The Anatomy of a Delegatable Task: Structured Definitions for AI Agents
Vague task descriptions produce vague results. A structured template ensures AI agents have everything needed for autonomous execution.
Ai-Planning
The Idea
A structured task definition template that ensures AI agents have everything needed for autonomous execution.
Context
The plan agent produced tasks with consistent structure that enabled clean delegation:
### Task N: [Name]
**What to do**:
- Specific actions with code examples
**Must NOT do**:
- Guardrails to prevent scope creep
**Recommended Agent Profile**:
- **Category**: `quick|visual-engineering|ultrabrain|...`
- **Skills**: [`skill-1`, `skill-2`]
- **Skills Evaluated but Omitted**: [with reasoning]
**Parallelization**:
- **Can Run In Parallel**: YES/NO
- **Parallel Group**: Wave N
- **Blocks**: Tasks X, Y
- **Blocked By**: Tasks A, B
**References**:
- File paths with line numbers
- External docs
- WHY each reference matters
**Acceptance Criteria**:
- Bash commands agent can run to verify
- Expected outputs
**Commit**: YES/NO
- Message format
- Files to include
- Pre-commit verification
Raw Exchange
Plan output included 7 tasks with this structure Each task was delegatable with zero ambiguity Verification commands were agent-executable (grep patterns, exit codes)
Read more