Specnaut

9 agents · 6 harnesses · 1 curl · v7.4.0

A coordinated AI agent team, in your project.

Specnaut scaffolds the files your AI coding harness consumes — slash-commands, a plan template, agents, and a backlog system — across Claude Code, Cursor, Codex CLI, GitHub Copilot CLI, OpenCode, Antigravity. One native binary. No Python.

Read the docs GitHub → 
curl -fsSL https://raw.githubusercontent.com/specnaut/specnaut-cli/main/install.sh | bash

macOS or Linux · verifies SHA256 · drops the binary into /usr/local/bin.

Speaks every major coding harness

  • Claude Code
  • Cursor
  • Codex CLI
  • GitHub Copilot CLI
  • OpenCode
  • Antigravity

From one command to a shipped feature

  1. 1

    Install & init

    One curl drops the binary; specnaut init scaffolds agents, slash-commands, and the backlog into your repo — shaped for your harness.

  2. 2

    Plan

    /task plan writes a short brief, then chains itself through implementation, review, merge and push. Expert agents settle the architecture, security and performance choices; it stops only for what only you can answer — then runs on autopilot.

  3. 3

    Review

    A dedicated review phase runs the code and security reviewers — and the accessibility, dependency and performance experts when the change calls for them — plus the quality gates — format, lint, typecheck, tests — before anything merges.

  4. 4

    Ship

    Merge on green, with the release skill and a GitHub-Project-backed backlog keeping the whole team in sync.

What Specnaut does

Scaffold

A whole agent team, scaffolded into your repo

Nine named agents — each with a clear remit and a tool allow-list — land as plain files your harness already understands. No service to run, no LLM calls of its own: Specnaut just sets the table so your harness can drive a structured workflow.

See what lands in your repo →

Review

A review phase and quality gates before every merge

The review phase dispatches the code and security reviewers from your session — and the accessibility, dependency and performance experts when the diff calls for them — then runs format, lint, typecheck, and tests as hard gates. An implement → review → fix → re-review loop keeps shipping honest — green before merge, every time.

How the review phase works →

Meet the cast

Every seat has a single job and a scoped tool allow-list, so every step of the workflow has an owner. All nine, below — the roster ships with the binary and is listed in your project's agents/README.md.

  • Product Owner
  • Developer
  • Specnaut Guide
  • Code Reviewer
  • Architect Expert
  • Security Expert
  • Performance Expert
  • Accessibility Expert
  • Dependency Expert

Specnaut Cloud

Run it headless. Answer from your phone.

The CLI puts a coordinated agent team in your repo. Specnaut Cloud takes it remote: run an agent unattended on a VM, and when it reaches a checkpoint — a clarification, a plan to approve, a merge to sign off — it opens a gate you answer from anywhere. The agent picks up where it left off. You were never at the terminal.

  • Remote-control gates — approve plans, answer questions, and sign off merges from your phone while the run continues headless.
  • Hosted backlog & team roles — a cloud Kanban with org / billing roles, no GitHub Project to wire up.
  • Built on the same CLI — specnaut cloud login links your project; everything you already scaffolded just works.
Start free → app.specnaut.com

Why teams trust Specnaut

  • Open source Public on GitHub — read every line the CLI ships.
  • Built on GitHub Spec Kit An enhanced fork of specify, not a black box.
  • No LLM of its own Specnaut scaffolds files; your harness does the thinking.
  • One native binary No Python, no runtime. curl, verify SHA256, done.
Read the docs 
curl -fsSL https://raw.githubusercontent.com/specnaut/specnaut-cli/main/install.sh | bash