THE PRODUCT
From “publish this” in chat to a controlled publishing pipeline.
The system connects a GPT-driven content workflow to an execution layer. Chat commands create queue records with APPROVED or PENDING status, support scheduled publishing, and keep the content workflow separate from the irreversible publishing step.
END-TO-END FLOW
Every step is explicit and inspectable.
The planner polls the queue on a schedule and supports delayed publishing through scheduled_at. Media is checked before publishing, then delivered through the Meta API.
RELIABILITY & SAFETY
Publishing is treated as an irreversible operation.
Checks file availability, location and media type before any publish call.
Previously published commands are blocked with duplicate protection rather than posted twice.
Persistent storage keeps publication state across deployments.
APPROVED, PENDING and scheduled actions separate planning from execution.
VERIFIED RESULT
A live end-to-end publish completed in 70 seconds.
In a verified test, the full chain ran from a chat command through the queue, preflight, signed URL, Meta processing, publication and ledger write. A repeated request was blocked as already published, confirming duplicate protection.