SLICE-001 Task Pack
This page breaks TASK-001 into smaller AI-friendly implementation tasks.
How To Use This Pack
- Run one task per AI chat thread.
- Finish tasks in dependency order unless you intentionally rebalance the plan.
- Give the AI only the task you want it to implement, not the whole epic.
- Reuse the copy/paste prompt from the task page as the starting point.
- Do not let one tool implement future tasks early unless the current task explicitly requires it.
Model Class Legend
Use these provider-neutral model classes so the guidance stays stable even when product names change:
BudgetFast and cheap coding model. Good for scaffolding, small UI work, and routine refactors.BalancedDefault high-quality coding model. Use this for most implementation tasks.High-capabilityExpensive top-tier model. Reserve for concurrency, cancellation, orchestration, or debugging tricky behavior.
If your tool exposes explicit reasoning levels, use the level named in each task:
LowPrefer speed and lower cost over deep exploration.MediumDefault depth for bounded feature work.HighUse for state transitions, cancellation, bounded streams, and race-prone behavior.
Provider Mapping
Map the model classes to whatever your tool currently offers:
- Claude-style tools Use a Sonnet-class model for most
Balancedtasks. Use an Opus-class model only forHigh-capabilitytasks. - Copilot-style tools Use the default strong coding model for
Balancedtasks. Switch to the premium or deepest reasoning option only forHigh-capabilitytasks. - GPT-style tools Use a mini model for
Budget, a standard frontier coding model forBalanced, and the strongest reasoning configuration forHigh-capability.
Recommended Execution Order
| Task | Goal | Depends On | Model Class | Reasoning | Cost Profile |
|---|---|---|---|---|---|
| TASK-001A | Solution and project bootstrap | None | Budget or Balanced | Low | Low |
| TASK-001B | Domain contracts and canonical app state | 001A | Balanced | Medium | Medium |
| TASK-001C | Fake vendor boundary and seeded machine data | 001A, 001B | Balanced | Medium | Medium |
| TASK-001D | Command handling and terminal workflow behavior | 001B, 001C | High-capability | High | High |
| TASK-001E | Bounded streams, diagnostics, and defect flow | 001C, 001D | High-capability or Balanced | High | Medium to High |
| TASK-001F | Main operator UI and state projection | 001B, 001D, 001E | Budget or Balanced | Low to Medium | Low to Medium |
| TASK-001G | Automated tests and verification hardening | 001D, 001E, 001F | Balanced | Medium | Medium |
Notes On Task Size
These tasks are intentionally sized so that:
- each task has one clear ownership boundary
- each task can usually be completed in one focused AI session
- the most expensive models are used only for the parts that benefit from them
- reviewers can verify results against a bounded scope