GitHub Spec Kit · slash commands
/speckit.specify/speckit.plan/speckit.tasks/speckit.implement/speckit.converge
Optional add-ons: clarify, analyze, checklist. Nothing requires the agent to run them.
Spec Kit adds a slash-command workflow to your agent and trusts you to follow it. Pulse runs the same workflow as a governed system: gates that block bad transitions, a memory that outlives the session.
Both apply Spec-Driven Development. They diverge on what’s optional and what’s enforced.
/speckit.specify/speckit.plan/speckit.tasks/speckit.implement/speckit.converge
Optional add-ons: clarify, analyze, checklist. Nothing requires the agent to run them.
Each step is a gate, not a suggestion. The next one doesn’t open until the one before it actually clears.
The same four numbers gate every task on a Pulse board, configured once and enforced every time.
/speckit.checklist can check similar things by hand, but nothing runs it automatically or blocks a merge if it’s skipped.
These four numbers are configured once when a board is set up, then Pulse enforces them on every task without anyone re-checking.
How each capability actually behaves, not just whether it exists.
| Capability | Okto Pulse | GitHub Spec Kit |
|---|---|---|
| COVERAGE & TRACEABILITY | ||
| SDLC coverage | Native Ideation → Sprint → Tasks/Tests/Bugs, one pipeline. |
Partial Spec → plan → tasks; no ideation intake or sprint layer. |
| Brownfield support | Native A Refinement stage forces investigation of existing code first. |
Partial/speckit.converge diffs code against spec on demand. |
| Traceability | Native Requirement → task → test → evidence, as queryable graph edges. |
Partial Markdown cross-references, checked on demand via /speckit.analyze. |
| GOVERNANCE & VALIDATION | ||
| Quality gates | Native 17 gates block a status transition until evidence clears. |
Partial Checklist/analyze commands exist, but are opt-in. |
| Governance | Native Locked specs, role-aware permissions, an audit trail via Okto Nexus. |
Partial Outsourced to git branch protection and PR review. |
| Enterprise readiness | Native Gates, audit trail, and role permissions enforced today. |
Partial Controls come from surrounding tooling, not Spec Kit itself. |
| OPERATIONS & EXTENSIBILITY | ||
| Developer experience | Native A real local board, not just a folder of files. |
Partial Pure CLI, no surface to inspect the work. |
| Extensibility | Native 200+ MCP tools, plus Okto Nexus for multi-agent coordination. |
Partial A well-built Extensions/Presets catalog, no tool surface of its own. |
| Deployment | Local-first: SQLite + embedded graph DB, Elastic License 2.0. | Git-native: markdown in your repo, MIT. |
MCP tool counts vary by release for both projects, rounded down on purpose. Rows reflect each project’s public docs and source as of August 2026.
…a small team, a single repo, a short-lived project, where a reviewed PR is real proof of done.
…more than one agent touches the work, the build outlives a sprint, or “the agent said it’s finished” stopped being good enough.
A Builder agent finishes a task and marks it done. Neither tool takes that claim the same way.
Nothing catches the gap. It surfaces later, if at all.
The task can’t reach done until the validator’s checks actually pass. A real rejection, not a hypothetical.
Both agents ran the same task. Spec Kit’s checklist step existed and was skippable; Pulse’s validator gate isn’t a step someone remembers to run, it’s the only path a status transition can take.
Spec Kit leaves traceability to your markdown conventions. In Pulse it’s a queryable graph.
Every one of these is a real relationship in the graph, not a comment or a filename convention. Queryable months after the session that created it.
Conventions get recorded as Decisions before new work starts, not rediscovered later.
The agent that builds a task can’t validate it. Enforced by identity.
A card can’t move to done without verified evidence attached first.
Shipped code traces back to its task, test, and originating spec.
Spec Kit’s equivalent lives in whichever markdown file someone thought to update. Pulse’s lives in the graph whether anyone remembers to look or not.
One install boots the whole runtime; every Spec Kit step maps to a governed stage.
# installs the CLI, boots the board + MCP server
pip install okto-pulse
okto-pulse init
okto-pulse serve
→ Web UI on port :8100, MCP server on port :8101/mcp. Connect two separate agent identities, Executor and Validator, each with its own credentials.
| Spec Kit command | Pulse equivalent |
|---|---|
/speckit.constitution | Board-wide governance settings and guidelines, configured once |
/speckit.specify | Ideation → Refinement |
/speckit.plan | Spec, gated on approval |
/speckit.tasks | Sprint & Tasks, opened once Spec is approved |
/speckit.implement | Builder executes a task |
/speckit.checklist, /speckit.analyze | Validator gate, runs automatically |
/speckit.converge | Refinement stage, on every brownfield board |
The commands are a rough map, not a 1:1 port. Spec Kit produces markdown files; Pulse produces board records with their own gates. The destination is a governed board, not a bigger specs/ folder.
Neither tool has its own chat UI. Both plug into the coding agent you already run.
Connects through its own CLI and per-agent templates. Each supported agent needs Spec Kit’s command set installed into it directly.
Connects over MCP: one server, and every agent above (plus any future one that speaks the protocol) queries the same board with zero extra setup on Pulse’s side.
Claude Code and Cursor work with both today. Spec Kit’s “30+” is a list of agents its maintainers have explicitly built a template for. Pulse’s isn’t a list at all. It’s the entire population of MCP-compatible agents, today and whatever ships next, with nothing for Pulse to build per agent.
Functionally, a replacement. Spec Kit gets a feature to a markdown spec. Pulse takes it further: enforced gates, a shared board, and a memory that survives the session.
Yes: Claude Code, Cursor, Windsurf, Cline, and any other MCP-compatible agent connect the same way they’d connect to Spec Kit.
Different license, same spirit of “free to run.” Spec Kit is MIT. Pulse is source-available under the Elastic License 2.0.
Spec Kit writes markdown and needs nothing else. Pulse runs a real local service: SQLite plus an embedded graph database, via pip install or Docker.
Spec Kit has less to install. Pulse asks for three extra commands and hands back a board, enforced gates, and a Knowledge Graph the moment it’s running.
Free to run locally, no account required, works with the coding agent you already have.
“GitHub” and “Spec Kit” are trademarks of GitHub, Inc. References to GitHub Spec Kit on this page are for comparison purposes only, based on its public documentation and source code, and do not imply any affiliation with or endorsement by GitHub.