gotalab/cc-sdd

cc‑sdd – Spec‑driven, long‑running AI coding agents

What it is

• An npm package that installs a set of Agent Skills for eight different AI coding assistants (Claude Code, Codex, Cursor, GitHub Copilot, Devin, OpenCode, Gemini CLI, Antigravity).
• The skills implement a spec‑driven software development lifecycle (SDLC) that lets the AI write a specification, have humans approve the contract, and then autonomously generate, test, review and debug code.
• The workflow is split into a series of commands (/kiro‑discovery, /kiro‑spec‑init, /kiro‑spec‑requirements, /kiro‑spec‑design, /kiro‑spec‑tasks, /kiro‑impl, etc.) that create markdown artefacts (brief.md, requirements.md, design.md, tasks.md) and drive a long‑running implementation loop.

Why it exists

• Traditional AI‑assisted coding often hands a single prompt to the model and expects the output to be the final product. cc‑sdd treats the specification as a contract that separates “what the system should do” from “how it is built”. This makes it possible for multiple agents (or humans) to work in parallel without constantly overwriting each other.
• By attaching a fresh sub‑agent to each task, the tool can run test‑driven development (RED → GREEN) behind a feature flag, perform an independent review, and automatically debug when a review fails. When native sub‑agents are unavailable the same process runs inline.

Key concepts

Concept	Role
Discovery (/kiro‑discovery)	Routes an idea to one of several paths: extend an existing spec, start a new spec, decompose into multiple specs, or implement directly without a spec. Produces brief.md (and roadmap.md when needed).
Spec files (requirements.md, design.md, tasks.md)	Structured markdown that follows the Kiro methodology: EARS‑style requirements, Mermaid diagrams, a File Structure Plan, and task annotations (_Boundary:_, _Depends:_).
Implementation (/kiro‑impl)	Executes each task in isolation. With native sub‑agents it spawns a fresh implementer, an independent reviewer, and an auto‑debug pass if the reviewer rejects the change. Results are recorded in tasks.md so the run can be resumed later.
Spec‑batch (/kiro‑spec‑batch)	Turns a multi‑spec roadmap into parallel spec creation waves, adding cross‑spec review to catch duplicated responsibilities or interface mismatches.
Agent Skills	A collection of 17 commands per agent (e.g., Claude Code, Codex, Cursor). The same skill set works across all supported agents; the default install uses Claude Code.

How to get started

# Install the default (Claude Code) skill set in any project folder
cd my‑project
npx cc‑sdd@latest

# Pick a different agent or language if you prefer
npx cc‑sdd@latest --codex‑skills --lang ja   # Codex, Japanese docs
npx cc‑sdd@latest --cursor‑skills --lang zh‑TW   # Cursor IDE, Traditional Chinese

Then run the first command inside the AI‑assistant you are using:

/kiro-discovery "Photo albums with upload, tagging, and sharing"

The tool will write brief.md and tell you the next command (usually kiro‑spec‑init). Follow the chain of commands until kiro‑impl finishes the implementation.

Supported agents & stability

• Claude Code – stable
• Codex – stable
• Cursor IDE, GitHub Copilot, Devin, OpenCode, Gemini CLI, Antigravity – beta
• Legacy “Cascade/Windsurf” mode is still present but deprecated.

Customization

• All templates (requirements, design, tasks) live under {{KIRO_DIR}}/settings/templates/.
• Rules that guide AI generation (e.g., acceptance‑criteria format, naming conventions) are in {{KIRO_DIR}}/settings/rules/.
• Teams can adapt these to match PRD processes, JIRA integration, domain‑specific standards, etc.

Documentation & guides

• Skill Reference – detailed description of each of the 17 skills and the /kiro‑impl sub‑agent flow.
• Command Reference – usage for the legacy /kiro:* commands.
• Customization Guide – concrete snippets for tailoring templates and rules.
• Spec‑Driven Guide – end‑to‑end walkthrough from idea to deployed code.
• Why cc‑sdd? – design rationale and when the tool is appropriate.

Who should use it

• Development teams that want to experiment with AI‑driven, contract‑first development at scale.
• Solo developers looking for a structured, repeatable way to let an LLM handle the heavy lifting while keeping a clear spec‑to‑code trace.
• Organizations that already use the Kiro spec methodology and want a CLI‑based bridge to multiple coding agents.

License

• MIT (see LICENSE).

---

All information above is taken directly from the repository’s README; no additional features have been inferred.