HUX-L1 — Ad creative factory

Part 1 — the plain-english view

Creative volume without brand drift.

What it watches

Nothing on a schedule — it's a goal-loop: it runs when given a brief and stops when the brief is satisfied
Its own output, against a scored brand checklist: palette, typography, tone, claim-safety — all mandatory checks must pass

When & how it speaks

Delivers 3 concepts × the required formats per brief, with an export manifest (sizes per placement, naming), into the photo library's review folder. Sherif approves in the library, not in a chat.

Why it exists

Sherif runs marketing alone; creative production is the bottleneck — and rushed creative drifts off-brand. The factory produces volume; the brand rulebook (encoded as a scoring skill) keeps every piece on-brand before a human ever reviews it.

3 ×

concepts per brief, each in every required placement format

100%

of pieces pass the mandatory brand checks before Sherif sees them

1 write

the only one it has: filing drafts into the review folder of the photo library

📝

Brief arrives

One line from Sherif is enough

🎨

Concepts + copy

Designed in the brand voice

📋

Brand gate

Scored checklist — fails are reworked, not shipped

📁

Filed for review

Export manifest + review folder in the library

✋

Sherif approves

Publishing is a human act, always

What it will never doNever uploads to Meta or Google — approved humans publish. Its one write is scoped to the photo library's review folder.

Where it stands today · July 2026Design complete, not yet built — deliberately queued behind the tracking-foundation owner decision, and one review cycle with Sherif is planned in the first month (this loop leaned hardest on the retired designer model's taste; Opus + the brand-rulebook verifier carry it now).

Part 2 — under the hood

How it's wired: systems, models, and the path a number takes.

For implementers and the technically curious. The full build sheet — verified queries, thresholds, and build notes — lives on the specs page.

StepBrief

One line from Sherif

→

StepOpus concepts + copy

3 concepts × formats, brand voice

→

StepBrand verifier

hux-brand-guidelines as a scored checklist

→

StepExport manifest

Sizes per placement, naming convention

→

StateDAM review folder

The scoped write — drafts land here

→

Human gateSherif approves

In the photo library; then humans publish

System we read Automated step State / memory Human decision

System	Role in this loop
Directus DAMphoto library	Where drafts land — a review folder with a dedicated agent role; the single sanctioned write, scoped and revocable.
Brand rulebookhux-brand-guidelines skill	The official brand guide encoded as a scored checklist — the machine version of “does this look like us?”
Agent Vaultcredential store	Where API keys live. Loops get read-only credentials from the vault; nothing is hardcoded in scripts.

Model / brain	What it does here
Claude Opusthe judgment model	Reviews alert wording, thresholds, and anything a human will read and act on. Post-Fable, Opus owns everything that ships.
GPT-5.5 via Codexthe bulk-work model	Wrote and maintains the mechanical parts — SQL, diffing, digest assembly. Effectively free on our existing subscription, so routine cycles cost almost nothing.
Claude Fablethe designer (retired)	Designed this loop's spec and encoded the judgment as thresholds and checklists before its retirement on 8 July 2026 — so cheaper models can run it without Fable-level reasoning.

State & memory

Brief history + which concepts were approved/rejected — the factory learns Sherif's taste over time.

Delivery

Drafts into the photo library review folder; a summary note to Sherif when a brief's set is ready.

Safety rails

One scoped write (review folder only); mandatory brand checks; claim-safety is part of the checklist so no unsubstantiated product claims ship.

Before it can run for real

Build the brand-verifier scoring skill
Sherif confirmed as the approval gate (pending his onboarding conversation)

Full build sheet →