Spekkr vs GitHub Spec Kit

They do different jobs, and you can use both. A lossless codec moves a spec between them.

9 minutes to read. Spec Kit checked 18 November 2026. Codec and catalogue checked 20 November 2026.

Looking for a GitHub Spec Kit alternative? Spekkr is not a drop-in replacement. Spec Kit plans a feature inside your repository. Spekkr hosts a finished spec any agent can fetch. Use both, and move the folder with dwrtz/spec-kit-codec. Spec Kit docs

At a glance

Spekkr and GitHub Spec Kit compared.
TopicSpekkrGitHub Spec Kit
What it isA catalogue of finished specs an agent can fetch. Each spec is a folder: requirements, design, tasks, and a manifest.A toolkit for spec-driven development with a coding agent.
Who makes itSpekkr makes it.GitHub publishes it, in the github/spec-kit repository.
Where specs liveOn spekkr.com at username/name. The page says who published it.specs/<directory>/ in the repo, plus .specify/feature.json.
How an agent gets onePaste the spec's line, or pull_spec. Search, get and pull need no account.specify init, then /speckit.plan in chat. Copilot's form is /speckit-plan. Command reference.
VersioningThe docs: a published version never changes. Raise version, add a changelog entry, and publish again.Files stay in the repo. Spec Kit does not prescribe how a team keeps them after requirements change.
Sharingpublish_spec lists the spec under the signed-in username. Any agent can fetch that address.A catalog of extensions, presets, workflows and bundles. Not feature specs.
Project rulesNo project constitution. The manifest says what a person may change, and which commands and secrets the build uses./speckit.constitution writes .specify/memory/constitution.md. Later commands read it.
Planning against your codeA spec is not produced from your repository. pull_spec records the choices the manifest declares./speckit.plan holds the stack. On an existing repo, the existing-project guide says to reuse its architecture, dependencies and tests.
Licence and costPricing: reading, pulling and publishing cost nothing in this version. Publishing needs an emailed sign-in code. No account is needed to build a spec.The license is MIT, copyright GitHub, Inc. The CLI needs Python 3.11 or newer.
Using them togetherPull a shared spec first. to-spekkr writes the Spekkr folder from a feature directory.dwrtz/spec-kit-codec moves the folder both ways. --activate writes .specify/feature.json. Plan setup keeps plan.md.

Eleven stages

Choose Spec Kit, Spekkr, or both. A stage neither covers stays visible.

Show stages covered by
  1. 1 Find a spec someone already wrote

    Spekkr

    Search on Spekkr, or search_specs. Spec Kit's catalog is extensions and presets, not a feature spec another project wrote.

  2. 2 Choose options

    Spekkr

    pull_spec records the choices the spec declares.

  3. 3 Set project-wide rules

    Spec Kit

    /speckit.constitution writes .specify/memory/constitution.md. Spekkr has none.

  4. 4 Write or adapt the spec

    Both

    dwrtz/spec-kit-codec copies spec.md byte for byte. Do not run /speckit.specify on that folder.

  5. 5 Plan against your codebase

    Spec Kit

    --activate writes .specify/feature.json. Plan setup keeps plan.md.

  6. 6 Break into tasks

    Both

    The codec copies tasks.md. Spec Kit's tasks command generates a new one.

  7. 7 Build

    Both

    /speckit.implement reads plan.md, where the codec puts the safety note.

  8. 8 Acceptance checks

    Both

    /speckit.checklist and /speckit.analyze are optional. juno/waitlist ends requirements with "Accept when".

  9. 9 Run the checks for you

    Neither

    /speckit.converge can append tasks. It does not run the checks, and Spekkr does not either.

  10. 10 Share across projects

    Spekkr

    publish_spec lists it. to-spekkr writes the Spekkr folder first.

  11. 11 Reuse a published version

    Spekkr

    A published version is not edited in place. Spec Kit does not prescribe how a team keeps the files.

Find a spec Your intent Choose options Spec Kit: specify Your repo Plan + tasks Agent builds Acceptance checks Share back
A Spekkr spec and your intent enter the repo. The codec writes the feature folder, Spec Kit plans, the agent builds, and to-spekkr can publish the result.

Where Spec Kit is stronger

Spec Kit is the stronger tool for work that stays in one repository.

The process

Specify, plan, tasks, implement, converge. Clarify, checklist and analyze are optional, and only specify is required before plan. /speckit.converge can append tasks. Bug fixing and idea assessment are extensions. The quickstart and the command reference describe this.

An existing repository

specify init --here --force --integration <key> adds .specify/ and the agent's files. --force may replace paths Spec Kit manages. It does not delete the app, and it does not infer specs for behavior already there. The existing-project guide says to commit or stash first.

Install

Python 3.11 or newer: uv tool install specify-cli. The integrations page lists 41 integrations, plus generic and none. The installation guide is the install page. Specify can also keep checklists/requirements.md.

Where Spekkr fits

On 20 November 2026 the catalogue listed 138 specs.

Fetch and publish

search_specs, get_spec, pull_spec. A missing choice stays at the default. An unknown key is refused. check_spec saves nothing. publish_spec lists the spec. Sign-in is an emailed code. The docs describes the tools.

juno/waitlist is a waitlist. vex/feature-flags is feature flags. Both are listed as app parts.

The plan for this repository stays in Spec Kit. A gist has no search, no declared choices and no MCP fetch.

Use them together

Pull, then convert

Pull a shared feature such as juno/waitlist (the call is below), move the folder with dwrtz/spec-kit-codec, then plan it with Spec Kit. No Spec Kit yet? Commit first, then specify init --here. The existing-project guide says so.

Python 3.11 or newer, standard library only. The GitHub repository holds the script. A pull includes reference/spec_kit_codec.py and reference/SKILL.md. From the directory that contains .specify/:

python3 spec_kit_codec.py to-speckit ./waitlist --activate

No --dir: to-speckit writes specs/<name>/, the id after the slash, so this pull is specs/waitlist/. --dir accepts specs/NNN-<name>. --activate writes .specify/feature.json as Spec Kit 1.4 reads it, for example {"feature_directory":"specs/pebble-jar"}. That file is outside the feature folder. Without it, The setup script reports "Feature directory not found." An older feature.json still names its directory.

File mapping

Spekkr folder to Spec Kit feature folder.
SpekkrSpec Kit
spec.md, tasks.md, and every other filethe same path, byte for byte
design.mdplan.md, with a fixed safety note at the top
(none)design.md, a short stub that points at plan.md
spekkr.json, spekkr.inputs.json, BUILD.mdthe same names, inside the feature folder

The safety note is the first bytes of plan.md. /speckit.implement reads it: show commands and secrets under safety, and ask before anything not listed. Do not run /speckit.specify. It replaces spec.md from the template.

Spec Kit's plan setup keeps plan.md, and the codec says so. Spec Kit's tasks command generates tasks.md. The codec does not say that keeps the copy: leave it and run /speckit.implement. Use /speckit.tasks for a new list. /speckit.converge checks the result. Copilot uses /speckit-plan.

Back to Spekkr

Run to-spekkr. The safety note comes off plan.md only when it is the codec's note, and the stub design.md is dropped only when it matches. An existing spekkr.json is restored unchanged. --author, --date and --platforms do not rebuild it.

A hand-written feature needs --author, written to author.id and author.name. No --date means changelog date 1970-01-01. Read the note: lines, edit the summary, then check_spec and publish_spec. The skill uses this shape:

python3 spec_kit_codec.py to-spekkr specs/003-widget --dir ./widget --author USERNAME

What lossless means

Byte for byte when spekkr.json already exists. A hand-written feature gets a stable manifest: the same folder and flags always produce the same file. The conversion is local. .specify/feature.json is outside the round trip. dwrtz/spec-kit-codec and the codec README are the sources.

The constitution stays

The codec does not write the constitution, and the plan does not replace the spec. Spec Kit keeps those as different commands.

Nearby tools

Kiro

Kiro writes requirements.md, design.md and tasks.md under .kiro/specs/, and can write a bugfix spec. Kiro's specs guide says sharing is through that repository.

OpenSpec

OpenSpec keeps each change under openspec/changes/<name>/: proposal.md, specs/, design.md, tasks.md. MIT licensed. Stores, marked beta, keep that shape in a separate repository.

Tessl

Tessl describes its registry as a searchable index of skills, for an organization and a private workspace. Publishing and installing skills is free. It is an index of agent skills, not feature specs.

BMad

BMad Method adds skills to a coding agent. bmad-build implements a change. A larger one can start with a product owner or a PRD. Stable v6.12, via npx bmad-method install (Node.js 20.12+, Python 3.10+). v7 is in preview. The BMAD-METHOD repository is MIT.

Pick in 30 seconds

Three questions. The recommendation is Spec Kit, Spekkr, or both. Every combination is on this page, and the result shows with no JavaScript.

Is this a feature many apps share (sign-in, a waitlist, a changelog)?
Does it have to fit an existing codebase or team rules?
Will anyone else want to reuse it?

Both. Pull it, convert with dwrtz/spec-kit-codec, plan with Spec Kit, and publish the result.

Both. Pull it, convert with dwrtz/spec-kit-codec, then plan it against your code.

Spekkr. Pull it, build it, and publish improvements.

Spekkr. Someone has probably specced it. Pull and build.

Both. Write it with Spec Kit, convert with dwrtz/spec-kit-codec, and publish it.

Spec Kit. It's yours alone and must fit your code.

Spekkr. Write a spec and publish it.

Spec Kit. Or a plain spec.md. Either way, write the checks first.

FAQ

Is Spekkr a Spec Kit alternative?

No. Spec Kit plans a feature in your repository. Spekkr hosts a spec any agent can fetch. dwrtz/spec-kit-codec moves the folder both ways.

Spekkr vs Spec Kit: which should I use?

Spec Kit, when it has to fit this repository. Spekkr, when a shared spec is already listed. Both, when you need the plan and the address.

Can I use both?

Yes. From the directory that contains .specify/, run python3 spec_kit_codec.py to-speckit SPEKKR_FOLDER --activate.

Do not run /speckit.specify on that folder. Leave the pulled tasks.md. The codec does not say /speckit.tasks keeps it.

Can I use Spec Kit specs with Spekkr?

Yes. to-spekkr copies an existing spekkr.json unchanged. A hand-written feature needs --author. With no --date the changelog date is 1970-01-01.

Read the note: lines, then check_spec and publish_spec.

Is the conversion lossless?

Yes, when spekkr.json is already there: byte for byte. A hand-written feature gets a stable generated manifest.

.specify/feature.json sits outside the round trip. The conversion is local.

Does Spekkr replace Spec Kit's constitution?

No. Spec Kit writes .specify/memory/constitution.md. dwrtz/spec-kit-codec does not.

Do I need an account to pull?

No. Publishing needs an emailed sign-in code. Search, get and pull do not.

Pull it

Connect https://spekkr.com/mcp in your agent, pull juno/waitlist into your repo, convert the folder with dwrtz/spec-kit-codec (the script on GitHub), then run your Spec Kit plan step on it. Search, get and pull need no account. Questions: support@spekkr.com.

pull_spec {"id": "juno/waitlist", "choices": {"referrals": false, "database": "sqlite"}}

python3 spec_kit_codec.py to-speckit ./waitlist --activate

Publishing a spec of your own needs sign-in. How to publish.