What is a playbook in Black Box?

A playbook is a pre-wired sequence of steps the CEO agent follows when it recognises a familiar goal. It covers context gathering, specialist delegation, evaluator review, and the owner-facing card. Playbooks are what turn "build my landing page" into a shipped URL without you writing any of the steps.

How is a playbook different from a prompt?

A prompt asks a model for text. A playbook is a deterministic orchestration — it gathers context from your config and Company Memory, decomposes into steps, delegates to specialists, handles failures, and only resurfaces to you when a decision needs a human.

How many playbooks ship today?

One — Landing Page Bootstrap. It is the v0.1 hero skill and is available on every tier. Ten additional playbooks are on the roadmap as installable Skill Packs in v0.2 and v0.3.

Can I write my own playbook?

Not yet. Enterprise-tier customers will be able to author custom playbooks in a future release. For now the CEO can still improvise — it will decompose any novel task using its orchestration primitives, it just will not be as polished as a first-class playbook.

Where do playbooks live in the code?

They are markdown specs the CEO consults at runtime. The canonical spec for the v0.1 playbook is at apps/engine/resources/playbooks/landing-page-bootstrap.md in the Black Box engine.

Docs · Playbooks · Overview

Playbook overview

A playbook is a pre-wired sequence your CEO agent follows when a familiar goal shows up. One sentence in. A shipped outcome out.

What you'll learn

What a playbook is, in plain English.
Why playbooks beat open-ended prompting.
The shape of a playbook: trigger, preconditions, steps, fallbacks.
Where the CEO improvises and where it follows the script.
The one playbook that ships today and the ten on the roadmap.

The one-sentence definition

A playbook is a deterministic sequence the CEO follows when it recognises a familiar goal — things like "build my landing page", "clear my inbox", "find me ten leads in Austin". Instead of improvising from scratch, the CEO loads a script, fills in your context, and runs it.

Why playbooks matter

Large-language-model agents are strong at creativity and weak at consistency. Ask a raw agent "build me a landing page" ten times and you'll get ten different journeys — some skipping the deploy, some inventing testimonials, some asking clarifying questions that a well-onboarded CEO already knows. A playbook pins the sequence. The magic moment becomes reliable. The evaluation gates protect you from bad output.

The shape of a playbook

Every playbook has the same five sections:

Trigger — the phrases and situations that should activate it ("build my landing page", "put me online", first ask after onboarding).
Preconditions — the connections and config the CEO must have before step 1. If a precondition fails, the CEO emits an owner card and pauses.
Steps — ordered, numbered, with the specialist each step delegates to and the exact output expected.
Fallbacks — what happens on step failure. Silent retry once; error-alert card on second failure; never silent loops.
Success criteria — the checklist the evaluator runs before the outcome reaches you.

How the CEO picks one

On every new owner message, the CEO consults a small classifier that compares the message against the triggers for each installed playbook. If the confidence clears a threshold, it loads that playbook. If not, it improvises using generic orchestration primitives. See triggering playbooks for the full selection logic.

Playbook vs. improvisation

A playbook is not a cage. The CEO still improvises inside each step — how it phrases the brief to the coding specialist, what follow-up questions it asks the evaluator, how it recovers from a Railway deploy flake. What's pinned is the shape of the sequence and the non-negotiable success criteria. Everything else flexes.

What ships today

One playbook ships in v0.1 — Landing Page Bootstrap. It's the hero skill. A typical run goes: 90 seconds of prompting, then seven minutes of background work, then one report-ready card in your inbox with a live Railway URL. No DNS, no Tailwind config, no fighting a build.

Ten more playbooks are planned for v0.2 and v0.3 as installable Skill Packs. See the playbooks index for the full roadmap.

The owner view

You never see the playbook spec. What you see is cards — decision_required, manual_task, report_ready. The playbook is what orchestrates the backstage work so the front stage stays clean.

Frequently asked

Is a playbook the same as a Skill?: Close. A Skill is an installable bundle that can contain one or more playbooks plus the prompts and tool wiring they need. Landing Page Bootstrap ships as a built-in Skill in v0.1; later Skill Packs will ship through the Skill Store.
Do I have to pick a playbook?: No. The CEO picks. You write in plain English; the orchestration happens on the inside of the box.
Can a playbook call another playbook?: Yes. Landing Page Bootstrap already composes with a lower-level deploy-to-Railway playbook. Larger Skill Packs will do more of this.
What if the playbook is wrong for my situation?: Tell the CEO. It can abort, re-plan, or switch to improvisation. It treats the playbook as a strong default, not a law.
Will I be able to write my own?: On the roadmap for Enterprise-tier customers. Until then the team adds new playbooks through the Skill Pack release cadence — see the version-by-version plan on pricing.

Key takeaways

Playbooks turn reliable workflows into one-sentence asks.
Every playbook has a trigger, preconditions, steps, fallbacks, and evaluator criteria.
The CEO picks — you don't have to know which playbook is running.
One ships in v0.1; ten more follow as Skill Packs.