All categories/Agents and automation
Level Advanced · Splitting Agent Acceptance
Task splitting: down to acceptance criteria
Hand an agent one big task and it nibbles at everything and finishes nothing.
Prompt
Split this task.
Task: {{task}}
Time available: {{time}}
Tools available: {{tools}}
Known constraints: {{constraints}}
Requirements:
1. Split into 4-8 subtasks; each must have a completion test (what becomes visible or checkable).
2. Mark dependencies: which wait, which can run in parallel.
3. Mark the step most likely to fail and which step to fall back to.
4. Give an execution order and say why.
5. List separately any subtask that depends on information I have not given. Do not assume it.What to state before it starts
Leave these unsaid and the model fills them with an assumption. Most 'wrong answers' start here.
- {{task}}: paste the raw request, do not pre-split it
- {{tools}}: what it can actually call, not vague 'internet access'
- {{constraints}}: time, budget, off-limits areas
Go through each of these
Where this kind of work most often goes wrong. Already in the prompt, so you do not have to remember it.
- The completion test is the point; a subtask without one drags forever.
- The unknown-dependency list surfaces traps early.