-
Notifications
You must be signed in to change notification settings - Fork 2
Home
Gregor Biswanger edited this page Aug 26, 2026
·
5 revisions
Spec-driven development that works in Claude Code and GitHub Copilot — from the same files.
The README gets you running in two minutes. This wiki holds everything else: the reference material, the reasoning behind the design, and the details you only need once you are actually using it.
Already running an older FeatherSpec? Since v1.2.0 the template is versioned and every existing project — however customized — can update safely. Copying two files in is the whole install: Updating & Versioning → Installing the updater. Cutting a template release with an AI agent: hand it Releasing.
| Page | What's in it |
|---|---|
| Getting Started | Install, open in either tool, verify what actually loaded |
| Hands-On Walkthrough | A full SDD cycle in ~45 minutes, PO and Dev in pairs |
| Commands | Every /sdd-* command in detail |
| Updating & Versioning | Check your version, update safely, what never gets touched |
| Page | What's in it |
|---|---|
| Specs and Plans | Document structure, lifecycle, acceptance criteria, traceability |
| Memory Bank | The four files and what belongs in each |
| Repository Layout | Every file and folder, and why it exists |
| Configuration | Shared vs. local settings, auto memory, MCP, hooks |
| Extending FeatherSpec | Add your own commands and rules |
| Page | What's in it |
|---|---|
| Interop Matrix | Exactly what VS Code Copilot reads from .claude/, with sources |
| Design Decisions | Why commands and not skills, why one source of truth, the documented exceptions |
| Committing to One Tool | Strip out the other tool later, mechanically |
| Migrating from copilot-spec-driven-template | What changed from the predecessor |
| Releasing | How a release is cut — give this page to your AI agent |
| Page | What's in it |
|---|---|
| Troubleshooting | Commands not showing up, rules not applying, diagnostics |
| FAQ | Short answers to the recurring questions |
-
Is: a repository scaffold that makes an AI coding assistant work spec-first — write a
spec, have a cold second pass attack it, plan against it, read the plan, implement, prove it,
and keep a small version-controlled Memory Bank in sync. The same
/sdd-*commands exist in both tools because each workflow is written down once, so it reads identically in docs, videos, and talks. - Is not: a program. There is nothing to install or run. If a feature cannot be expressed in Markdown plus native tool configuration, it does not belong here.
- Is not: a speed tool. The honest claim is variance reduction, not velocity: fewer confident wrong turns, a written record of why the code looks like it does, and a chain from every acceptance criterion to the test that decides it. If you want it to be faster on day one, it will disappoint you; if you want it to still make sense in month six, that is the trade.
-
Is not: ceremony for everything. A typo does not get a spec.
AGENTS.mdsays work spec-first for anything that changes behaviour and documents a fast path for the rest — because a workflow that demands three Markdown files for a config value teaches people to stop using it for the changes that needed it.
Start here
Reference
Design & interop
- Interop Matrix
- Design Decisions
- Specify Method
- Committing to One Tool
- Migrating from the predecessor
- Releasing
Help