How does the CEO know which playbook to run?

Every playbook declares trigger phrases and situations. On each owner message the CEO compares intent against installed playbooks; if confidence clears a threshold, it loads that playbook. Otherwise it improvises.

Can I force a specific playbook?

Yes — ask for it by name. "Run Landing Page Bootstrap" or "use the landing-page playbook" both route directly. Skill Packs will expose friendlier names in the app once the Skill Store lands in v0.2.

What if two playbooks match?

The CEO picks the most specific and confirms with a short decision_required card if the match is ambiguous. It will never silently run the wrong one.

What happens when no playbook matches?

The CEO falls back to improvisation — it decomposes the task using generic orchestration primitives and delegates to specialists without a canned script. The work still ships; it just will not be as polished as a named playbook.

Does the CEO learn which playbooks I use most?

Yes. The outcomes log and Company Memory bias future routing toward playbooks that have worked for you before. The more you run, the sharper the routing.

Docs · Playbooks · Triggering

Triggering playbooks

How the CEO decides whether to run a named playbook or improvise from scratch — and how to steer the choice yourself.

What you'll learn

How trigger phrases are declared on each playbook.
The classifier and confidence threshold the CEO uses.
What preconditions do before a playbook actually runs.
When the CEO falls back to improvisation.
How to force a specific playbook when you want to.

Trigger phrases

Every playbook declares a short set of trigger phrases and situational hints. Landing Page Bootstrap declares these triggers: "build my landing page", "make me a site", "landing page", "website", "put my business online", "get me online". It also declares a situational hint: the owner's first request after onboarding.

You don't have to match the wording exactly. The classifier is a small model that compares intent, not strings. "Can you put me on the internet?" routes to the same playbook as "build my landing page."

The confidence threshold

On every new owner message the CEO runs the classifier against each installed playbook. The highest-scoring match has to clear a confidence threshold to be auto-selected. Below threshold, the CEO either asks a clarifying question or improvises.

In v0.1 there is only one installed playbook, so the decision is binary: does this ask look like a landing page? If yes, run it. If no, improvise. As more playbooks install through Skill Packs in v0.2+, the classifier will start disambiguating between them.

Preconditions before the first step

A matched playbook is not the same as a running playbook. After matching, the CEO runs the playbook's preconditions. If any fail, it pauses the playbook and emits an owner card — a manual_task to connect a missing service or a decision_required to fill in a missing field. The playbook resumes from step 1 once the card is resolved.

For Landing Page Bootstrap the preconditions are GitHub connected, Railway connected, and all five onboarding answers present. See the Landing Page Bootstrap walkthrough for exact wording.

Playbook vs. improvisation

When nothing matches, the CEO improvises. That means it decomposes the task using the generic orchestration playbook (which is not a specific playbook — it's the meta-plan for breaking arbitrary work into steps and delegating), picks specialists on the fly, and runs it. The work still ships; the evaluator still gates it. The difference is that a named playbook has been hand-tuned for a common case — better copy rules, tighter error paths, more-polished owner cards.

Forcing a specific playbook

Sometimes you know what you want. Three ways to force a specific playbook:

Name it. "Run Landing Page Bootstrap."
Use a trigger phrase verbatim. "Build me a landing page."
Pick it from the Skill Store (v0.2+). Installed Skills expose a run button in the app.

Ambiguity handling

If two playbooks match with similar confidence, the CEO emits a decision_required card: "Did you mean Landing Page Bootstrap or Custom Domain Setup?" You pick; the playbook runs. The CEO never silently picks when the choice is close, because that is how agents ship the wrong outcome.

Learning over time

The outcomes log and Company Memory bias future routing. If you've run Landing Page Bootstrap three times and always approved the result, the classifier's threshold for that playbook drops a little — it'll route faster. If you've rejected an output, the threshold climbs. Routing gets sharper with use.

Frequently asked

How does the CEO know which playbook to run?: A small classifier over installed playbook triggers, gated by a confidence threshold. High-confidence matches route automatically; ambiguous ones ask you.
Can I force a specific playbook?: Yes — name it, quote a trigger phrase, or pick it from the Skill Store once that view lands.
What if two match?: You get a decision_required card asking which one. No silent misrouting.
What if none match?: The CEO improvises. The work still ships, just without a named, hand-tuned sequence.
Does routing improve over time?: Yes. Company Memory biases future matches toward playbooks that have succeeded for you.

Key takeaways

Triggers are declared on each playbook and matched by intent, not strings.
A matched playbook still runs its preconditions before step 1.
No match means improvisation, not failure.
You can always name a playbook to force it.
Routing sharpens with use as Company Memory grows.