output quirks,
particularly observed with Perplexity.
- Adjusts issue in commands.js where successfulRemovals would be undefined. It's properly invoked from the result variable now.
- Updates supported models for Gemini
These changes address issues observed during E2E tests, enhance the
reliability of AI-driven task analysis and expansion, and promote
consistent telemetry data across multiple AI providers.
This commit updates to more robustly handle responses from .
Previously, the module strictly expected the AI-generated object to be nested under . This change ensures that it now first checks if itself contains the expected task data object, and then falls back to checking .
This enhancement increases compatibility with varying AI provider response structures, similar to the improvements recently made in .
This commit introduces two key improvements:
1. **Google Provider Telemetry:**
- Updated to include token usage data (, ) in the responses from and .
- This aligns the Google provider with others for consistent AI usage telemetry.
2. **Robust AI Object Response Handling:**
- Modified to more flexibly handle responses from .
- The add-task module now check for the AI-generated object in both and , improving compatibility with different AI provider response structures (e.g., Gemini).
These changes enhance the reliability of AI interactions, particularly with the Google provider, and ensure accurate telemetry collection.
This commit applies the standard telemetry pattern to the analyze-task-complexity command and its corresponding MCP tool.
Key Changes:
1. Core Logic (scripts/modules/task-manager/analyze-task-complexity.js):
- The call to generateTextService now includes commandName: 'analyze-complexity' and outputType.
- The full response { mainResult, telemetryData } is captured.
- mainResult (the AI-generated text) is used for parsing the complexity report JSON.
- If running in CLI mode (outputFormat === 'text'), displayAiUsageSummary is called with the telemetryData.
- The function now returns { report: ..., telemetryData: ... }.
2. Direct Function (mcp-server/src/core/direct-functions/analyze-task-complexity.js):
- The call to the core analyzeTaskComplexity function now passes the necessary context for telemetry (commandName, outputType).
- The successful response object now correctly extracts coreResult.telemetryData and includes it in the data.telemetryData field returned to the MCP client.
This commit applies the standard telemetry pattern to the update-subtask command and its corresponding MCP tool.
Key Changes:
1. Core Logic (scripts/modules/task-manager/update-subtask-by-id.js):
- The call to generateTextService now includes commandName: 'update-subtask' and outputType.
- The full response { mainResult, telemetryData } is captured.
- mainResult (the AI-generated text) is used for the appended content.
- If running in CLI mode (outputFormat === 'text'), displayAiUsageSummary is called with the telemetryData.
- The function now returns { updatedSubtask: ..., telemetryData: ... }.
2. Direct Function (mcp-server/src/core/direct-functions/update-subtask-by-id.js):
- The call to the core updateSubtaskById function now passes the necessary context for telemetry (commandName, outputType).
- The successful response object now correctly extracts coreResult.telemetryData and includes it in the data.telemetryData field returned to the MCP client.
This commit applies the standard telemetry pattern to the update-tasks command and its corresponding MCP tool.
Key Changes:
1. Core Logic (scripts/modules/task-manager/update-tasks.js):
- The call to generateTextService now includes commandName: 'update-tasks' and outputType.
- The full response { mainResult, telemetryData } is captured.
- mainResult (the AI-generated text) is used for parsing the updated task JSON.
- If running in CLI mode (outputFormat === 'text'), displayAiUsageSummary is called with the telemetryData.
- The function now returns { success: true, updatedTasks: ..., telemetryData: ... }.
2. Direct Function (mcp-server/src/core/direct-functions/update-tasks.js):
- The call to the core updateTasks function now passes the necessary context for telemetry (commandName, outputType).
- The successful response object now correctly extracts coreResult.telemetryData and includes it in the data.telemetryData field returned to the MCP client.
This commit applies the standard telemetry pattern to the command and its corresponding MCP tool.
Key Changes:
1. **Core Logic ():**
- The call to now includes and .
- The full response is captured.
- (the AI-generated text) is used for parsing the updated task JSON.
- If running in CLI mode (), is called with the .
- The function now returns .
2. **Direct Function ():**
- The call to the core function now passes the necessary context for telemetry (, ).
- The successful response object now correctly extracts and includes it in the field returned to the MCP client.
This commit implements AI usage telemetry for the `expand-all-tasks` command/tool and refactors its CLI output for clarity and consistency.
Key Changes:
1. **Telemetry Integration for `expand-all-tasks` (Subtask 77.8):**\n - The `expandAllTasks` core logic (`scripts/modules/task-manager/expand-all-tasks.js`) now calls the `expandTask` function for each eligible task and collects the individual `telemetryData` returned.\n - A new helper function `_aggregateTelemetry` (in `utils.js`) is used to sum up token counts and costs from all individual expansions into a single `telemetryData` object for the entire `expand-all` operation.\n - The `expandAllTasksDirect` wrapper (`mcp-server/src/core/direct-functions/expand-all-tasks.js`) now receives and passes this aggregated `telemetryData` in the MCP response.\n - For CLI usage, `displayAiUsageSummary` is called once with the aggregated telemetry.
2. **Improved CLI Output for `expand-all`:**\n - The `expandAllTasks` core function now handles displaying a final "Expansion Summary" box (showing Attempted, Expanded, Skipped, Failed counts) directly after the aggregated telemetry summary.\n - This consolidates all summary output within the core function for better flow and removes redundant logging from the command action in `scripts/modules/commands.js`.\n - The summary box border is green for success and red if any expansions failed.
3. **Code Refinements:**\n - Ensured `chalk` and `boxen` are imported in `expand-all-tasks.js` for the new summary box.\n - Minor adjustments to logging messages for clarity.
This commit integrates AI usage telemetry for the `expand-task` command/tool and resolves issues related to incorrect return type handling and logging.
Key Changes:
1. **Telemetry Integration for `expand-task` (Subtask 77.7):**\n - Applied the standard telemetry pattern to the `expandTask` core logic (`scripts/modules/task-manager/expand-task.js`) and the `expandTaskDirect` wrapper (`mcp-server/src/core/direct-functions/expand-task.js`).\n - AI service calls now pass `commandName` and `outputType`.\n - Core function returns `{ task, telemetryData }`.\n - Direct function correctly extracts `task` and passes `telemetryData` in the MCP response `data` field.\n - Telemetry summary is now displayed in the CLI output for the `expand` command.
2. **Fix AI Service Return Type Handling (`ai-services-unified.js`):**\n - Corrected the `_unifiedServiceRunner` function to properly handle the return objects from provider-specific functions (`generateText`, `generateObject`).\n - It now correctly extracts `providerResponse.text` or `providerResponse.object` into the `mainResult` field based on `serviceType`, resolving the "text.trim is not a function" error encountered during `expand-task`.
3. **Log Cleanup:**\n - Removed various redundant or excessive `console.log` statements across multiple files (as indicated by recent changes) to reduce noise and improve clarity, particularly for MCP interactions.
Implements AI usage telemetry capture and propagation for the command and MCP tool, following the established telemetry pattern.
Key changes:
- **Core ():**
- Modified the call to include and .
- Updated to receive from .
- Adjusted to return an object .
- Added a call to to show telemetry data in the CLI output when not in MCP mode.
- **Direct Function ():**
- Updated the call to the core function to pass , , and .
- Modified to correctly handle the new return structure from the core function.
- Ensures received from the core function is included in the field of the successful MCP response.
- **MCP Tool ():**
- No changes required; existing correctly passes through the object containing .
- **CLI Command ():**
- The command's action now relies on the core function to handle CLI success messages and telemetry display.
This ensures that AI usage for the functionality is tracked and can be displayed or logged as appropriate for both CLI and MCP interactions.
This commit introduces a standardized pattern for capturing and propagating AI usage telemetry (cost, tokens, model used) across the Task Master stack and applies it to the 'add-task' functionality.
Key changes include:
- **Telemetry Pattern Definition:**
- Added defining the integration pattern for core logic, direct functions, MCP tools, and CLI commands.
- Updated related rules (, ,
Usage: mcp [OPTIONS] COMMAND [ARGS]...
MCP development tools
╭─ Options ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ --help Show this message and exit. │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─ Commands ─────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ version Show the MCP version. │
│ dev Run a MCP server with the MCP Inspector. │
│ run Run a MCP server. │
│ install Install a MCP server in the Claude desktop app. │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯, , ) to reference the new telemetry rule.
- **Core Telemetry Implementation ():**
- Refactored the unified AI service to generate and return a object alongside the main AI result.
- Fixed an MCP server startup crash by removing redundant local loading of and instead using the imported from for cost calculations.
- Added to the object.
- ** Integration:**
- Modified (core) to receive from the AI service, return it, and call the new UI display function for CLI output.
- Updated to receive from the core function and include it in the payload of its response.
- Ensured (MCP tool) correctly passes the through via .
- Updated to correctly pass context (, ) to the core function and rely on it for CLI telemetry display.
- **UI Enhancement:**
- Added function to to show telemetry details in the CLI.
- **Project Management:**
- Added subtasks 77.6 through 77.12 to track the rollout of this telemetry pattern to other AI-powered commands (, , , , , , ).
This establishes the foundation for tracking AI usage across the application.
- Enhance error validation in parse-prd.js and update-tasks.js
- Fix bug where mcpLog was incorrectly passed as logWrapper
- Improve error messages and response formatting
- Add --skip-verification flag to E2E tests
- Update MCP server config that ships with init to match new API key structure
- Fix task force/append handling in parse-prd command
- Increase column width in update-tasks display
This commit introduces several improvements and refactorings across MCP tools, core logic, and configuration.
**Major Changes:**
1. **Refactor updateSubtaskById:**
- Switched from generateTextService to generateObjectService for structured AI responses, using a Zod schema (subtaskSchema) for validation.
- Revised prompts to have the AI generate relevant content based on user request and context (parent/sibling tasks), while explicitly preventing AI from handling timestamp/tag formatting.
- Implemented **local timestamp generation (new Date().toISOString()) and formatting** (using <info added on ...> tags) within the function *after* receiving the AI response. This ensures reliable and correctly formatted details are appended.
- Corrected logic to append only the locally formatted, AI-generated content block to the existing subtask.details.
2. **Consolidate MCP Utilities:**
- Moved/consolidated the withNormalizedProjectRoot HOF into mcp-server/src/tools/utils.js.
- Updated MCP tools (like update-subtask.js) to import withNormalizedProjectRoot from the new location.
3. **Refactor Project Initialization:**
- Deleted the redundant mcp-server/src/core/direct-functions/initialize-project-direct.js file.
- Updated mcp-server/src/core/task-master-core.js to import initializeProjectDirect from its correct location (./direct-functions/initialize-project.js).
**Other Changes:**
- Updated .taskmasterconfig fallback model to claude-3-7-sonnet-20250219.
- Clarified model cost representation in the models tool description (taskmaster.mdc and mcp-server/src/tools/models.js).
Refactors the JSON array parsing logic within
in .
The previous logic primarily relied on extracting content from markdown
code blocks (json or javascript), which proved brittle when the AI
response included comments or non-JSON text within the block, leading to
parsing errors for the command.
This change modifies the parsing strategy to first attempt extracting
content directly between the outermost '[' and ']' brackets. This is
more robust as it targets the expected array structure directly. If
bracket extraction fails, it falls back to looking for a strict json
code block, then prefix stripping, before attempting a raw parse.
This approach aligns with the successful parsing strategy used for
single-object responses in and resolves the
parsing errors previously observed with the command.
Problem: expand_task & expand_all MCP tools failed with .env keys due to missing projectRoot propagation for API key resolution. Also fixed a ReferenceError: wasSilent is not defined in expandTaskDirect.
Solution: Modified core logic, direct functions, and MCP tools for expand-task and expand-all to correctly destructure projectRoot from arguments and pass it down through the context object to the AI service call (generateTextService). Fixed wasSilent scope in expandTaskDirect.
Verification: Tested expand_task successfully in MCP using .env keys. Reviewed expand_all flow for correct projectRoot propagation.
Modified update-subtask-by-id core, direct function, and tool to pass projectRoot for .env API key fallback. Removed check preventing appending details to completed subtasks.
Modified update-task-by-id core, direct function, and tool to pass projectRoot. Reverted parsing logic in core function to prioritize `{...}` extraction, resolving parsing errors. Fixed ReferenceError by correctly destructuring projectRoot.
Modified parse-prd core, direct function, and tool to pass projectRoot for .env API key fallback. Corrected Zod schema used in generateObjectService call. Fixed logFn reference error in core parsePRD. Updated unit test mock for utils.js.
Modified add-task core, direct function, and tool to pass projectRoot for .env API key fallback. Fixed logFn reference error and removed deprecated reportProgress call in core addTask function. Verified working.
Modified analyze-task-complexity.js core function, direct function, and analyze.js tool to correctly pass projectRoot. Fixed import error in tools/index.js. Added debug logging to _resolveApiKey in ai-services-unified.js. This enables the .env API key fallback for analyze_project_complexity.
Modified ai-services-unified.js, update.js tool, and update-tasks.js direct function to correctly pass projectRoot. This enables the .env file API key fallback mechanism for the update command when running via MCP, ensuring consistent key resolution with the CLI context.
Problem:
- Task Master model configuration wasn't properly checking for API keys in the project's .env file when running through MCP
- The isApiKeySet function was only checking session.env and process.env but not inspecting the .env file directly
- This caused incorrect API key status reporting in MCP tools even when keys were properly set in .env
Solution:
- Modified resolveEnvVariable function in utils.js to properly read from .env file at projectRoot
- Updated isApiKeySet to correctly pass projectRoot to resolveEnvVariable
- Enhanced the key detection logic to have consistent behavior between CLI and MCP contexts
- Maintains the correct precedence: session.env → .env file → process.env
Testing:
- Verified working correctly with both MCP and CLI tools
- API keys properly detected in .env file in both contexts
- Deleted .cursor/mcp.json to confirm introspection of .env as fallback works
- Enhance E2E testing and LLM analysis report and:
- Add --analyze-log flag to run_e2e.sh to re-run LLM analysis on existing logs.
- Add test:e2e and analyze-log scripts to package.json for easier execution.
- Correct display errors and dependency validation output:
- Update chalk usage in add-task.js to use bracket notation (chalk[color]) compatible with v5, resolving 'chalk.keyword is not a function' error.
- Modify fix-dependencies command output to show red failure box with issue count instead of green success box when validation fails.
- Refactor interactive model setup:
- Verify inclusion of 'No change' option during interactive model setup flow (task-master models --setup).
- Update model definitions:
- Add max_tokens field for gpt-4o in supported-models.json.
- Remove unused scripts:
- Delete prepare-package.js and rule-transformer.test.js.
Release candidate
- Refactors the core `removeTask` function (`task-manager/remove-task.js`) to accept and iterate over comma-separated task/subtask IDs.
- Updates dependency cleanup and file regeneration logic to run once after processing all specified IDs.
- Adjusts the `remove-task` CLI command (`commands.js`) description and confirmation prompt to handle multiple IDs correctly.
- Fixes a bug in the CLI confirmation prompt where task/subtask titles were not being displayed correctly.
- Updates the `remove_task` MCP tool description to reflect the new multi-ID capability.
This addresses the previously known issue where only the first ID in a comma-separated list was processed.
Closes#140
Adds the ability for users to specify custom model IDs for Ollama and OpenRouter providers, bypassing the internal supported model list.
- Introduces --ollama and --openrouter flags for the 'task-master models --set-<role>' command.
- Updates the interactive 'task-master models --setup' to include options for entering custom Ollama/OpenRouter IDs.
- Implements live validation against the OpenRouter API when a custom OpenRouter ID is provided.
- Refines the model setting logic to prioritize explicit provider flags/choices.
- Adds warnings when custom models are set.
- Updates the changeset file.
- Add OpenAI provider implementation using @ai-sdk/openai.\n- Update `models` command/tool to display API key status for configured providers.\n- Implement model-specific `maxTokens` override logic in `config-manager.js` using `supported-models.json`.\n- Improve AI error message parsing in `ai-services-unified.js` for better clarity.
The add-task command handler in commands.js was incorrectly passing null for the manualTaskData parameter to the core addTask function. This caused the core function to always fall back to the AI generation path, even when only manual flags like --title and --description were provided. This commit updates the call to pass the correctly constructed manualTaskData object, ensuring that manual task creation via the CLI works as intended without unnecessarily calling the AI service.
This commit completes the major refactoring initiative (Task 61) to migrate all AI-interacting task management functions to the unified service layer (`ai-services-unified.js`) and standardized configuration (`config-manager.js`).
Key Changes:
1. **Refactor `update-task-by-id` & `update-subtask-by-id`:**
* Replaced direct AI client logic and config fetching with calls to `generateTextService`.
* Preserved original prompt logic while ensuring JSON output format is requested.
* Implemented robust manual JSON parsing and Zod validation for text-based AI responses.
* Corrected logger implementation (`logFn`/`isMCP`/`report` pattern) for both CLI and MCP contexts.
* Ensured correct passing of `session` context to the unified service.
* Refactored associated direct function wrappers (`updateTaskByIdDirect`, `updateSubtaskByIdDirect`) to remove AI client initialization and call core logic appropriately.
2. **CLI Environment Loading:**
* Added `dotenv.config()` to `scripts/dev.js` to ensure consistent loading of the `.env` file for CLI operations.
3. **Obsolete Code Removal:**
* Deleted unused helper files:
* `scripts/modules/task-manager/get-subtasks-from-ai.js`
* `scripts/modules/task-manager/generate-subtask-prompt.js`
* `scripts/modules/ai-services.js`
* `scripts/modules/ai-client-factory.js`
* `mcp-server/src/core/utils/ai-client-utils.js`
* Removed corresponding imports/exports from `scripts/modules/task-manager.js` and `mcp-server/src/core/task-master-core.js`.
4. **Verification:**
* Successfully tested `update-task` and `update-subtask` via both CLI and MCP after refactoring.
5. **Task Management:**
* Marked subtasks 61.38, 61.39, 61.40, 61.41, and 61.33 as 'done'.
* Includes other task content/status updates as reflected in the diff.
This completes the migration of core AI features to the new architecture, enhancing maintainability and flexibility.
Completes the refactoring of the AI-interacting task management functions by aligning `update-tasks.js` with the unified service architecture and removing now-unused helper files.
Key Changes:
- **`update-tasks.js` Refactoring:**
- Replaced direct AI client calls and AI-specific config fetching with a call to `generateTextService` from `ai-services-unified.js`.
- Preserved the original system and user prompts requesting a JSON array output.
- Implemented manual JSON parsing (`parseUpdatedTasksFromText`) with Zod validation to handle the text response reliably.
- Updated the core function signature to accept the standard `context` object (`{ session, mcpLog }`).
- Corrected logger implementation to handle both MCP (`mcpLog`) and CLI (`consoleLog`) contexts appropriately.
- **Related Component Updates:**
- Refactored `mcp-server/src/core/direct-functions/update-tasks.js` to use the standard direct function pattern (logger wrapper, silent mode, call core function with context).
- Verified `mcp-server/src/tools/update.js` correctly passes arguments and context.
- Verified `scripts/modules/commands.js` (update command) correctly calls the refactored core function.
- **Obsolete File Cleanup:**
- Removed the now-unused `scripts/modules/task-manager/get-subtasks-from-ai.js` file and its export, as its functionality was integrated into `expand-task.js`.
- Removed the now-unused `scripts/modules/task-manager/generate-subtask-prompt.js` file and its export for the same reason.
- **Task Management:**
- Marked subtasks 61.38, 61.39, and 61.41 as complete.
This commit finalizes the alignment of `updateTasks`, `updateTaskById`, `expandTask`, `expandAllTasks`, `analyzeTaskComplexity`, `addTask`, and `parsePRD` with the unified AI service and configuration management patterns.
Refactors the `expandTask` and `expandAllTasks` features to complete subtask 61.38 and enhance functionality based on subtask 61.37's refactor.
Key Changes:
- **Additive Expansion (`expandTask`, `expandAllTasks`):**
- Modified `expandTask` default behavior to append newly generated subtasks to any existing ones.
- Added a `force` flag (passed down from CLI/MCP via `--force` option/parameter) to `expandTask` and `expandAllTasks`. When `force` is true, existing subtasks are cleared before generating new ones.
- Updated relevant CLI command (`expand`), MCP tool (`expand_task`, `expand_all`), and direct function wrappers (`expandTaskDirect`, `expandAllTasksDirect`) to handle and pass the `force` flag.
- **Complexity Report Integration (`expandTask`):**
- `expandTask` now reads `scripts/task-complexity-report.json`.
- If an analysis entry exists for the target task:
- `recommendedSubtasks` is used to determine the number of subtasks to generate (unless `--num` is explicitly provided).
- `expansionPrompt` is used as the primary prompt content for the AI.
- `reasoning` is appended to any additional context provided.
- If no report entry exists or the report is missing, it falls back to default subtask count (from config) and standard prompt generation.
- **`expandAllTasks` Orchestration:**
- Refactored `expandAllTasks` to primarily iterate through eligible tasks (pending/in-progress, considering `force` flag and existing subtasks) and call the updated `expandTask` function for each.
- Removed redundant logic (like complexity reading or explicit subtask clearing) now handled within `expandTask`.
- Ensures correct context (`session`, `mcpLog`) and flags (`useResearch`, `force`) are passed down.
- **Configuration & Cleanup:**
- Updated `.cursor/mcp.json` with new Perplexity/Anthropic API keys (old ones invalidated).
- Completed refactoring of `expandTask` started in 61.37, confirming usage of `generateTextService` and appropriate prompts.
- **Task Management:**
- Marked subtask 61.37 as complete.
- Updated `.changeset/cuddly-zebras-matter.md` to reflect user-facing changes.
These changes finalize the refactoring of the task expansion features, making them more robust, configurable via complexity analysis, and aligned with the unified AI service architecture.
Refactored the `expandTask` feature (`scripts/modules/task-manager/expand-task.js`) and related components (`commands.js`, `mcp-server/src/tools/expand-task.js`, `mcp-server/src/core/direct-functions/expand-task.js`) to integrate with the unified AI service layer (`ai-services-unified.js`) and configuration management (`config-manager.js`).
The refactor involved:
- Removing direct AI client calls and configuration fetching from `expand-task.js`.
- Attempting to use `generateObjectService` for structured subtask generation. This failed due to provider-specific errors (Perplexity internal errors, Anthropic schema translation issues).
- Reverting the core AI interaction to use `generateTextService`, asking the LLM to format its response as JSON containing a "subtasks" array.
- Re-implementing manual JSON parsing and Zod validation (`parseSubtasksFromText`) to handle the text response reliably.
- Updating prompt generation functions (`generateMainSystemPrompt`, `generateMainUserPrompt`, `generateResearchUserPrompt`) to request the correct JSON object structure within the text response.
- Ensuring the `expandTaskDirect` function handles pre-checks (force flag, task status) and correctly passes the `session` context and logger wrapper to the core `expandTask` function.
- Correcting duplicate imports in `commands.js`.
- Validating the refactored feature works correctly via both CLI (`task-master expand --id <id>`) and MCP (`expand_task` tool) for main and research roles.
This aligns the task expansion feature with the new architecture while using the more robust text generation approach due to current limitations with structured output services. Closes subtask 61.37.
Refactored the feature and related components (CLI command, MCP tool, direct function) to integrate with the unified AI service layer ().
Initially, was implemented to leverage structured output generation. However, this approach encountered persistent errors:
- Perplexity provider returned internal server errors.
- Anthropic provider failed with schema type and model errors.
Due to the unreliability of for this specific use case, the core AI interaction within was reverted to use . Basic manual JSON parsing and cleanup logic for the text response were reintroduced.
Key changes include:
- Removed direct AI client initialization (Anthropic, Perplexity).
- Removed direct fetching of AI model configuration parameters.
- Removed manual AI retry/fallback/streaming logic.
- Replaced direct AI calls with a call to .
- Updated wrapper to pass session context correctly.
- Updated MCP tool for correct path resolution and argument passing.
- Updated CLI command for correct path resolution.
- Preserved core functionality: task loading/filtering, report generation, CLI summary display.
Both the CLI command ([INFO] Initialized Perplexity client with OpenAI compatibility layer
[INFO] Initialized Perplexity client with OpenAI compatibility layer
Analyzing task complexity from: tasks/tasks.json
Output report will be saved to: scripts/task-complexity-report.json
Analyzing task complexity and generating expansion recommendations...
[INFO] Reading tasks from tasks/tasks.json...
[INFO] Found 62 total tasks in the task file.
[INFO] Skipping 31 tasks marked as done/cancelled/deferred. Analyzing 31 active tasks.
Skipping 31 tasks marked as done/cancelled/deferred. Analyzing 31 active tasks.
[INFO] Claude API attempt 1/2
[ERROR] Error in Claude API call: 400 {"type":"error","error":{"type":"invalid_request_error","message":"max_tokens: 100000 > 64000, which is the maximum allowed number of output tokens for claude-3-7-sonnet-20250219"}}
[ERROR] Non-overload Claude API error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"max_tokens: 100000 > 64000, which is the maximum allowed number of output tokens for claude-3-7-sonnet-20250219"}}
Claude API error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"max_tokens: 100000 > 64000, which is the maximum allowed number of output tokens for claude-3-7-sonnet-20250219"}}
[ERROR] Error during AI analysis: 400 {"type":"error","error":{"type":"invalid_request_error","message":"max_tokens: 100000 > 64000, which is the maximum allowed number of output tokens for claude-3-7-sonnet-20250219"}}
[ERROR] Error analyzing task complexity: 400 {"type":"error","error":{"type":"invalid_request_error","message":"max_tokens: 100000 > 64000, which is the maximum allowed number of output tokens for claude-3-7-sonnet-20250219"}}) and the MCP tool () have been verified to work correctly with this revised approach.
