https://github.com/enzodevs/go-lean-sdd
Go Lean SDD — a lean spec-driven development framework for AI coding agents, packaged as a Claude Code skill
https://github.com/enzodevs/go-lean-sdd
agentic-coding ai-agents claude-code claude-skill developer-tools llm sdd spec-driven-development tdd
Last synced: about 19 hours ago
JSON representation
Go Lean SDD — a lean spec-driven development framework for AI coding agents, packaged as a Claude Code skill
- Host: GitHub
- URL: https://github.com/enzodevs/go-lean-sdd
- Owner: enzodevs
- License: cc-by-4.0
- Created: 2026-06-25T20:42:19.000Z (5 days ago)
- Default Branch: main
- Last Pushed: 2026-06-25T21:10:24.000Z (5 days ago)
- Last Synced: 2026-06-25T22:05:05.134Z (5 days ago)
- Topics: agentic-coding, ai-agents, claude-code, claude-skill, developer-tools, llm, sdd, spec-driven-development, tdd
- Size: 157 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
Go Lean SDD
The leanest path that still ships tested code.
A spec-driven development framework for AI coding agents, packaged as a Claude Code skill.
---
Go Lean SDD turns a request or PRD into working, tested code. It sizes the process to the task instead of running the same ceremony on every change. A one-line fix gets one line of spec. A multi-component feature earns a full spec, a design, and a task breakdown.
## Why it exists
Spec-driven frameworks tend to grow heavy. You get directory trees, status files, and multi-step rituals applied to every change, whether it is a typo fix or a new subsystem.
A benchmark of four popular SDD frameworks pointed the other way. A strong model given nothing but a good `AGENTS.md` scored 0.89, matching the best framework while spending fewer tokens. That result shaped this skill: a capable model does better with clear, tight instructions than with heavy scaffolding. Go Lean SDD is built around that idea.
## What it does
The skill sizes planning to the task. Complexity earns depth, so trivial and medium work stays in the conversation and only a large feature writes anything to disk.
Tests are not optional here. Every change ships with co-located tests, and a discrimination check then mutates the new code to confirm a test actually fails. That is how you tell a test that catches regressions from one that only looks green.
It also keeps the repo clean. There is no scaffolding tree and no status file to go stale, just a single `AGENTS.md` that holds the durable context. Before writing new code, the skill climbs a reuse ladder through your project, your dependencies, the standard library, the framework, and the platform, and it writes something new only when nothing above it fits.
## Install
```sh
git clone https://github.com/enzodevs/go-lean-sdd.git ~/.claude/skills/go-lean-sdd
```
Restart your session. The skill loads on its own when you plan, build, or verify a change, and you can also call it as `/go-lean-sdd`.
## How it works
Four phases, applied only as deep as the task earns:
| Scope | Specify | Design | Tasks | Execute |
|---|---|---|---|---|
| Trivial (3 files or fewer, one sentence) | one line, inline | skip | skip | implement and test |
| Medium (clear feature, under 10 steps) | brief, with edge cases | inline | implicit | implement and test each step |
| Large (multi-component or ambiguous) | full, with requirement IDs | approach and reuse | atomic breakdown | implement, test, and verify each task |
Specify and Execute always run. Design and Tasks switch on only when the work needs them. The full skill lives in [`SKILL.md`](SKILL.md), and the deeper guidance loads on demand from [`references/`](references/).
## Recommended tooling
The skill reaches for two command-line tools when they are present:
- [`code-context-v2`](https://github.com/enzodevs/code-context-v2) (`cc2`) for semantic code search across a project.
- [`xray`](https://github.com/enzodevs/xray) for reading a file's structure and dependencies before editing.
Neither is required. Without them the skill falls back to ordinary search and file reading; with them, the reuse-first checks are faster and more precise.
## Credits
Go Lean SDD distills a public benchmark of spec-driven development frameworks, run by Waldemar Neto, down to the parts that earned their keep: auto-sizing, strict test enforcement, tight scope control with delta specs, early surfacing of unknowns, interface contracts, and the reuse-first build ladder.
The logo features the Go gopher, a mascot created by [Renée French](https://reneefrench.blogspot.com/) and licensed under [CC BY 3.0](https://creativecommons.org/licenses/by/3.0/).
## License
[Creative Commons Attribution 4.0 International](LICENSE) (CC BY 4.0). You can share, adapt, and build on it, including commercially, as long as you credit "Go Lean SDD by rrghost" and link the license.