neumie Complexity Assessment
Evaluate a task's complexity using a structured heuristic. This is a pure analysis skill — it reports a tier and rationale but does NOT execute anything.
Process
- Read the task description — what the user wants to accomplish
- Explore the codebase — grep for relevant terms, read related files, understand what's involved. When invoked from
task-start, exploration has already been done — use those findings directly rather than re-exploring. - Score each dimension (see heuristic below) with a one-line justification
- Report the tier and recommended approach
Heuristic
Score each dimension 1-3:
Scope — How many files and modules are affected?
- 1: 1-3 files in one module
- 2: 4-15 files or 2-3 modules
- 3: 15+ files or 4+ modules or new module creation
Clarity — How specific is the task?
- 1: Exact file, function, or error referenced
- 2: Clear goal, but approach needs figuring out
- 3: Vague goal, significant design decisions needed
Risk — What breaks if this goes wrong?
- 1: Isolated, easily reversible
- 2: Touches shared code or has moderate blast radius
- 3: Auth, payments, data integrity, or public API changes
Novelty — Has the codebase done this before?
- 1: Pattern already exists, just replicate
- 2: Extends existing patterns in new directions
- 3: No precedent in the codebase
Tiers
| Total Score | Tier | Meaning |
|---|---|---|
| 4-5 | Trivial | Straightforward — just do it |
| 6-8 | Moderate | Multi-step — break it down and work through sequentially |
| 9-12 | Complex | Significant — plan first, then execute phase by phase |
Output Format
## Complexity: <TIER> (score: <N>/12)
| Dimension | Score | Rationale |
|-----------|-------|-----------|
| Scope | <1-3> | <one line> |
| Clarity | <1-3> | <one line> |
| Risk | <1-3> | <one line> |
| Novelty | <1-3> | <one line> |
Recommended: <one-line approach based on tier>
Rules
- Always explore the codebase before scoring — don't guess from the task description alone
- Be honest about uncertainty — if you can't determine scope without deeper exploration, that itself suggests higher complexity
- Scores should reflect the actual codebase state, not abstract difficulty
- This skill can be invoked standalone ("how complex would X be?") or as part of
task-start