Agents and media

TL;DR

A media actor turns a brief into a reviewed video by filling an existing template and rendering it. It picks a template, drafts the slot values, validates them, renders, and the result waits for a human yes. It never authors templates, never calls an image or video generation model, and never publishes on its own.

Rules

• Video is always a template render. No text-to-video, no image models.
• Actors do not create or edit templates. When no template fits the brief, the actor writes one "asset blocked" note (what the brief needed, which templates it checked) and stops.
• Every render and every post goes to review. Actors never publish directly and never change a channel without an approved review.
• Pick files, do not generate them. Image and clip slots take files that already exist in Files.

From a brief to a reviewed video

1. Find a template with data.video.template.list. Prefer templates by key (for example acme/product-update) over names. Read its slots, sizes and looks with data.video.template.get.
2. Draft the slot values from the brief, only for slots the template has.
3. Validate with video.render.validate (gates: "error" for agents). Fix every error and try again; after two attempts, write an "asset blocked" note with the errors.
4. Render with video.render.video (template, version, values, size, look). The MP4 and poster land in Files and the video waits in the review inbox.
5. Report the template key and version, size, look, values and the video's id.

For posting, the actor uses publisher.post.create-with-approval with a look and the rendered file; the post waits for the reviewer like any other.

What actors fill

Copy rules are errors for agents: no en or em dashes, no retired positioning. People see the same findings as warnings.

When something stops the actor

See also Staff and agents and Media actors.