AI UGC Test Loop

Run a research-first AI UGC testing loop. Agent Media makes the video take; you handle reference intake, outlier scoring, prompt writing, review, finishing, and the publish packet. This workflow is inspired by public AI UGC workflows in general terms. Do not copy exact scripts, protected likenesses, trademarks, or shots from references.

Use only Agent Media MCP tools that are listed in the user's client. This skill may use get_account, list_models, list_characters, open_upload_panel, get_uploads, upload_image, quote_ugc, make_ugc, quote, generate_image, generate_video, generate_audio, get_run_status, and rate_run. Missing steps are handled with normal files, local editing, the user's pasted references, and your own analysis. There is no Agent Media scraper, reference analyzer, campaign tool, batch tool, app-screen insertion tool, caption exporter, or publish packet exporter.

Operating rules

• Research first. Ask for product context, account baseline, and 3 to 10 references before rendering.
• Score references against the account's own baseline. Use outlier_score = reference_views / account_baseline_views when both numbers exist.
• Change one variable per test. Good variables are hook, creator type, starting frame, objection, proof moment, offer, and CTA.
• Render one test take first through Agent Media. New OAuth connector accounts may use the free first ad when get_account.free_first_ad_available is true and the quote confirms a 0-credit, 10 second, 480p, watermarked video.
• Batch only after the user approves the first take and says to render more variants.
• Finish final variants with a real app screen recording and captions. Do not fake app screens when the user can supply a recording.
• Build a publish packet with final URLs or files, variant IDs, post copy, CTA, thumbnail notes, baseline metric, and UTM links that include utm_source=skill.

1. Product context

Ask for only what you need: