Implementation Guide

Overview

The Implementation Guide takes a SpecWriter specification and breaks it into a structured, multi-session plan. Instead of dumping an entire spec into one Claude Code session and hoping for the best, the guide divides the work into ordered sessions — each with a clear scope, a list of files to read, a list of files to create or modify, a prompt you can send to Claude, and a verification checklist to confirm the session's work is complete.

How It Works

Session Cards

Each session card in the guide contains:

• Session title — A short name describing the unit of work (e.g. "Database schema and migrations", "API endpoints", "Frontend components")
• Status icon — Shows whether the session is done, active, or pending
• Scope section — What this session covers and its boundaries
• Read section — Files Claude should read for context before making changes
• Files list — Specific files to create or modify (collapsible)
• Copy Prompt button — Copies a pre-built prompt to your clipboard, ready to paste into a Claude Code session
• Send to Chat button — Sends the prompt directly to your active Claude Code session

Progress Tracking

A progress bar at the top of the guide shows overall completion. As you mark sessions complete, the bar fills and the session cards update:

Verification Checklists

Each session card includes a Verify section with concrete, testable assertions:

- [ ] User table migration runs without errors
- [ ] API returns 401 for unauthenticated requests
- [ ] Login form validates email format before submission

Check items off as you verify them. When all items in a session are checked, the Mark Session Complete button turns green — click it to advance.

Verify for Me

Click the Verify for me button to have Claude Code automatically run through the verification checklist. Claude reads the relevant files, checks each assertion, and reports which items pass and which need attention. This saves time on mechanical verification steps like "file exists" or "endpoint returns correct status code."

Working with the Guide

Recommended Workflow

1. Start from the top. Sessions are ordered by dependency — later sessions build on earlier ones.
2. Open a new Claude Code session for each guide session. This keeps context focused and prevents hitting context limits.
3. Use Copy Prompt to get a pre-built prompt that includes the scope, files to read, and expected output for that session.
4. Check the verification items after Claude finishes. If something doesn't pass, ask Claude to fix it before moving on.
5. Mark the session complete once all verification items pass.

Reopening the Guide

The guide persists as long as it's active. You can:

• Dismiss the guide (with a confirmation dialog) to close it
• Open Spec via the footer button to jump back to the original specification
• Navigate away and return — the guide remembers your progress

Self-Drive: Automatic Implementation

Instead of working through sessions manually, you can let Self-Drive automate the entire process. Self-Drive is an AI orchestrator that sends prompts, verifies output, fixes errors, runs builds and tests, and advances through sessions — all without manual intervention.

When Self-Drive is active, a status strip appears at the top of the Guide panel showing the current phase, session number, and elapsed time. Decision Cards appear in the Chat Panel showing each orchestrator decision (advance, fix, verify, pause) with color-coded icons and summaries.

To use Self-Drive:

1. Configure an orchestrator provider and model in Settings > Self-Drive
2. Click Start Self-Drive at the top of the Guide panel
3. Monitor progress via Decision Cards and the Guide panel — or walk away and come back later

Self-Drive automatically pauses when it encounters an issue it cannot resolve, letting you step in, fix it manually, and resume.

When to Use It

The Implementation Guide is most valuable for medium-to-large features that would exceed a single session's context window. If your spec describes work that touches multiple files across different layers (database, API, frontend), the guide ensures nothing gets lost between sessions.

For small, single-session features, you can skip the guide and use SpecWriter's Send to Chat button directly.

For guides with 3 or more sessions, consider using Self-Drive to automate the entire implementation — it handles the repetitive prompt-verify-fix cycle for you.

Tips

• Don't skip verification. The checklist is your safety net — it catches issues early before they compound across sessions.
• Use "Verify for me" for mechanical checks (file exists, type compiles, endpoint responds). Save manual verification for visual and behavioral checks.
• Copy Prompt includes context. The generated prompt tells Claude which files to read and what to build, so you don't need to repeat the spec manually.
• Mark sessions complete honestly. Skipping verification and marking done defeats the purpose — later sessions may depend on earlier work being correct.
• Use Self-Drive for larger guides. For 3+ session guides, Self-Drive automates the repetitive work and catches errors automatically with build and test checks.

Related Articles

• Self-Drive: Autonomous Implementation — Automate the entire Implementation Guide
• SpecWriter: Writing Better Specs — Create the specification that feeds the Implementation Guide
• SpecWriter: Feature Mode — Codebase-aware specs for existing projects
• Working with Multiple Sessions — Strategies for managing parallel and sequential sessions
• Building a Feature from Scratch — End-to-end guide using SpecWriter, the Implementation Guide, and Self-Drive