skills/davidkiss/smart-ai-skills/subagent-task-execution

subagent-task-execution

SKILL.md

Subagent Task Execution

Overview

This skill is used to execute a set of tasks (typically from a task breakdown or plan). It focuses on high-quality execution by dispatching specialized subagents for each task, ensuring each task is performed correctly and verified before moving to the next.

Core Principle

One Task, One Subagent, Multi-Stage Review. By isolating each task and applying a structured review process, we ensure high quality and prevent context pollution.

The Process

  1. Preparation:

    • Read the entire plan or task breakdown.
    • Extract all tasks with their full context.
    • Create a structured todo list using the TodoWrite tool.
  2. Per-Task Execution Loop:

    • Skill Discovery: Before starting a task, identify which specialized skill(s) are needed for its execution (e.g., coding, brainstorming, or any project-specific skills).
    • Dispatch Implementer: Dispatch a subagent to perform the task. Provide the subagent with the full task description, any identified skills, and the necessary context.
    • Interactive Clarification: If the subagent has questions, answer them clearly before they proceed with implementation.
    • Task Implementation: The subagent performs the task, following the identified best practices and skills.
    • Verification & Review:
      • Spec Compliance Review: A separate subagent verifies that the result exactly matches the task requirements (no more, no less).
      • Quality Review: A final review stage to ensure the output meets the highest standards of the domain (e.g., code quality, documentation clarity, etc.).
    • Completion: Mark the task as completed in the todo list.
  3. Finalization:

    • Once all tasks are complete, perform a final holistic review of the entire body of work to ensure consistency and overall quality.

Key Principles

  • Discovery First: Always look for and apply the most relevant skills for the specific task at hand.
  • Isolation: Each task should be treated as a distinct unit of work.
  • Review Loops: Never move to the next task if the current one has open issues from either the spec or quality reviews.
  • Subagent Specialization: Use subagents as specialized workers, providing them with all the context they need upfront.

Red Flags

  • Skipping the skill discovery phase.
  • Combining multiple tasks into a single subagent invocation.
  • Proceeding with a task while reviews are still pending or failing.
  • Ignoring subagent questions or providing vague answers.
  • Failing to update the todo list as progress is made.
Weekly Installs
106
First Seen
Feb 15, 2026
Installed on
opencode106
gemini-cli75
amp75
github-copilot75
codex75
kimi-cli75