Front page / Concepts / Spec-driven work
Spec-driven work
Write down what you want before anything gets built: the goal, the constraints, the acceptance criteria, the open questions. Then the assistant plans against the spec, you approve the plan, and it builds to it. It is the discipline that turns a chat into a project, and it applies to a deck or a research report as much as to software.
Where it stands
Among the 14 concepts. Bars: GitHub stars gained this week.
- MCP and tool use+3.9k
- Agentic loops+1.3k
- Knowledge graphs (GraphRAG)+1.2k
- Retrieval (RAG)+1.2k
- Prompt engineering+1.1k
- Vibe coding+980
- Workflows, not agents+565
- Spec-driven work+16k
- Memory+7.3k
- Agent skills+5.7k
- Multi-agent orchestration+5.6k
- Computer use+4.7k
- Context engineering+3.3k
- Evals and verification+929
How to use it today
In ChatGPT or Claude, as a person doing the work, not as a developer.
- 1Before a sizeable task, write a one-page spec: outcome, audience, constraints, what done looks like, what is out of scope.
- 2Ask the assistant to turn the spec into a plan with numbered steps and to list the questions it has; answer them before it starts.
- 3In Claude Code use plan mode (or ask for a plan and approval before changes); in a chat app, ask for the outline first and approve it.
- 4Keep the spec open while reviewing the result; check the acceptance criteria one by one rather than reading for feel.
- 5When the result is wrong, change the spec and rerun rather than patching the output by hand.
What changes day to day
The first thirty minutes of a task become writing, not prompting. Rework drops, because disagreements surface in the plan rather than in the finished thing.
Where it goes wrong
A spec that lists steps instead of outcomes removes the assistant's ability to find a better route.
Over-specified small tasks cost more than doing them.
Listings that use it well
Chosen by hand after reading them, each with the reason. Install from the listing's own page.
| Listing | Kind | Why it is here | Stars |
|---|---|---|---|
| superpowers | Agent skills | brainstorming, writing-plans then executing-plans: design first, plan from the spec, then build. | 315k |
| skills | Agent skills | grill-me interviews until the spec is shared understanding; implement-spec builds from it. | 319k |
| agent-skills | Agent skills | spec-driven-development creates the spec before any code when none exists. | 114k |
| engineering | Agent skills | spec-driven-workflow defines acceptance criteria first and generates tests from the spec. | 29k |
| ccpm | Agent skills | PRD to epic to GitHub issues to parallel agents. | 8.4k |
| spec-workflow-mcp | MCP server | An MCP server with spec workflow tools and a dashboard for requirements, design and tasks. | 4.3k |
Elsewhere
Repositories that embody the idea and are not in the catalog.
- github/spec-kit ↗GitHub's toolkit: specify, plan, tasks, implement, with templates for most coding agents.
- Fission-AI/OpenSpec ↗Delta-spec workflow (propose, apply, archive) for AI coding assistants, built for existing codebases.
Standing
Our judgment, with the reason and the sources it rests on.
Rising since 2025-07. Plan modes and spec kits appeared across the coding assistants in 2025 as the answer to work that drifted; the pattern is spreading to non-code work. AWS's Kiro launched in July 2025 with spec-driven development as its headline, and GitHub open-sourced Spec Kit in September 2025 for Copilot, Claude Code and Gemini CLI; the vendors present it as the step you add to describing what you want when the result has to be relied on. Adopted faster each month; the vendors are building for it and new tools keep appearing.
- Introducing Kiro2025-07-14
- Spec-driven development with AI: Get started with a new open source toolkit2025-09-02
Related: Evals and verification, Agentic loops, Workflows, not agents