skills.rest
alatyshau
alatyshauCOMMUNITY
Explore all skills by this author →
GitHub

execute

Executes implementation plan steps with status tracking and commit discipline.

Updated Nov 21, 2025
Search Tags:#implementation plan#task status#git commit#state machine#human review#ai agent workflow#step execution
One-click install
npx skills add https://github.com/alatyshau/duet --skill execute-alatyshau
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: execute
Source: https://github.com/alatyshau/duet/tree/main/packages/skills/modes/execute
Command: npx skills add https://github.com/alatyshau/duet --skill execute-alatyshau

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When an AI agent implements a multi-step plan, it often drifts: starting steps without permission, marking work done without human review, or losing track of what changed. This Skill enforces a disciplined execution workflow where every plan step moves through explicit statuses (TODO, WIP, IN_REVIEW, DONE) and only the human can close a step. ## Core Features & Use Cases - Status state machine: Each step transitions TODO → WIP → IN_REVIEW → DONE via explicit commands like /next, /done, and /done-next, so the agent never auto-continues. - Step-per-commit discipline: Every completed step ends with a commit whose message comes from the step's commit field, keeping code and spec/ documentation in sync. - Human-in-the-loop review: The agent stops at IN_REVIEW and waits for feedback, treating interjections like "ok" as feedback rather than commands. - Use Case: You have a topic file with an IMPLEMENTATION PLAN of five steps. You type /next, the agent executes step 1, updates its status to IN_REVIEW, commits the code plus spec changes, and waits for your /done before touching step 2. ## Quick Start Enter EXECUTE mode and type /next to have the agent start the first TODO step of your implementation plan.

Frequently Asked Questions about execute

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make an AI agent follow an implementation plan step by step?▼

Define steps in an IMPLEMENTATION PLAN with statuses TODO, WIP, IN_REVIEW, and DONE. The agent starts a step only on /next, moves it to IN_REVIEW when finished, and waits for /done before closing it.

How do I prevent an AI agent from marking tasks done without review?▼

Use a rule that only the human can set DONE status. The agent completes work, sets the step to IN_REVIEW, commits, and explicitly waits for /done or feedback instead of self-approving.

Can the agent start the next step automatically after /done?▼

No. The /done command only closes the current IN_REVIEW step. Starting the next step requires an explicit /next, or /done-next to close and start in one command.

What happens when a new session starts with a WIP step?▼

The agent starts in DIALOGUE mode, reports that a WIP step exists from the previous session, and asks whether to continue. It enters EXECUTE mode only after explicit confirmation.

How are git commits tied to plan steps?▼

Each step equals one commit made when the step moves to IN_REVIEW, using the message from the step's commit field. The spec/ directory must be updated before committing so code and documentation stay consistent.

Related Skills

Explore More Software Engineering Skills→
DecentralizedGeoDecentralizedGeo
1

writing-plans

Draft step-by-step implementation plans for multi-step software engineering tasks.

Official
Advanced
tincoppertincopper
10

trellis-finish-work

Archive completed Trellis tasks and write a session journal.

Community
Intermediate
TeamSloppyTeamSloppy
26

mode-auto

Select and execute the best route from the Auto route catalog.

Official
Advanced
BoChao1ZhangBoChao1Zhang

self-driven

Automate long-running tasks with a self-driven agent without human intervention.

Community
Advanced