Task Slicing Plan
Turn one larger spec into small, safe, independently reviewable tasks.
Source spec
- Spec file:
- Related issue:
- Human owner:
Slicing principles
- Prefer small diffs
- Prefer tests before implementation when behavior is known
- Keep discovery separate from implementation
- Avoid speculative refactoring
- Make dependencies explicit
- Make each task independently reviewable where possible
Tasks
Task 1: <short imperative title>
Goal:
Scope:
Out of scope:
Implementation notes:
Evals:
Risk: Low / Medium / High
Dependencies: None / Task N
Agent prompt:
textImplement only this task. Do not solve later tasks. Do not perform unrelated refactoring. Run the listed evals and report results.
Recommended execution order
- ...
Parallelizable tasks
- ...
Stop conditions
- unclear interface ownership
- failing existing tests unrelated to the task
- missing dependency
- contradictory specs
- security-sensitive behavior not specified