incremental-implementation

name	incremental-implementation
description	Build features in small, verifiable steps with continuous test validation. Use when implementing any feature from a plan. Each step should produce a working, tested increment. Prevents large, unverified code drops.
version	1.0.0
permissions	["fs.read","fs.write","shell.exec"]
triggers	[{"context":"implementing a feature"},{"context":"following a plan"},{"command":"/build"}]
platforms	["claude-code","cursor","gemini-cli","copilot","codex","opencode","windsurf","kiro"]
dependencies	["test-driven-development@^1.0.0","planning-and-task-breakdown@^1.0.0"]
model_variants	{"claude":{"enforcement_style":"human-partner collaborative"},"gemini":{"enforcement_style":"structured-checklist"},"gpt":{"enforcement_style":"directive-imperative"}}
author	codehands-core
signed	true
tier	1

Overview

Incremental implementation turns plans into code through small, verified steps. Each step adds ONE behavior, proves it works with a test, and leaves the codebase in a working state. No large code drops. No "I'll test it all at the end."

When to Use

• When implementing tasks from an approved plan
• When building any feature larger than a single function
• When modifying existing code that has tests
• NOT for exploratory spikes (use a throwaway branch)

Process

The Increment Cycle

For every plan task, repeat this cycle:

1. Read the task → 2. Write failing test → 3. Implement minimum →
4. All tests pass → 5. Commit → 6. Next task

Rules

1. One behavior per increment. Each increment adds exactly one testable behavior. Not two. Not "this and also while I'm here..."
2. Tests pass after every increment. The FULL test suite, not just the new test.
3. Commit after every green increment. Small commits with clear messages. Each commit is a valid rollback point.
4. Never break existing tests. If your change breaks an existing test, understand why before proceeding. Don't just update the test to match.
5. Follow the plan order. Tasks have dependency ordering for a reason. Don't skip ahead.

When Things Go Wrong

• Test won't pass → Use codehands:systematic-debugging — Don't thrash.
• Task is bigger than expected → Split it. Re-plan mid-implementation if a task will take > 15 minutes.
• Requirements are unclear → Ask. Don't implement an assumption.
• Existing code needs changes you didn't plan → Note it, complete the current task, then add a cleanup task.

Common Rationalizations

Rationalization	Reality
"I'll implement the whole feature and test at the end"	Large code drops are where bugs hide. Every increment that isn't tested is a gamble.
"These two changes are related, I'll do them together"	Related is not the same as inseparable. Implement one, test it, commit, then implement the other.
"Committing after every small change creates noise"	Small commits are not noise — they're rollback points. git log --oneline is easy to read. Debugging a 500-line commit is not.

Red Flags

• Uncommitted changes spanning multiple unrelated behaviors
• Tests not run between increments
• "I'll add tests at the end" for any increment
• Skipping plan task order without re-planning

Verification

• Each increment adds exactly one testable behavior
• All tests pass after every increment
• Each increment is committed with a descriptive message
• Plan tasks completed in dependency order

See Also

• codehands:test-driven-development — TDD cycle within each increment
• codehands:planning-and-task-breakdown — Plan that drives increments
• codehands:source-driven-development — Read before writing each increment