plan-dev

Creates implementation plans in plan mode and persists plan and research artifacts under docs/agents.

Updated Apr 28, 2026
One-click install
npx skills add https://github.com/byron1st/personal-harness --skill plan-dev-byron1st
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: plan-dev
Source: https://github.com/byron1st/personal-harness/tree/main/skills/plan-dev
Command: npx skills add https://github.com/byron1st/personal-harness --skill plan-dev-byron1st

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? It turns vague feature or refactoring requests into reviewed, persistable implementation plans with acceptance contracts and authority boundaries, so a separate executor session can implement them without re-deriving the direction. ## Core Features & Use Cases - Plan-mode interview workflow: Refines requirements through questions and an optional read-only planner persona, keeping everything read-only until human approval. - Structured artifacts: Writes plan files, optional multi-step sub-plans, and research files with an index under docs/agents, following enforced naming and frontmatter conventions. - Acceptance Contract and Authority Boundaries: Records observable completion criteria, executor discretion limits, and stop conditions so independent reviewers can judge the finished work. - Use Case: Before refactoring a service layer, run plan-dev to interview for goals and non-goals, mine an existing SPEC.md, consult prior research, and persist an approved plan that an implement-dev session executes cold. ## Quick Start Ask the agent to plan the implementation of your feature before writing any code, then approve the plan it presents.

Frequently Asked Questions about plan-dev

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

FAQPage Schema
How do I create an implementation plan before coding with an AI agent?▼

Invoke plan-dev and describe the work; it interviews you in plan mode, optionally consults a read-only planner persona, and presents a plan for approval. After approval it writes the plan and research files under docs/agents.

What is the difference between single-step and multi-steps plan modes?▼

Single-step is the default and produces one plan file for features, refactors, and small-to-medium work. Multi-steps is explicit opt-in only and produces a main plan plus standalone STEP-N sub-plans, each executed individually by implement-dev.

Can plan-dev write code or modify files during planning?▼

No. Everything before approval is read-only: reads, searches, questions, and the planner persona. The first writes happen only after approval, persisting research files and their index first, then the plan files.

When should I not use a multi-step plan breakdown?▼

Avoid multi-steps unless the user explicitly requests it, such as for new projects or large initiatives. For typical features and bug fixes, a single-step plan is the default and keeps review overhead low.

Why does the plan avoid line-level code details?▼

The executor runs in a separate session with no memory of planning, so the plan fixes direction (goals, non-goals, key decisions) while mechanics are left to the implementer. Deep investigation goes into research files instead.