| name | incremental-implementation |
| description | Delivers changes incrementally. Use when implementing any feature or change that touches more than one file. Use when you're about to write a large amount of code at once, or when a task feels too big to land in one step. |
Incremental Implementation
Execution-discipline skill for building changes in thin, verifiable slices.
When to Use
- Implementing any multi-file change.
- Building a new feature from an approved task breakdown.
- Refactoring existing code without losing control of scope.
Load / Do Not Load
- Load this skill during implementation for multi-file work.
- Do not use it for trivial one-file edits where the scope is already minimal.
Core Workflow
- Pick the Smallest Useful Slice: Choose one complete, testable step instead of a wide partial rewrite.
- Prefer the Simplest Viable Change: Implement the minimum code that solves the slice. Do not add speculative abstractions, extra configurability, or future-proofing that the request does not need.
- Keep Changes Surgical: Avoid mixing feature work with adjacent cleanup, formatting churn, or unrelated refactors. Remove only the imports, variables, or helpers that the current slice made unused. When current-scope cleanup appears to meet Phase 3.5 triggers, finish the current slice first and route it through
refactoring-and-simplification instead of mixing refactor work into the feature slice.
- Verify Before Expanding: Run the relevant tests or checks after each slice.
- Carry Forward Only Stable State: Leave the repo compilable and reviewable between slices.
- Repeat Until Done: Expand slice-by-slice instead of front-loading the entire feature.
Ask / Escalate
- Ask when the implementation cannot be sliced safely without a feature flag, migration plan, or contract-first step.
- Escalate when the requested scope would force a broken intermediate state or broad cross-cutting refactor.
References
- Behavioral overlay:
../../references/behavioral-guidelines.md
- Detailed guidance:
references/guidance.md
