incremental-coding— agent skill

Overview

The biggest risk in software development is building a lot of code that doesn't work. Incremental coding limits this risk: build a little, verify it works, build more. At every step, the system is in a known-good state.

When to Use

• Any implementation that will take more than 2 hours
• When building in a complex domain you're uncertain about
• When multiple components need to integrate

Process

Step 1: Define the First Increment

1. What is the smallest possible thing you can build that provides value and can be verified?
2. It doesn't have to be feature-complete — just correct and verifiable.
3. Example: "Add the endpoint skeleton with hardcoded response" before adding business logic.

Verify: The first increment can be verified in under 5 minutes.

Step 2: Build → Verify → Commit

1. Build only the first increment.
2. Run tests. Verify manually if needed. Confirm it works.
3. Commit this working state.
4. Repeat for the next increment.

Verify: There is a working commit after each increment.

Step 3: Integration Continuously

1. Integrate with the real system as early as possible — not at the end.
2. Test against real dependencies (DB, API, etc.) as early as possible.
3. Fake integrations (mocks) should be replaced with real ones by the end.

Verify: By completion, all mocks replaced with real integration.

Common Rationalizations (and Rebuttals)

Verification

• [ ] Implementation built in verifiable increments
• [ ] Working commit exists after each increment
• [ ] No long stretches of "broken" state in git history
• [ ] All mocks replaced with real integrations by completion

References

• test-driven-development skill
• task-decomposition skill