| name | model-handoff |
| description | At the end of writing an implementation plan, evaluate the full plan's complexity and append a Model Handoff Recommendation block with the recommended model and escalation triggers. |
Model Handoff
After writing an implementation plan, evaluate the plan's concrete complexity and recommend which model should execute it.
When This Skill Fires
At the end of writing an implementation plan — after all tasks are defined but before the plan is handed off for execution.
Why This Exists
The thinking phase (brainstorming + planning) always runs on max-power/Opus. But implementation may not need Opus. This skill evaluates the concrete plan (not just the user's description) and recommends the cheapest safe model for execution.
This is more accurate than initial triage because:
- The exact files to change are known.
- The algorithmic approach is decided.
- Edge cases have been identified during planning.
- Domain complexity is concrete (e.g., "modifies GLSL shader" vs "something in mv/").
Evaluation
Score the same 6 dimensions as model-triage, but against the concrete plan:
| Dimension | What to evaluate in the plan |
|---|
| Algorithm | Do any tasks require non-trivial logic design? |
| Scope | How many files does the entire plan touch? |
| Domain | Do any tasks touch specialized domains (GPU, audio, protocols)? |
| Ambiguity | Are all tasks fully specified or do some need interpretation? |
| Edge cases | Do any tasks involve concurrency, timing, or complex error handling? |
| Risk | Could any task corrupt data, break core features, or be hard to revert? |
Decision Rules
- Total >= 15 → recommend strong model (Opus).
- Any single dimension >= 4 → recommend strong model (Opus) (one hard step makes the whole plan hard).
- Total < 15 AND all dimensions < 4 → recommend fast model (Sonnet).
Output
Append this block to the end of the plan document:
## Model Handoff Recommendation
**Recommended model:** Sonnet (or Opus)
**Score:** X/30
**Reasoning:** [1-2 sentences explaining why this model is appropriate]
### Escalation Triggers
If you encounter any of these during implementation, warn the user to consider switching to the strong model (Opus):
- [Plan-specific trigger 1]
- [Plan-specific trigger 2]
- [Plan-specific trigger 3]
The escalation triggers must be specific to this plan — not generic. Generate them based on the plan's identified risk areas.
Integration with Planning Skills
This skill runs AFTER the planning skill (e.g., superpowers:writing-plans) completes. It reads the full plan, evaluates it, and appends the handoff block. It does NOT modify the plan's tasks.
