FeatherSpec workflow
FeatherSpec drives spec-driven development in Claude Code and GitHub Copilot: 13 /sdd commands and a Markdown project memory
- 70 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The template is instructions and commands in Markdown, but by them the agent writes and changes project code and files in `.specs` and `.memory-bank`
Install
- Works in
Manual install
npx degit GregorBiswanger/featherspec my-projectDownloads the template files with no git history. Then open the project in Claude Code or Copilot and run /sdd-setup. For an existing repository copy the files by hand.
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The template is instructions and commands in Markdown, but by them the agent writes and changes project code and files in `.specs` and `.memory-bank`
This is third-party code. Review the repository files before installing.
What it does
FeatherSpec is a featherweight spec-driven development template with no CLI, install or dependencies: only Markdown and folders. You describe what you want, the agent turns it into a spec with testable acceptance criteria, then a plan of baby steps, and only then writes code. Specs, plans, decisions and progress live in the repository, so the next session, teammate or tool picks up where you stopped. Thirteen `/sdd-*` commands work identically in Claude Code and GitHub Copilot because both execute the same files: `/sdd-setup`, `/sdd-specify`, `/sdd-clarify`, `/sdd-plan`, `/sdd-compile`, `/sdd-lifecycle`, `/sdd-overview`, `/sdd-clean`, `/sdd-architecture-scan`, `/sdd-architecture-update`, `/sdd-reverse-specify`, `/sdd-style-update`, `/sdd-featherspec-update`. Rules live in `AGENTS.md`, which `CLAUDE.md` imports, and scout subagents in `.claude/agents` and `.github/agents` read code in isolation. The first `/sdd-setup` asks the documentation language, project description, quality rules after each step and TDD mode, then where tickets live: GitHub, Jira, another tracker or nowhere. For an existing codebase the wizard runs a deep architecture scan and distills it into a lean fingerprint, and `/sdd-reverse-specify` recovers the behavior of code already written. Only a constitution of about 200 lines is always in context, while module maps of up to 40 lines load when a module is touched.
Who it is for. Developers and teams who mix Claude Code and GitHub Copilot and want a shared order of specs, plans and project memory right in the repository.
Good fit when
- A new feature should go from a spec with acceptance criteria to a plan of baby steps and only then to code
- A colleague opens the project in Copilot while you use Claude Code and both should see the same `/sdd-*` commands
- An existing service has to be described: `/sdd-architecture-scan` builds a fingerprint and `/sdd-reverse-specify` recovers behavior
- Specs should become Jira or GitHub Issues tickets and statuses should become transitions
- A new agent session should continue from the Memory Bank without a recap of yesterday's chat
Not a fit when
- The task is a small one-off: the spec and plan cycle is overkill
- You need a CLI or a versioned package: FeatherSpec is only files copied into a project
- You work in Cursor or Codex: the template targets Claude Code and GitHub Copilot in VS Code
- Copying ten folders and files into a large repository is not allowed by team rules
Example request
/sdd-specify draft an invoice CSV export feature with a spec and acceptance criteria, then /sdd-planLimitations
- It is a template of files: for a new project take it with Use this template or `npx degit GregorBiswanger/featherspec my-project`, for an existing one copy `AGENTS.md`, `CLAUDE.md`, `.claude/`, `.github/prompts/`, `.github/instructions/`, `.github/agents/`, `.specs/`, `.memory-bank/` and `.vscode/settings.json`.
- In Copilot open the project folder itself in Agent mode or the slash commands will not appear.
- Tracker links use the access the tool already has (`gh`, the GitHub or Atlassian MCP server), and the template carries no tracker knowledge of its own.
- A deep scan of a large repository is costly: its cost is tuned through the worklist.
How to disable. Delete the template files from the repository: `AGENTS.md`, `CLAUDE.md`, `.claude/`, `.github/prompts/`, `.github/instructions/`, `.github/agents/`, `.specs/` and `.memory-bank/`. There is no install or background process, so nothing of FeatherSpec remains after deleting the files.
FAQ
How do I start with FeatherSpec?
Take the template with Use this template or `npx degit GregorBiswanger/featherspec my-project`, open the project in Claude Code or Copilot and run `/sdd-setup`.
Does anything need installing?
No. There is no CLI or dependency, only Markdown and folders. Setup means copying files and running the `/sdd-setup` wizard once.
How do I adopt an existing project?
Copy the template files into the repository and run `/sdd-setup`, choosing existing software. The wizard offers a deep architecture scan and `/sdd-reverse-specify` recovers specs from code and tests.
Does the template work with Jira?
Yes, if you name the tracker in the wizard: a finished spec can become a ticket and each status move gives the matching transition through access your tool already has. A solo developer answers `none`.
How much context does the documentation take?
Only a constitution of about 200 lines is always loaded, and module maps of up to 40 lines load when the agent touches that module.
Discussion
No questions or reviews yet.
Related
WorkflowsTwo skills for spec-driven development: /spec asks questions and saves a document to specs/, /spec-impl implements an approved spec step by step
A CLI that keeps the spec, tasks and proof of done in git so Claude, Codex, Grok and Cursor can hand work to each other
A skill that interviews you and writes requirements.md, design.md and tasks.md so Claude Code and Cursor work from one plan
A plugin with three skills for large development work: project analysis, phased tasks tracked as GitHub Issues, parallel agents and code review
A terminal environment and CLI for spec-driven development: Gherkin, a task board and several agents, written in Rust.
GitHub toolkit for spec-driven development: the specify CLI adds agent commands and skills to a project, from principles to implementation
README in short
The README describes a spec, plan and code loop and promises work with no install. Three start steps, a section on adopting an existing codebase with architecture scan and reverse spec, the `/sdd-*` command list and tracker integration follow.