Stackific · open source

Stax

An evidence-based, spec-driven agent skillset. Enterprise accuracy at startup speed with a two-skill vocabulary.

A tiny workflow. Mighty results.

Most agent skill libraries pile up commands and turn into noise. Stax ships exactly two skills: one to file one or more new work-items, one to work the stack. Everything else is the scaffolding that keeps your agent honest.

               
/scopeFiles a new work-item.
/scope opens one or more fresh work-items in the current folder — Goal, Inputs, Changes, Systems, and EARS acceptance criteria slotted into a numbered markdown your agent fills in.
         
/shipWorks the stack.
/ship picks up the next work-item in line and executes it against your codebase. Scoping in one window, shipping in another — the loop runs until the stack is empty.
                               
One install, every agent.Nine agents and counting.
Anthropic Claude Code, Cursor, GitHub Copilot, Google Antigravity, Kilo Code, OpenAI Codex, OpenCode, Pi, Zed — one stax init reaches every agent on the machine.

Built to develop serious softwares.

If you are building serious software, you must prioritize quality and accuracy over quantity — every single time. These are the design choices in Stax that make that possible.

Spec-driven, not vibe-driven.

Every scoping exercise let you and AI agree on EARS criteria, which makes ambiguity theoritically impossible.

Catch contradictions and misunderstandings in minutes, not days. Default, yet configurable, 30-line cap keeps each scope reviewable; per-task or per-work-ttem review is your call.

Plans live in git, forever.

Every plan is a markdown in .stax/. Versioned, reviewable, auditable. Work-items extend, supersede, or retire each other — and the chain is preserved in your history.

A central system registry in YAML names every system. EARS criteria resolve against it. No AI interpretation drift, even months later.

Guard rails on by default.

Four small mechanics keep Stax projects on rails without the ceremony of a heavyweight process.

 30-line plan cap
Default, yet configurable plan-size guard rail. Forces your agent to split sprawling work into smaller, reviewable plans.
 
 Central system registry
One YAML lists every system in your project. EARS criteria resolve against it. No AI interpretation drift.
 
 Per-task or per-plan review
Pause for review after each EARS criterion or only at work-item boundaries. Tight loop or loose loop — your call.
 
 AGENTS.md as constitution
Stax respects every agent's native rules file — CLAUDE.md and AGENTS.md — as the governing layer over every plan.
 

A local dashboard for the plans your agent writes.

Run stax from any project. A browser opens at http://localhost:7829. No signup, no cloud round-trip, the bytes never leave your machine.

Open tasks jump out

Colored flag icons on any plan with unchecked - [ ] items. Scan the home feed and you know where to pick up.
Navigate, don't grep

Work-items link to systems. Systems link back to every work-item that powers them. Superseded or deprecated work-items are clear — history is one click away in either direction.
Find any work-item from a half-remembered phrase

A system name, a title fragment, or just a few words you wrote in the body — the matches show up the moment you type them. No more opening files you half-remember writing last week.
Local by design

Loopback only. Scope markdown is rendered server-side with raw HTML escaped, so pasted snippets cannot execute.

Install in under a minute.

One command per platform. The installer drops the stax binary and the bundled skills into your home directory and adds itself to PATH.

macOS / Linux

curl -fsSL https://stackific.com/stax/install.sh | sh

Windows (PowerShell)

Set-ExecutionPolicy Bypass -Scope Process -Force
iex (irm https://stackific.com/stax/install.ps1)

Inside WSL2, use the macOS / Linux command.

First run

cd your-project
stax init        # install bundled skills + seed .stax/

# in your agent
/scope <describe what you want to build>
/ship            # continuous execution loop

# back in your shell
stax             # opens the local Stax web UI in your browser

See the Stax reference for flags, exit codes, and CI examples.

Open source. Apache 2.0.

Stax is free for any team to use. If you want hands-on help adopting it across your engineering org, that is exactly the kind of work we do.

Roll Stax out across a team or rebuild an existing AI workflow around spec-driven planning. We can help.