Video and post

The end-to-end loop for one video item: content → script.json → render → review → publish.

Antigravity / Gemini: same steps, run the commands directly.

1. Pick the item

From the idea queue (node scripts/queue-client.mjs pull), from today's PLAYBOOK phase, or from what the user handed you. Check history.json — skip anything already done.

Work in brain/<slug>/.

2. Make the video

Invoke the create-video skill (or follow it as a document). It ends with a validated script.json and a rendered video.mp4, and prints VIDEO=<path>.

3. Write the caption

Separate from the narration — this is the text of the social post, not what is spoken:

• post — the caption. Follows the project's WRITING_CRAFT.md voice, not the video's. Emoji and hashtags are fine here.
• comment — the first comment (engagement / link / source credit).
• title — required by YouTube; ≤100 characters.
• hashtags — e.g. #ai #coding.

Never put the source's copyrighted text in the caption. Summarize in your own words and link.

4. Review gate (mandatory)

Invoke the review-gate skill on the item — an independent reviewer, scoring the video section of the rubric plus the caption. Watch specifically for:

• validator clean at --strict
• hook lands in the first 3 seconds
• narration reads naturally aloud — every number spoken correctly
• template variety; on-screen text not overflowing its limits
• outro carries the right brand and URL

Pass → continue. Fail → fix (2–3 rounds) → drop and log why.

5. Publish

Check the payload first — it costs nothing:

node scripts/social/make-post.mjs --video brain/<slug>/video.mp4 \
  --post brain/<slug>/caption.txt --comment brain/<slug>/comment.txt \
  --title "…" --hashtags "#ai #coding" \
  --platforms tiktok,youtube_shorts,facebook_reels,instagram_reels --dry-run

Then either post now (drop --dry-run), or schedule it — the better default, because rendering is slow and the posting slot should be instant:

node scripts/scheduler/build-queue.mjs items.json     # assign times → queue.json
node scripts/scheduler/register-tasks.mjs             # schtasks / cron lines

The queue item must be type: "video" and carry videoPath (already rendered), post, comment, title, hashtags, platforms — see templates/state/queue.example.json. run-item.mjs then only uploads and calls the webhook when the slot fires.

6. Record and report

• Add the title to history.json (dedup).
• Mark the source posted in the idea queue: node scripts/queue-client.mjs posted <url>.
• Clean per-run scratch files; keep script.json, video.mp4, voice.mp3, script.txt.
• Write brain/<slug>/report.md: what was made, the review verdict, where it went.

Notes

• Rendering needs FFmpeg + Chrome + a local OmniVoice server, and cannot run on GitHub Actions — see docs/06-scheduling.md.
• The Make.com scenario is a Webhook → Router, one filtered route per platform. Payload contract: docs/11-social-posting.md.