[x] SpecBox · harness for coding agents

From spec
to evidence.

SpecBox makes your coding agents build what was asked and proves it criterion by criterion, with a receipt your client understands too.

Install the VSCode extensionSee a real delivery

Engine, extension and CLI are free and open. For agencies, tech leads and founders who build with agents and have to prove what was built.

A verified criterion from our own board

US-53 › UC-5301 › AC-03

UC-5301 · AC-03verificada

Terminado ese periodo, projects.embed.build/<ruta>?<parámetros> responde 301 a projects.specbox.build/<ruta>?<parámetros>.

  • Pull requestPR #41: projects.embed.build/<ruta>?<parámetros> → 301 a projects.specbox.build (check-redirects 12/12)

    por Jesús ·

verificada por Jesús ·

UC-53013/3 criteria with a receiptSnapshot from 2026-10-04 21:29 UTC

Real data from our board, which we write in Spanish.

The LLM brings the speed. SpecBox brings the control and guarantees the quality.

Spec frameworks

suggest.

AI code reviewers

check afterwards, and only the code.

[x] SpecBox

blocks what doesn't comply and checks it against the requirement.

SpecBox is built with SpecBox

2,510 use cases. Each square is one.

Everything on the ecosystem's boards: what's done, what's waiting for review, what's in progress and what's still queued. No rounding.

projects
31
developers
18
stories
503
acceptance criteria
10,077
use cases done
80 %
  • Done2,011
  • In review47
  • In progress44
  • Queued408

Snapshot from 2026-10-04 21:29 UTC· Aggregated data, no personal information; archived ones are left out. See activity

Evidence, not stories

A real delivery, step by step.

The history of a use case on our board: when it started, which criterion was verified with which receipt, when it closed and, if a person accepted it, when. The board records it, not a summary written afterwards.

History of UC-5301

  1. Jesússtarted the use caseUS-53UC-5301Estado: en curso

  2. Jesúsverified the criterionUS-53UC-5301AC-01Estado: hecho

    PR #75 desplegada: projects.specbox.build sirve el portal; contraseña y código llegan a Auth

  3. Jesúsverified the criterionUS-53UC-5301AC-02Estado: hecho

    projects.embed.build sigue sirviendo y autenticando (sin 301 hasta 2026-10-31)

  4. Jesúsverified the criterionUS-53UC-5301AC-03Estado: hecho

    PR #41: projects.embed.build/<ruta>?<parámetros> → 301 a projects.specbox.build (check-redirects 12/12)

  5. Jesúscompleted the use caseUS-53UC-5301Estado: hecho

Snapshot from 2026-10-04 21:29 UTCReal data from our board, which we write in Spanish.

US-53 › UC-5301 › AC-01

UC-5301 · AC-01verificada

https://projects.specbox.build sirve el portal y permite entrar con contraseña y con código por correo (Google queda fuera mientras el proveedor no esté activo: UC-5303, decisión de Jesús 2026-10-01).

  • Pull requestPR #75 desplegada: projects.specbox.build sirve el portal; contraseña y código llegan a Auth

    por Jesús ·

verificada por Jesús ·

Without a receipt, a criterion is not verified. And if the receipt says it fails, the card shows it in red: it doesn't hide it.

An instructions file is not an enforcement mechanism

Hooks don't suggest. They block, and explain why.

  • Read before you writeThe agent can't touch a file it hasn't read in the session.
  • Quality gates between phasesZero lint, coverage that never drops and no previously passing test in red.
  • No receipt, no acceptanceEvery criterion closes with a test, a diff, a screenshot, a URL or a PR.
  • Every block explains itselfWhat was blocked, why and what to do. A silent guardrail only breeds rejection.
PreToolUse · Edit · quality-first-guardblocked
============================================================
QUALITY FIRST: Read before you write
============================================================
File: src/components/home/Hero.astro
You are trying to modify an existing file without reading
it first. …
To proceed:
1. Use the Read tool to read this file
2. Understand what's already there
3. Then make your changes
============================================================
[x] The agent reads the file and tries again.

Done means the same for both of you

The same data, in the language of whoever looks at it.

Your team works with IDs, branches and states. Your client sees stories in plain language and accepts each deliverable with its evidence. Same board: nobody translates by hand.

Real data from our board, which we write in Spanish.

A real epic from our board

How your team sees it in SpecBox Cloud

  1. US-50Todo el ecosistema tiene su dirección decidida bajo specbox.build
    2/3 UCen curso
  2. US-51El servidor MCP responde en mcp.specbox.build y los clientes se mudan solos
    2/2 UChecho
  3. US-52El site de SpecBox vive en specbox.build
    2/2 UChecho
  4. US-53El portal de clientes vive en projects.specbox.build
    3/3 UChecho
  5. US-54El correo del ecosistema sale y se contesta desde specbox.build
    0/1 UCbacklog
Snapshot from 2026-10-04 21:29 UTC

How it works

Four commands. Each leaves something you can review.

  1. 01/discovery

    Who it's for and which job it does for them, before writing a line.

    Leaves icp_jtbd.md

    Before moving on: the discovery has to be complete.

  2. 02/prd

    Stories, use cases and measurable acceptance criteria, straight on the board.

    Leaves US · UC · AC

    Before moving on: every criterion has to be checkable.

  3. 03/plan

    Phases, components and a design for each screen. Work is estimated before it starts.

    Leaves plan + designs

  4. 04/implement

    Branch, code, tests, quality gates and PR. Each criterion closes with its receipt.

    Leaves PR + evidence

    Before moving on: zero lint, coverage that doesn't drop, no test in red.

Speed that survives review

Use cases closed, week by week.

Each bar is the use cases that moved to done that week, across all projects. We publish the slow weeks too.

Weeks starting on the Monday shown (MM-DD); the last one may still be in progress. Snapshot from 2026-10-04 21:29 UTC

What's free and what isn't

Open for those who build. Paid for those who govern.

  • Free and open

    Engine, extension and CLI

    The full harness in VSCode and Claude Code: commands, hooks, quality gates and a receipt per criterion.

    Install the extensionCode on GitHub
  • For teams · per seat

    SpecBox Cloud

    Multi-user board, your agents' sessions, audit trail and cost per use case.

  • For your clients · per project

    Projects portal

    Your client follows progress by story and accepts each deliverable with its evidence, without seeing a single ID.

Let your agents prove what they did.

Install the extension, open Claude Code and start with /discovery. The first receipt arrives with the first closed criterion.

Install the VSCode extensionI'm an agency or a team

Or from the VSCode command palette:ext install EmbedBuild.specbox-engine