The MySpec workflow turns an idea into four spec files a coding agent can build from. It is the default for Greenfield projects and the right choice for most new products and features. In the picker it is MySpec generate-workflow (v1).
At a glance
- Best for: A new product or feature, built from scratch
- Project type: Greenfield (or Auto)
- Stages: Interview → Constitution → Requirements → Solution → Tasks
- Files:
constitution.md,requirements.md,solution.md,tasks.md - Folder:
specs/<bundle>/, where the bundle is a short name the AI picks, such asacme-portal
Each file builds on the ones before it. The constitution sets the rules, the requirements say what the system must do within those rules, the solution says how to build it, and the tasks say in what order.
Stages
Every stage after the Interview runs Collecting → Generating → Review. In Collecting the AI only asks about real gaps: it reads your earlier answers and files first and doesn’t ask the same thing twice.
Interview
The AI learns about the product, not the technology:
- the problem it solves and why now;
- what users interact with (a web app, a mobile app, an API, a command-line tool, and so on);
- the domain, the target users and their main pain points;
- the core features and the two or three key user journeys;
- the scope of the first version and how you will know it succeeds.
It asks in short rounds of multiple-choice questions, usually two or three. It doesn’t ask about languages, frameworks or architecture yet: the Constitution stage covers those. When it has enough, it names the session and the bundle.
MySpec may then show Recommended domain knowledge: Custom Skills that match your product, such as banking or healthcare. See Custom Skills.
Constitution
Collecting asks about engineering constraints: your core engineering principles, the technologies you must or must not use, architecture style, testing approach, coding standards, security, performance targets and integration points.
Generating writes constitution.md: the project vision, core principles and
the constraints for technology, architecture, testing, coding standards,
security, performance and integrations.
Review checks that the constitution is complete, specific and consistent. Every later file must stay consistent with it, and every later review checks that it does.
Requirements
Collecting asks about functional requirements and the journeys behind them, acceptance criteria, measurable non-functional targets, user roles and access, data, integrations, errors and edge cases.
Generating writes requirements.md: user roles, functional requirements
(FR-001, FR-002 …) with acceptance criteria, and non-functional requirements
(NFR-001 …) with measurable targets. Acceptance criteria follow the EARS+
patterns. See Writing requirements (EARS+).
Review checks that the requirements follow the constitution and the EARS+ patterns, cover every functional area, edge case and error, and stay technology-neutral: they say what the system does and how well, never which framework does it.
Solution
Collecting asks about architecture style, integrations and protocols, data model hints and the trade-offs that matter to you. Questions offer two to four standard options and stay within the technologies your constitution allows.
Generating writes solution.md: the technology stack and design decisions,
the architecture with a diagram, the system modules, the data model with an
entity-relationship diagram, API design, security, deployment and operations,
and success criteria. Every functional requirement gets a matching part of the
solution.
Review checks requirements coverage and constitution compliance first, then scalability, security, maintainability and documentation.
Tasks
Collecting asks about delivery: milestones, sequencing, which features come first, how fine-grained tasks should be, and testing, deployment and documentation priorities.
Generating writes tasks.md: milestones of checkbox tasks, each with
implementation details, acceptance criteria, dependencies, the requirements it
covers and a size (Small, Medium or Large). A Dependency Graph at the end
shows the critical path and what can run in parallel.
Review checks that every requirement is covered, and that tasks are well-formed, correctly sized and in a workable order.
When you finish the Tasks review, the session is complete. See Finish a session and build.
Tips
- Say what you already know. Tech choices you mention during the Interview become defaults that later stages confirm instead of asking from scratch.
- Spend time on the constitution. It is the shortest file but steers everything else, and every review checks against it.
- Go back instead of starting over. If the solution shows a requirement was
wrong, use
/gobackto revise Requirements. MySpec asks follow-up questions and then works forward through the later stages again.
When to pick another workflow
- Your organization needs measurable quality targets, compliance requirements and release gates: use MySpec Enterprise.
- Your team already works with GitHub’s Spec Kit: use SpecKit.
- You are changing an existing codebase: use MySpec brownfield or OpenSpec.
See Spec files for what each file contains in detail.