Complex Task Decomposition

For tasks that are too large or risky for a single implementation run, the aitasks framework supports decomposition into child subtasks. This gives you controlled, disciplined execution of complex features while maintaining full context across all subtasks.

How It Works

• During the planning phase of /aitask-pick, if a task is assessed as high complexity, the skill automatically offers to break it into child subtasks
• You can also force decomposition by adding a line like “this is a complex task: please decompose in child tasks” in the task description
• Each child task is created with detailed context: key files to modify, reference patterns, step-by-step implementation instructions, and verification steps. This ensures each child can be executed independently in a fresh code agent context

Context Propagation Between Siblings

When implementing a child task, /aitask-pick automatically gathers context from previously completed siblings. The primary reference is the archived plan files in aiplans/archived/p<parent>/, which contain the full implementation record including a “Final Implementation Notes” section with patterns established, gotchas discovered, and shared code created. This means each successive child task benefits from the experience of earlier ones.

Typical Decomposition Flow

1. Create a parent task describing the full feature
2. Run /aitask-pick <parent_number> — during planning, choose to decompose
3. Define child tasks with descriptions and dependencies
4. Implement children one at a time with /aitask-pick <parent>_<child> (e.g., /aitask-pick 16_1, /aitask-pick 16_2)
5. When all children are complete, the parent is automatically archived