Skip to content

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:

  • Budget Fast and cheap coding model. Good for scaffolding, small UI work, and routine refactors.
  • Balanced Default high-quality coding model. Use this for most implementation tasks.
  • High-capability Expensive 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:

  • Low Prefer speed and lower cost over deep exploration.
  • Medium Default depth for bounded feature work.
  • High Use 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 Balanced tasks. Use an Opus-class model only for High-capability tasks.
  • Copilot-style tools Use the default strong coding model for Balanced tasks. Switch to the premium or deepest reasoning option only for High-capability tasks.
  • GPT-style tools Use a mini model for Budget, a standard frontier coding model for Balanced, and the strongest reasoning configuration for High-capability.
TaskGoalDepends OnModel ClassReasoningCost Profile
TASK-001ASolution and project bootstrapNoneBudget or BalancedLowLow
TASK-001BDomain contracts and canonical app state001ABalancedMediumMedium
TASK-001CFake vendor boundary and seeded machine data001A, 001BBalancedMediumMedium
TASK-001DCommand handling and terminal workflow behavior001B, 001CHigh-capabilityHighHigh
TASK-001EBounded streams, diagnostics, and defect flow001C, 001DHigh-capability or BalancedHighMedium to High
TASK-001FMain operator UI and state projection001B, 001D, 001EBudget or BalancedLow to MediumLow to Medium
TASK-001GAutomated tests and verification hardening001D, 001E, 001FBalancedMediumMedium

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

Docs-first project memory for AI-assisted implementation.