Spec Kit writes the documents.DotMD is where you review them.

What GitHub Spec Kit is

Spec Kit is GitHub's open-source toolkit for spec-driven development. It does not replace your agent. It adds a command line tool, specify, that sets up your project, and a set of commands you run in your agent's chat. The agent does the writing; Spec Kit gives it the steps and the structure.

It supports many agents, including Claude Code, Codex CLI and Cursor. The full list is on Spec Kit's integrations page.

Install and initialise

You need Python 3.11 or newer and uv. Then:

uv tool install specify-cli
specify init my-project --integration claude

The --integration flag takes the key of your agent: claude for Claude Code, codex for Codex CLI, cursor-agent for Cursor. Spec Kit's own page lists the key for every other agent.

The commands, in order

You type these in the agent's chat, not in the terminal:

1. /speckit-constitution
2. /speckit-specify: writes spec.md.
3. /speckit-plan: writes plan.md.
4. /speckit-tasks: writes tasks.md.
5. /speckit-implement: builds from the tasks.
6. /speckit-converge

The spelling of a command differs by agent: some use /speckit.specify, others a hyphen or a $. Spec Kit explains the forms for each agent in its command invocation notes.

Review each step in DotMD

Spec Kit asks you to review each result before you continue. DotMD makes that a proper review instead of a skim in the terminal.

1. Open the project

   Press ⌘O and choose the project folder. The files Spec Kit writes appear in the sidebar as they are created, rendered with their headings, lists and tables.

2. Comment on the exact line

   Read spec.md, select a requirement that is wrong or vague and press ⌥⌘M. Write what to change. Comments are kept by DotMD, not in the file.

3. Have the agent fix them

   Tell the agent: "Go through my comments in DotMD." It reads them over MCP and edits the file. Each fix is shown next to its comment, the old text red and the new green. Accept or reopen each one.

4. Only then run the next command

   When the spec is right, run /speckit-plan, and repeat for the plan and the tasks. An error caught in the spec costs one comment; caught in the code, it costs a rework.

Checking tasks.md

The task list is a Markdown checklist, and DotMD renders task lists with their boxes, so you can see at a glance what is done and what is left. Before you run /speckit-implement, read the tasks as the agent will: in order, each one small, none missing. If you ran Spec Kit's analysis step, treat its findings the same way: open them in DotMD and comment on any you disagree with.

Connecting the agent to DotMD takes one line; the guides for Claude Code, Codex and Cursor show it.

Why review each file

Spec Kit moves fast: one command, and a whole document appears. That speed is the risk. If the spec has a wrong requirement and you run the next command without reading it, the plan and the tasks are built on it, and so is the code. Reading spec.md carefully once is the cheapest check in the whole process.

Questions about Spec Kit

More questions