# Implement

> Implement approved work from a PRD or issue set with tight feedback loops.

- Skill: `youssefguba/implement` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add youssefguba/implement`
- Raw SKILL.md: https://api.skillmd.com/api/skills/youssefguba/implement/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Product & Planning
- Author: Youssefguba (https://skillmd.com/u/youssefguba)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/youssefguba/implement

---


# Implement

Implement the work described by the user's PRD or approved issues. Work the
frontier: begin only issues whose blockers are complete, and finish one thin
vertical slice before starting the next.

## 1. Ground the work

Read the source PRD, every selected issue, comments, repository instructions,
domain documentation, and relevant architecture decisions. Confirm the fixed
test seam and the exact acceptance criteria before changing code.

## 2. Build with TDD

If a TDD skill is installed, invoke it at the agreed seam. Otherwise run the
same discipline directly:

1. **Red**: write one external-behavior test and observe the expected failure.
2. **Green**: write only enough implementation to pass that test.
3. Repeat one behavior at a time.
4. **Refactor** only while green, rerunning the targeted test after each change.

Do not write all tests before all implementation. Avoid mocks except at real
system boundaries.

## 3. Keep feedback tight

- Run formatting, linting, and type checking regularly.
- Run the current targeted test after every meaningful edit.
- Run the full validation suite once the complete slice is green.
- Verify acceptance criteria through the highest available external seam.

## 4. Review on two axes

If a code-review skill is installed, invoke it against the fixed point captured
before implementation. Otherwise review the diff independently on two axes:

- **Standards**: repository instructions, maintainability, security, and common
  code smells.
- **Spec**: missing requirements, incorrect behavior, and scope creep against
  the originating PRD and issues.

Fix confirmed findings, then rerun targeted and full validation.

## 5. Finish

Commit the completed work to the current branch using a message that references
the implemented issue where practical. Push only when the user has requested
publication. Report delivered behavior, verification evidence, remaining
risks, and links to the source work.

