recursive_task_decomposer

Recursive Task Decomposer Example

Demonstrates non-tail recursion by decomposing a complex task into subtasks, recursively expanding the non-atomic ones, and merging results into a cohesive plan.

Key Features

• Non-tail recursion: subtask calls return results that are merged into the final plan
• Structured input: task, context, and remaining depth are explicit
• Bounded depth: MAX_DEPTH caps recursion; runtime --max-depth is a safety cap

Running

export LLM_DO_MODEL="anthropic:claude-haiku-4-5"
llm-do examples/recursive_task_decomposer

The manifest configures max_depth: 10 and includes a default input for the sample task.

Sample Input

TASK: Plan a 2-day onboarding workshop for new backend engineers
CONTEXT: Remote team, 15 attendees, limited to 4 hours per day, include a hands-on lab
REMAINING_DEPTH: 2

Notes:

• REMAINING_DEPTH values <0 are treated as 0.

Expected Behavior

1. Break the workshop into agenda design, logistics, and lab preparation.
2. Recursively expand non-atomic subtasks while remaining depth > 0.
3. Merge returned subtasks into a numbered plan with success criteria within the (REMAINING_DEPTH + 1) * 4 line budget.
4. Never call the planner on the original task; only call it on specific subtasks.