Shrimp Task Manager MCP

Deeply analyze task requirements and systematically examine the codebase, assess technical feasibility and potential risks, if code needs to be provided use pseudocode format and only provide high-level logic flow and key steps avoiding complete code

Clear incomplete tasks and reset the task list

Delete incomplete tasks, but does not allow deleting completed tasks, ensuring the integrity of system records

Retrieve the instructional guidance for a specific task. You will complete the programming task based on this guidance. You must strictly follow the feedback and instructions provided by the tool — it is designed to guide you toward perfect task completion, not to execute the task for you. Severe Warning: Calling the executeTask tool does not mean you have completed the task. You must follow the step-by-step guidance returned by the tool to complete the task properly.

Retrieve complete detailed information of a task based on task ID, including untruncated implementation guides and verification criteria

Initialize project standards, call this tool when the user requests to generate or initialize a project standards document, also call this tool if the user requests to change or update project standards

Generate structured task lists, including complete status tracking, priorities, and dependencies

When you need to plan tasks or construct complex features, you can use this tool to receive task planning guidance. You are expected to strictly follow the step-by-step instructions provided by the tool when organizing your tasks. You may optionally choose to reference existing tasks for extended planning.

Critical Warning: All forms of assumptions, guesses, and imagination are strictly prohibited. You must use every available tool at your disposal to gather real information.

Conduct flexible and evolvable thinking processes, progressively deepen understanding and generate effective solutions through establishing, questioning, verifying, and correcting ideas. When encountering situations requiring data collection, analysis, or research, prioritize reviewing project-related code; if relevant code does not exist, query the web rather than speculate. Set nextThoughtNeeded to false when thinking is sufficient, otherwise adjust total_thoughts to extend the process

Search for tasks based on keywords or ID, display abbreviated task information

Critically review analysis results, assess solution completeness and identify optimization opportunities, ensure solutions conform to best practices, if code needs to be provided use pseudocode format and only provide high-level logic flow and key steps avoiding complete code

When you need to conduct in-depth research on programming-related topics, you can use this tool to enter a specialized research mode. This tool will guide you on how to use web search and code search tools to systematically research technical topics, ensuring research depth and breadth while avoiding topic deviation. Suitable for technical research, best practice exploration, solution comparison, and other scenarios.

Break down complex tasks into independent subtasks, establishing dependencies and priorities.

1. Granularity Control (Required Reading)

• ### Minimum Viable Task

Each subtask should be completable and verifiable by a single developer within 1–2 working days (approximately 8–16 hours).

• ### Maximum Complexity Limitation

A single subtask should not span multiple technical domains such as frontend, backend, and database. If cross-domain work is required, split it into multiple subtasks.

• ### Recommended Number of Tasks

Avoid splitting into more than 10 subtasks at once. If more are needed, submit them in prioritized batches (6–8 tasks per batch).

• ### Recommended Task Length

Each split should not exceed 5,000 characters. If it does, divide and submit in multiple batches.

• ### Depth Limitation The task tree should not exceed 3 levels:
• Level 1: Functional Modules
• Level 2: Main Processes
• Level 3: Key Steps

2. Task Splitting Example

• Identify core functionality points, and create a subtask for each.
• Annotate each subtask with:
• Input/Output
• Acceptance Criteria
• If needed, provide pseudocode:
• Only outline high-level logic and key steps.
• Avoid providing complete source code.
• Check dependencies between subtasks and specify them in the dependencies field.
• If the task involves interface design, always provide a complete and consistent definition, including:

• Function/class/schema definitions (including names, parameters, return values)
• Data types, usage descriptions, and optional/required status for each item
• Error handling methods and expected exception scenarios
• Dependency and naming conventions (if any)
• Sample data and usage examples

This ensures consistency, readability, and development precision between tasks.

3. Dependencies and Prioritization

• Mark each subtask with its dependencies list.
• Automatically compute and enforce execution order based on the dependency graph to prioritize the critical path.

4. Update Mode Explanation (updateMode)

When you need to create a new task that is not related to the current task list, be sure to use clearAllTasks to avoid task confusion.

• append: Keep existing unfinished tasks and add new ones.
• overwrite: Delete all unfinished tasks, keep completed ones.
• selective: Smart-match and update tasks by name.
• clearAllTasks: Clear all tasks and automatically back up the current list.

---

5. Strict JSON Rules

• ### No Comments Allowed

JSON does not support comments. Any use of # or // will cause parsing failures.

• ### Proper Escaping Required All special characters (e.g., double quotes \", backslashes \\) must be properly escaped, or they will be considered invalid.

6. Important Notes

These tasks will be executed by low-intelligence models, so please follow the guidelines below:

• Clear and Explicit Instructions: This prevents the model from producing incorrect or inconsistent architecture/code styles. Provide clear commands or specifications.
• Encapsulated Interfaces: Each task runs independently. Define the interfaces clearly — such as function names, parameters, return values — so that other task-executing models can easily understand how to interact with or integrate these functions.
• Dependencies: If there are dependencies between tasks, define the interaction interfaces first. Tasks do not need to know each other's implementation, but must know how to interact with one another.

Update task content, including name, description and notes, dependent tasks, related files, implementation guide, and verification criteria, completed tasks only allow updating summary and related files

Verify Task

Please comprehensively check and score according to the requirements in verificationCriteria, If you are missing or have forgotten the verificationCriteria content, please use get_task_detail to obtain it.

Please score according to the following rules:

Verification Standards

1. Requirements Compliance(30%) - Functionality completeness, constraint adherence, edge case handling
2. Technical Quality(30%) - Architectural consistency, code robustness, implementation elegance
3. Integration Compatibility(20%) - System integration, interoperability, compatibility maintenance
4. Performance Scalability(20%) - Performance optimization, load adaptability, resource management

score Parameter Guidelines

Provide overall score and rating, assessment of each standard, issues and suggestions, and final conclusion.

Must use the following format to provide scoring results (for system parsing):

score: [number from 0-100]

summary Parameter Guidelines

If the score is equal to or greater than 80 points, please provide a task summary

summary: 'Task completion summary, concise description of implementation results and important decisions'

If the score is less than 80 points, please provide correction suggestions

summary: 'List task issues and correction suggestions'