Implement Spec: Build from Blueprint

The implement-spec command is the final step in the repeatable feature development cycle in the spec-driven development workflow with Agent OS.

This phase involves:

• Executing task groups in logical order
• Delegating to specialized implementer roles (agents) to appropriate tasks
• Analyzing and reusing existing code patterns
• Documenting all implementation work
• Verifying quality through specialized verifier roles (agents)

What gets created

The implement-spec command—in addition to the actual code changes that are made—also creates implementation reports and verification documents in agent-os/specs/[this-spec]/:

• implementation/[task-group]-implementation.md - One report per task group documenting files modified, implementation decisions, database changes, test coverage, standards compliance, and known issues.
• verification/backend-verification.md - Backend quality verification report.
• verification/frontend-verification.md - Frontend quality verification report.
• verification/final-verification.md - Final verification ensuring all tasks complete, tests pass, and quality standards are met.

Note: In single-agent mode, a prompts/ folder is also created containing individual prompt files for each task group (more on that below).

How to run

Multi-Agent Mode

Run command, /implement-spec

When you're running Agent OS in multi-agent mode (typically in Claude Code), you can initiate and run the entire implement-spec phase with a single command:

/implement-spec

This automatically plans assignments, coordinates and delegates to implementer and verifier subagents, runs work in parallel (when applicable), and produces a final report.

Single-Agent Mode

Generate and run prompts sequentially

First, generate the implementation prompts:

@agent-os/commands/implement-spec/implement-spec.md run this

This creates individual prompt files in agent-os/specs/[this-spec]/implementation/prompts/.

Then run each prompt sequentially in your AI tool. You can simply @ reference each prompt markdown file and append "run this" (or similar) to your prompt.

@agent-os/specs/[date-feature]/implementation/prompts/1-[task-group].md run this

@agent-os/specs/[date-feature]/implementation/prompts/2-[task-group].md run this

Continue with remaining prompts in order. These will cover implementation of each task group along with verification of the work.

For full control, you can run each prompt individually and review the work before proceeding to the next prompt. But if your AI tool supports it, you can also queue them all up to run one after another automatically.

Next step

Congratulations! Your feature is now complete and ready to ship. Implementation is finished when all tasks are checked off, implementation documentation exists for each group, all tests pass, and verification reports confirm quality.

Important: Building with AI agents—with or without Agent OS—is rarely a one-and-done process. You should always review the implementation, put your final polish on it before shipping.

Your review should also serve as your opportunity to spot patterns in how your agents perform, so that you can optmize your standards and workflows for better performance and alignment.

Return to the workflow overview