Trial post: Instagram Trial Reels + TikTok via Zernio

Step 5 of the pipeline: idea -> script -> record -> "the card is in" -> edit-style -> trial-post -> reel-scorecard.

Why trials: a Trial Reel is shown to people who do not follow you. Every hook gets tested on strangers first; only winners are shared to your main feed (you tap "Share to everyone" in the Instagram app, there is no API for it). Three hooks per idea = three trial reels = the scorecard tells you which hook wins.

Requirements

• Claude Code with the Zernio MCP connector: claude mcp add --transport http -s user zernio https://mcp.zernio.com/mcp (sign in when asked). Instagram must be a professional account connected in Zernio; TikTok connected too.
• The connector has full access to every connected account (posting, DMs, comments, deletes). The rules below are the only guard: keep them, and never let a skill send DMs, comments, edits or deletes.
• cp config.example.json config.json, then fill accounts from Zernio accounts_list (one id per platform), your timezone, spacing and daily caps. Start low (3 to 5 a day per platform) and ramp.

Before any post (per reel)

1. The edit passed every edit-style gate, including face-first, dead-air and join-envelope, on BOTH files: final.mp4 (Instagram, comment-KEYWORD ending) and final-bio.mp4 (TikTok, "link in my bio" ending).
2. Captions from reel-captions passed the checker. The Instagram caption carries the keyword CTA; the TikTok caption never says "comment" and does carry your bio line.
3. Every claim in the video was checked live today. A false or outdated line is cut, not posted.
4. Never an identical clone of something already posted. A new hook, a new edit, a genuinely different video. No fingerprint-evasion tricks (mirror flips of posted footage, speed or pitch nudges, noise): they read as spam and get accounts restricted.

Post (one platform at a time)

1. Upload. mcp__zernio__call_tool name media_get_media_presigned_url, arguments {"filename": "<short>.mp4", "content_type": "video/mp4", "size": <bytes>}. Save the result. PUT the file to uploadUrl with curl -X PUT -H "Content-Type: video/mp4" --data-binary @final.mp4 "<uploadUrl>" (no auth header), then download publicUrl and compare sha256 with the local file. Mismatch = upload again.
2. Build. python3 scripts/build_post.py --platform instagram --idea R6 --hook A --video final.mp4 --media-url <publicUrl> --caption CAPTION-IG.txt --now (--at 2026-11-02T18:05:00 for a slot, --ai for avatar or AI-generated footage). It refuses on a STOP file, a duplicate file, a spacing gap, a used-up cap, or a caption that breaks the rules, and prints the arguments.
3. Call. mcp__zernio__call_tool name posts_create_post with EXACTLY the printed arguments. Save the response.
4. Verify. python3 scripts/verify_post.py --plan <plan> --response response.txt. Exit 0 = recorded (Instagram: trial confirmed). Exit 4 = still publishing: wait a minute, posts_get, verify again. Exit 3 = STOP.
5. Repeat for TikTok with final-bio.mp4 and the TikTok caption.

THE TRAP (read twice)

dry_run exists for TikTok only; an Instagram dry run is rejected. So the proof of a trial is the create response: the instagram target must show trialParams.graduationStrategy: "MANUAL" (and isTrialReel: true once listed). Payload shapes and response examples: references/zernio.md.

Stop rules (any one: stop the batch, tell the creator the same minute)

• An Instagram response without MANUAL trial proof, or with SS_PERFORMANCE (auto graduation).
• A failed status, no post id, or a create error.
• A trial-cap error ("maximum number of trial reels"), a TikTok spam_risk error, any account-status or recommendation-eligibility notice.
• A post still scheduled or publishing 30 minutes after its slot (over-cap posts can be held silently).

verify_post.py writes $HOME/.trial-post/STOP on the first three; build_post.py refuses until a human has looked in the app and deleted the file. Never "fix" a post with an edit, delete or retry call.

Never

X or LinkedIn from this skill, DMs, comments, deletes, edits, unpublishing, paid-partnership flags, collaborator tags, a sponsor label burned into the video (disclosure goes in the caption), or posting to an account not in config.json.

Other schedulers (adapter)

Not on Zernio, or also scheduling the main feed elsewhere? references/scheduler-adapter.md lists what any scheduler must support (trial reels or not, the AI flag, per-platform captions, future-dated only) and a worked example for Metricool. Fill the adapter section for your tool; keep every rule above.

After posting

Hand the ledger ($HOME/.trial-post/ledger.jsonl) to reel-scorecard at 72 h. Graduation stays the creator's tap.