decompose

name	decompose
description	Decompose a complex task into atomic, independently verifiable steps before execution. Use when: the user invokes /decompose, a task has many dependent steps, the task is complex enough to risk compositional failure, or reasoning chains exceed 5 dependent steps. Targets: compositional failure, logic collapse, long reasoning chain drift.
user-invocable	true

Decompose — Anti-Compositional-Failure Breakdown

You are breaking a complex task into the smallest independently verifiable steps. LLMs fail when reasoning chains get long — error compounds at each step. The solution is to make each step atomic and checkpointed.

Step 1: Identify the Task

State the task being decomposed. If the user provided a task via $ARGUMENTS, use that. Otherwise, use the most recent task or request from the conversation.

Step 2: Atomic Decomposition

Break the task into steps where each step:

• Has a clear input (what it needs)
• Produces a clear output (what it creates or changes)
• Has a verification method (how to confirm it's correct)
• Is independently testable (can be validated without completing later steps)

Format each step as:

Step N: [action verb] [specific thing]
  Input:    [what this step needs]
  Output:   [what this step produces]
  Verify:   [how to confirm correctness]
  Depends:  [which prior steps must complete first]

Step 3: Dependency Analysis

Draw the dependency graph:

• Which steps can run in parallel? (no dependencies between them)
• Which steps are sequential? (output of one feeds input of next)
• What is the critical path? (longest chain of dependent steps)

Flag any step that depends on more than 3 prior steps — this is the compositional danger zone where errors compound.

Step 4: Checkpoint Strategy

Identify natural checkpoints — points where you should stop, verify accumulated work, and confirm direction before continuing:

• After completing each independent branch
• Before any step that is hard to reverse
• At the midpoint of the critical path
• Before any step that combines outputs from multiple branches

Step 5: Risk Assessment

• Chain length: How many dependent steps in the longest chain?
• Reversal cost: Which steps are hardest to undo?
• Failure cascade: If step N fails, which downstream steps are affected?
• Simplification: Can any multi-step sequence be replaced with a single operation?

Output

Present the decomposed plan as a numbered task list with dependencies, parallel opportunities, and checkpoints clearly marked. Include an estimated "compositional risk" rating:

• Low: Critical path <= 3 steps, all easily verified
• Medium: Critical path 4-6 steps, some verification requires judgment
• High: Critical path > 6 steps or verification is difficult — consider breaking into multiple sessions