implement-step

Dispatches a single plan step to an isolated worker context and relays a structured report.

Updated May 26, 2026
One-click install
npx skills add https://github.com/robot-denny/ai-sketchlab --skill implement-step-robot-denny
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: implement-step
Source: https://github.com/robot-denny/ai-sketchlab/tree/main/.agents/skills/implement-step
Command: npx skills add https://github.com/robot-denny/ai-sketchlab --skill implement-step-robot-denny

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Long implementation plans clutter the main conversation with step-by-step execution details, making it hard to track progress and review work. This Skill executes one plan step in a clean, isolated worker context so the main conversation stays focused while still enforcing the plan's TDD and validation contract. ## Core Features & Use Cases - Isolated Step Execution: Parses a plan file, extracts the target step plus its Context and Key Decisions sections, and dispatches a self-contained prompt to a fresh worker context. - Contract Enforcement: The worker prompt enforces TDD red-green ordering, runs every validation command, requires evidence for non-mechanical checks, and forbids out-of-scope changes or unrequested commits. - Structured Reporting: Relays a verbatim DONE or BLOCKED report with files changed, validation results, and a Next pointer for continuing the plan. - Use Case: After writing a spec and plan for a feature, run the skill with the plan name and step number to execute step 3 without polluting your main session, then review the diff and proceed to step 4. ## Quick Start Ask the AI to run implement-step with your plan name and step number, for example implement step 2 of the auth-refactor plan.

Frequently Asked Questions about implement-step

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

FAQPage Schema
How do I execute a single step from an implementation plan?

Invoke the skill with two arguments: the plan identifier (slug or path) and the step number. It reads the plan, extracts that step plus the Context and Key Decisions sections, and dispatches the work to an isolated worker context.

How does the skill enforce TDD during plan execution?

If the step says "Test first", the worker prompt requires writing a failing test, confirming RED, implementing, then confirming GREEN. It also requires running every command listed under the step's Validation section and reporting each result.

What happens if the working tree is dirty before dispatching?

The skill runs git status --short and, if uncommitted changes exist, warns that the worker will edit files on top of them and asks for confirmation before proceeding. A clean tree skips the prompt.

Does the worker commit changes after completing a step?

By default no — changes are left in place for review via git diff and /code-review. If the step explicitly instructs a commit, the step wins and the worker notes the commit in its report.

What happens when a plan step is blocked or fails validation?

The worker stops and reports BLOCKED with what it tried and what blocked it, rather than thrashing. The orchestrator relays the report verbatim and suggests resolving the blocker, then re-invoking the same step.

When should I not use implement-step for a plan item?

Do not use it when the step is a spell-cast (e.g., running /feature update) rather than implementation work — the skill detects this and tells you to cast the spell directly instead of dispatching a code worker.