dev-workflow_skill

Overview

This skill orchestrates document-driven AI agent development as a per-task, phase-gated workflow. It enforces artifact-first outputs, heals missing dependencies automatically, and updates artifacts incrementally instead of overwriting them. It is designed for controlled, review-gated progress across intake, research, plan, todo, implement, and review phases.

How this skill works

The skill interprets user input to identify the task, requested phase(s), and mode (create or update). It checks required upstream artifacts and, if missing, performs minimal dependency healing before proceeding. Each phase produces or updates explicit artifacts (task setup, research.md, plan.md, todo.md, implementation notes, review.md) and records assumptions, risks, and next steps.

When to use it

• Starting a new agent development task and wanting structured phases
• Continuing or refining an existing task with incremental artifact updates
• Jumping into a later phase but requiring minimal upstream artifacts generated
• Enforcing review-gated progression to avoid premature big-bang implementation
• Keeping task scope isolated and recording adjacent work as follow-ups

Best practices

• Always provide or confirm a task id; if none, accept the inferred task and state the assumption
• Request one or a small number of phases per run for focused, inspectable outputs
• Allow the skill to auto-heal dependencies rather than asking for full rework
• Prefer research and planning phases before broad implementation
• Request explicit incremental updates rather than asking to overwrite artifacts

Example use cases

• User provides a feature request and asks for a plan: the skill runs intake->research->plan, generating research.md and plan.md
• User requests implement but only has intake: the skill auto-heals research and plan, creates todo.md, then produces implementation notes
• User iteratively refines existing todo items: the skill updates todo.md and implementation notes without overwriting prior artifacts
• User asks for a review of partial implementation: the skill reviews existing artifacts, lists gaps, and recommends next steps
• Team adoption: use the skill to standardize per-task artifact templates and review gates across projects