Skip to main content

SpecWeave vs GitHub Spec Kit

Both tools are built on the same idea: an AI coding agent does better work from a written specification than from a chat message. They differ in what happens after the spec is written.

GitHub released Spec Kit in 2025. It gives an agent a set of slash commands that walk a feature from principles to code. SpecWeave 3.0 keeps each piece of work in one spec file with an append-only task ledger, so the work can be claimed, proved, handed off to another tool and closed.

This page describes Spec Kit as its public repository documents it. Check the repository for its current state.

At a glance​

Spec KitSpecWeave 3.0
Unit of workA feature folder under specs/An increment in .specweave/increments/NNNN-slug/
Files per featurespec.md, plan.md, tasks.md, plus supporting files such as research and contractsOne spec.md (Problem, Scope, Acceptance Criteria, Approach, Open questions, Tasks); plan.md optional
Workflow/speckit.constitution, /speckit.specify, /speckit.plan, /speckit.tasks, /speckit.implement, with optional clarify and analyze stepsspecweave create-increment, then task claim and task done, verify, complete
Task progressCheckboxes in tasks.mdAppend-only ledger.jsonl; nothing written back into markdown
Proof a task is doneUp to the agenttask done --run "<test>" stores the real output; a failing command is refused
ClosingNo closure stepcomplete requires a passing specweave verify (tests, lint, build, every AC met) or an explicit reason
Several agents at onceNot addressedTask claims with a lease, file-overlap checks, merge=union on the ledger
Switching tool or account mid-taskNot addressedspecweave handoff and specweave pickup
Instruction filePer-agent command filesOne AGENTS.md, imported by CLAUDE.md
Issue trackersNoneOptional: GitHub, Jira, Azure DevOps, only on specweave sync push
RuntimePython CLI (specify) that scaffolds the templatesNode.js CLI (specweave) that also runs the loop

Where Spec Kit is strong​

  • A guided path from idea to plan. The constitution, specify, clarify and plan steps push you to think about principles and unknowns before any code. SpecWeave's spec.md asks for the same things in fewer sections, and its brainstorm skill covers the early exploration, but Spec Kit makes the stages more explicit.
  • Richer design artifacts. Spec Kit plans can include research notes, data models and API contracts as separate files. SpecWeave keeps the design to an Approach section and one optional plan.md.
  • Nothing to run after scaffolding. Once the files exist, the agent works from them with no extra CLI in the loop.

Where SpecWeave is different​

  • One file to read. An agent resuming work reads one spec.md and runs specweave pickup, not three or more files.
  • State that survives more than one agent. Claims, completions and evidence live in an append-only ledger in git. Two agents cannot both take the same task, a stale claim can be taken over, and merges keep both sides.
  • Evidence instead of checkboxes. A task is done when its test command exited 0 and the output was stored. An acceptance criterion is met when the tasks covering it are done; nobody ticks boxes.
  • Handoff across tools and accounts. When one tool runs out of tokens, say "hand off": specweave handoff pushes everything the next session needs through your git remote, and saying "pick up" in any other tool or account continues from there. See Switch from Claude Code to Codex without losing your place.
  • A closing gate. specweave complete refuses to close work whose verification failed, unless you record why.

When to use which​

Spec Kit fits when one developer and one agent are building a feature in one sitting, and you want a structured path through the design first.

SpecWeave fits when work spans sessions, tools or people: you switch between Claude Code and Codex, run parallel threads, need proof that tests ran, or want issues in GitHub or Jira to follow the work.

Nothing stops you from using both. A Spec Kit spec.md and plan.md are good input for a SpecWeave increment: copy the requirements into Acceptance Criteria, the design into Approach, and the task list into the Tasks section as ### T-01 Title entries.

See also​