Implement — A Claude Skill for Implementing Code According to Specifications or Tickets

Skill Overview

Implement is a Claude Code engineering skill that enables Claude to write code strictly according to specifications or ticket lists you have already prepared. It runs type checks and individual test files as it works, then runs the full test suite, performs a code review, and commits the changes to the current branch.

Use Cases

1. Specifications or tickets are ready and it is time to start coding. After writing a specification with to-spec or breaking the work down into tickets with to-tickets, use implement to have Claude execute the plan instead of improvising on the fly. It is the implementation stage of the “think first, then code” workflow.
2. You need to implement a complete body of work rather than make scattered changes. It is suitable for a complete feature, a refactoring task, or a batch of related tickets. Claude will continue working until the task is complete, repeatedly running type checks and individual-file tests along the way instead of leaving all validation until the end.
3. You want quality safeguards when changes are finalized. The skill includes three checkpoints: type checking, testing, and code review, followed by a commit. It is suitable for engineering teams and individuals who do not want to rely on vibe coding and prefer every step to be traceable.

Core Features

1. Implementation according to specifications/tickets: Uses the user-provided specification or tickets as the sole basis for implementing the described work. The skill itself is not responsible for clarifying requirements or breaking down tasks; those are the responsibilities of the upstream to-spec and to-tickets workflows.
2. Layered verification cadence: Requires regular type checks, regular runs of individual test files, and one full test-suite run after everything is complete. This “incremental validation + final full validation” cadence helps uncover issues earlier.
3. End-to-end completion: Once the work is complete, it invokes /code-review to review the changes and then commits the results to the current branch. TDD (/tdd) is used when possible, but only at pre-agreed seams, rather than being enforced throughout the entire process.

Frequently Asked Questions

Is implement triggered automatically?

No. This skill is configured with disable-model-invocation: true, so Claude will not decide to invoke it on its own; you must explicitly initiate it. This design ensures that the action of “starting to write code” is always triggered by a person, preventing AI from starting work without permission while you are still discussing the approach.

Do I need a specification or tickets before using implement?

Yes, this is a prerequisite. The skill is intended for “implementation,” and its input is a finalized specification or set of tickets. If you only have a vague idea, you should first use to-spec or to-tickets to clarify and break down the work, and then hand it over to implement. Skipping this step leaves it without a blueprint to follow.

Does it always use TDD? Will it commit the code automatically?

TDD is used “when possible and at pre-agreed seams”—in other words, you and Claude decide in advance which areas require tests to be written first; the skill does not independently decide to enforce test-first development throughout. Code is committed only after the work is complete and has passed /code-review. The commit targets the current branch; no separate branch is created, and nothing is pushed to the remote.