SDD Skills

★ Star on GitHub

Install

Claude Code plugin

claude plugin marketplace add georgeracu/sdd-skills
claude plugin install sdd@georgeracu

Recommended for Claude Code. The skills install as /sdd:spec-workflow, /sdd:setup and so on.

npx skills add

npx skills@latest add georgeracu/sdd-skills

For other agents, such as Codex or Cursor. This route installs each skill under its bare name, such as spec-generator, without the sdd: prefix that the skills use to call each other.

Manual copy

git clone https://github.com/georgeracu/sdd-skills.git
mkdir -p .claude/skills/sdd
cp -R sdd-skills/skills/. .claude/skills/sdd/

Copy into .agents/skills/sdd/ instead for agents that read that layout. Keep the sdd folder, because it gives the skills their prefix.

Choose one route per agent. Two routes install every skill twice.

Then run /sdd:setup once in your project. It copies the work-item scaffold and records your build and test commands in AGENTS.md.

The pipeline

Seven skills take a feature from an agreed problem to an audited implementation. spec-workflow is the map of the steps, and setup runs once per project. Every sdd: step writes a file into the work item, project/work-items/{NN}-{name}/, so the record of what was agreed, reviewed and verified stays in the repository. The three gates block the next step: the design and task reviews on a CRITICAL finding, and the implementation audit on a CRITICAL or MAJOR finding.

1. 1
   sdd:spec-discovery

   discovery.md in the work item, after user approval

2. 2
   sdd:spec-generator Phase 1

   requirements.md

3. 3
   sdd:spec-generator Phase 2

   design.md

4. 4
   sdd:spec-design-reviewGate

   design-review-summary.md

5. 5
   sdd:spec-generator Phase 3

   tasks.md

6. 6
   sdd:spec-task-reviewGate

   task-review-summary.md

7. 7
   sdd:spec-qa-review plan mode

   test-plan.md

8. 8
   subagent-driven-development (superpowers)

   Implementation

9. 9
   sdd:spec-implementation-auditGate

   implementation-audit.md

10. 10
    sdd:spec-qa-review review mode

    Updated test-plan.md

11. 11
    sdd:spec-maintenance

    Updated previous specs, spec-maintenance-report.md

12. 12
    finishing-a-development-branch (superpowers)

    PR

Steps 8 and 12 hand off to the superpowers plugin. A bugfix runs a shorter path with no gates: spec-generator writes bugfix.md, design.md and tasks.md, and implementation follows. The spec-workflow page has both tables and the rules that override superpowers’ default order.

Who it’s for

Teams that need to show why the code does what it does. Each work item keeps its requirements in EARS notation, the design, the review verdicts, the test plan and an implementation audit backed by pasted build and test output. A reviewer, an auditor or the next engineer can read the decision trail without asking the agent that wrote it.

Teams that want a gate to stop the agent, not only to advise it. The design review, the task review and the implementation audit each return a blocking status, and a second blocking run hands the decision back to a person.

It is a poor fit for a quick prototype. The pipeline asks for your approval at every phase, and that costs time you may not want to spend on throwaway code.

How it compares

All four tools write a spec before code. The differences are in what can stop the work and what happens after the code is written.

	SDD Skills	Spec Kit	Kiro	OpenSpec
Form	Agent skills (SKILL.md)	CLI and agent commands	IDE, CLI and web app	CLI and agent commands
Spec format	EARS requirements, design, tasks	Spec, plan, tasks, checklists	EARS requirements, design, tasks	Proposal, design, tasks, spec deltas
Blocking review gates	Design review and task review block on CRITICAL	plan stops on failed constitution checks; analyze only reports	You approve each phase; no automated review gate confirmed	Validation blocks malformed changes; verify does not block
Audit after coding	Implementation audit, FAILED on CRITICAL or MAJOR	converge adds the gaps as new tasks	Optional property-based tests	verify grades findings, does not block
Older specs	spec-maintenance annotates superseded specs	By team convention	Sync Files updates tasks.md	Deltas merge into the live specs
Licence	MIT	MIT	Proprietary	MIT

Choose Spec Kit or OpenSpec if you want a CLI that works across many agents with less ceremony. Choose Kiro if you want specs built into the editor. Choose these skills if you want review verdicts and an audit kept as files, and gates that stop the agent. They need superpowers for implementation today, and they are tested mainly with Claude Code.

Read the series

The series explains why each skill works the way it does. The posts describe the skills as they were when written, and each skill page here is current.

1. Beyond Kiro: Engineering Quality Gates for AI Agents, 10 May 2026
2. Inside my Spec Generator: From a Half-Baked Brainstorm to Full Specs, 7 September 2026
3. How I Review Specs Against the Codebase, 16 September 2026
4. Inside Spec Task Review, 27 September 2026
5. Inside Spec QA Review, 1 October 2026
6. Inside Spec Implementation Audit, 9 October 2026
7. How I Maintain My Old Specs When They Are Superseded, 17 October 2026
8. Inside Spec Workflow, 20 October 2026

Follow the work

New posts cover the skills as they change and the experiments behind each change. Subscribe on Substack to get them by email, or follow the changelog feed.