跳到正文
DEV Community· Frank Heijdenrijk·· 55 分鐘前AI 評分61

UGC Creator Skill for Claude Code:由付費活動簡報到影片發布及收益追蹤

UGC Creator Skill for Claude Code: Paid Campaign Brief In, Your Own Disclosed Video Posted, Every Payout Tracked

AI 導讀

UGC Creator Skill 是免費、MIT 授權的 Claude Code Skill,可將付費 UGC 活動簡報轉成核對清單、腳本、影片檢查、發布安排和收益記錄。

正文

UGC creator work on paid per-view campaigns looks like filming from the outside. From the inside, filming is the short part. The rest is reading a brief with fourteen rules in it, writing a script that hits every talking point without saying anything that isn't true for you, cutting out the six tries of your second line, posting the same video to TikTok, Reels and Shorts with the right caption on each, getting every link submitted inside 30 minutes, and finding out three weeks later whether the campaign paid you $6 an hour.

This post shows how to run that from an agent with UGC Creator Skill, a free, MIT-licensed skill for Claude Code (it also runs in Codex, Cursor, OpenClaw and Gemini CLI). You film. The agent reads the brief, writes the script with you, edits what you filmed, checks every video against the brief, posts after one yes, gives you the submission deadlines and keeps the ledger. Everything before posting runs on your own computer with Python and ffmpeg and works without signing up for anything.

npx skills add social-media-skills/ugc-creator-skill -g -a claude-code -y

Repo: github.com/social-media-skills/ugc-creator-skill

The plan page for a made-up UGC campaign: an EXAMPLE ONLY banner with the plan id, then a card each for TikTok, Instagram Reels and YouTube Shorts with the video's first frame, its hook, caption, settings, checks and expected earnings

From the example in the repo: a made-up app, "Example Orbit", on public-domain NASA footage with no people in it. The creator's half of every video is a labelled placeholder where your own recording goes. Nothing here is a real ad, and NASA endorses nothing here.

Who this is for, and what it won't do

This is for a real person who takes paid UGC campaigns, the kind you find on marketplaces like Content Rewards on Whop, where a brand pays per 1,000 views for a video you film and post on your own TikTok, Instagram and YouTube. It also has a mode for creators paid per video who hand the files to the brand.

It is not an AI UGC tool. The skill never generates, swaps, clones or alters a person, a face or a voice. A brief that requires an AI presenter, an avatar or a cloned voice is refused. So is posting as someone you aren't ("pretend you're a regular user"), posting from an account the brand owns, a testimonial about results you didn't get, and any brief that asks you to hide that the video is an ad.

Step 1: the brief becomes a checklist

I got a UGC campaign, brief is in brief.txt. Split screen with their footage or face plus screen recording. Accounts: TikTok @example.creator, Reels @example.creator, Shorts @examplecreator, all verified in the marketplace. I've opened the app every morning this week.

The agent copies the brief's own numbers and words into campaign.json and runs:

python3 <skill>/scripts/campaign.py campaign.json
## Verdict: GO

Conditions:
- tone notes ("keep it natural", "a regular post, not an ad") are followed as style: a relaxed, unscripted read. The disclosure stays in every video
- at most 3 submissions for this campaign

1. **Platforms and rates:** TikTok $2.00 per 1,000 views, Instagram Reels $2.00 per 1,000 views, YouTube Shorts $1.00 per 1,000 views.
2. **Payout:** minimum payout per video $5.00; cap per video $200.00; marketplace fee 10%; total budget $5,000.00; $3,820.00 left when checked 18 Nov 13:30 UTC.
3. **Pays from:** TikTok 2,500, Instagram Reels 2,500, YouTube Shorts 5,000 views per video; under that a video earns nothing.
...
8. **Never say:** "free forever", "best app ever", "official NASA app".
...
11. **Disclosure:** said out loud, ending by second 10 ...; "Ad" on screen from the first frame for 3 seconds; first in the caption; the platform's paid-partnership setting.
...
21. **Usage rights:** not stated in the brief. On most marketplaces an approved video gives the brand a license to run it, with your face and voice, in its own ads ...

The verdict is GO, NO-GO or REFUSED, with the reason in one sentence. Line 3 is the one people miss: under the minimum payout, a video earns nothing, so at $2 per 1,000 views and a $5 minimum, a TikTok has to clear 2,500 views before it pays a cent. The checklist also says what one video earns at 10,000 views after the fee, and that most videos get fewer. You confirm the checklist before anything else happens.

"Keep it natural, it shouldn't feel like an ad" is a tone note, not a reason to drop the disclosure. The skill writes a relaxed read and keeps the ad label. A brief that actually asks you to hide the ad ("no #ad", "keep the paid label off") is a NO-GO, with a line to send the brand.

Step 2: a script where every line is tagged

python3 <skill>/scripts/script.py script.json --confirm l2 --note "opened it every morning this week"
python3 <skill>/scripts/script.py script.json

The agent writes 3 to 5 hooks and the script in your voice, and every line gets a tag. From the example's script.md:

Time Line Tag
3 to 6 s Example Orbit is paying me to show you this. spoken disclosure
6 to 9 s It shows a new space photo every day. brand claim, from talking point tp1
9 to 14 s I've opened it every morning this week to see the new one. your experience, confirmed
14 to 18 s You can save any photo as your phone wallpaper in two taps. brand claim, from talking point tp2
18 to 21 s Download Example Orbit, link in bio. call to action

"Honestly it's the best space app out there" is cut, because it isn't in the talking points. An experience line only goes in after you confirm it is true for you, and --note keeps what you said. script.py refuses a brand claim that doesn't trace to the brief, a banned word, a disclosure that can't be said inside the window at a natural pace, a missing call to action, quick-money framing, and any line that denies the ad ("I'm not being paid to say this"), which no confirmation clears.

It also writes shotlist.md: per layout, how to set up the phone, where your face goes so the crop doesn't cut it, and what to say and do for each hook.

Step 3: you film

One continuous clip: all the hooks, then the script, restarting any line as often as you like. Plus a screen recording when the format needs one. Before you film, make_video.py edit.json --placeholder --width 540 renders a preview of the layout with the script as captions, so you see the framing first.

Step 4: the edit

python3 <skill>/scripts/transcribe.py takes/creator.mp4 --out work
python3 <skill>/scripts/make_video.py edit.json --out videos

transcribe.py gives a time for every word (faster-whisper on your own computer, your own Groq or OpenAI key, or an .srt you already have). make_video.py finds the opening hook and the last full try of each line, cuts the flubs and pauses, and builds one of four layouts: split screen with the brand's supplied footage on top and you below, face plus screen recording, talking head, or hands and b-roll under your voiceover. Captions are your own words. A small "Ad" label sits in the corner from the first frame, and safe zones stay clear of each app's buttons. It prints what it cut, so you don't put a flubbed line back in by accident.

The first frames of three finished vertical videos from the example campaign: NASA launch footage or a mock app screen on top, a labelled placeholder where your own recording goes below, each with its hook and the on-screen Ad label

Already edit your own videos? Hand in the finished file. The skill transcribes it, runs the same checks, and shows you the first frames to confirm the on-screen label.

Step 5: checks on what you actually said

python3 <skill>/scripts/check_videos.py job.json
orbit-tiktok (TikTok @example.creator, videos/tiktok.mp4, 22.3 s): ok
orbit-reels (Reels @example.creator, videos/reels.mp4, 22.3 s): ok
orbit-shorts (Shorts @examplecreator, videos/shorts.mp4, 21.9 s): ok

0 error(s), 0 warning(s)

This is the step a script check alone can't do. People ad-lib. The check reads the transcript of the exact take in each video, so if you said "I use it every day" on camera after confirming "this week", it is flagged before anything posts. The same pass checks the disclosure is said inside the window and on screen in time, banned words in speech and captions, #ad first in every caption, every required hashtag, @mention and link, length and 9:16 per platform, a music declaration (none, licensed with where the license is, or the brief's track), the brand's approval of this exact file when the brief asks for one, and the submission limit.

It also stops duplicates the way campaigns actually judge them. One edit on TikTok, Reels and Shorts is normal cross-posting. The same video on two TikTok accounts, or a video (or a re-cut of the same take) in another campaign, is refused, and the skill compares sound and frames, not only file bytes. It never uses speed, mirror or filter tricks to make one video look like two.

Step 6: the plan page and one yes

python3 <skill>/scripts/plan.py job.json
python3 <skill>/scripts/plan.py job.json --yes <plan id>

plan.html is the image at the top: per account the first frame, hook, caption, settings, checks, and what the video earns at 1,000, 10,000 and 100,000 views after the fee, minimum and cap. It also warns when it is night in the brief's audience countries. Your yes is bound to those exact files and captions and holds for 3 hours. Re-edit a video and the yes no longer covers it.

Step 7: posting, with the paid partnership label on

After the yes, the agent uploads each file once and sends one request with a child per account, each with its own video and caption. Trimmed from the example's request:

{
  "schedule": {"type": "PUBLISH_NOW"},
  "socialAccounts": [
    {"platform": "TIKTOK", "socialAccountId": "example-tiktok-1", "postType": "VIDEO",
     "privacyLevel": "PUBLIC_TO_EVERYONE", "allowComment": true, "allowDuet": false, "allowStitch": false,
     "isBrandedContent": true, "isYourBrand": false, "autoAddMusic": false, "...": "..."},
    {"platform": "INSTAGRAM", "socialAccountId": "example-instagram-1", "postType": "REEL", "...": "..."}
  ]
}

isBrandedContent: true is TikTok's paid partnership label, on for every campaign post. The audience, comment, duet and stitch settings are the ones you chose, never defaults. Right before sending, plan.py --check refuses a request that was edited, or whose files or captions changed after the yes. With "stagger": true you post one platform at a time, so you can do its app step and submit it before the next one goes live.

An account you post by hand gets a ready page in by-hand/ with the file, caption and settings. In the example, TikTok and Reels go through the request and Shorts is posted by hand. The example campaign can never be sent: --check refuses every request from it.

An honest note: the posting path was run end to end in WoopSocial's test sandbox (upload, validate, post, read back the link, submission page). The TikTok body with the paid partnership label passed validation, but we haven't posted a real branded TikTok with it, because that would mean posting a fake ad on a real account.

Step 8: submit within 30 minutes

Once posts are live, submit.md lists per post the link, the clock time its 30 minutes run out, and the steps in order. From the example:

2. **Reels @example.creator** (Instagram Reels), post `orbit-reels`
   - Link: https://www.instagram.com/reel/EXAMPLEorbit1/
   - Live since Wed 18 Nov 09:02 America/New_York (EST). Submit by **Wed 18 Nov 09:32 America/New_York (EST)** (13 min left).
   - Step 1, first: Instagram: edit the Reel in the app and turn on the paid partnership label ...
   - Step 2: submit the link in the campaign, then tell your agent ...

Instagram's paid partnership label and YouTube's paid promotion box can't be set through the API, so they are listed as app steps inside the window. deadlines.ics puts the deadline and the proof screenshot times (24 hours, 72 hours, 7 days) in your calendar. You paste each link into the campaign yourself: the skill never submits for you and never drives your logged-in browser.

Step 9: the ledger and earnings per hour

python3 <skill>/scripts/ledger.py --submitted <link> --campaign example-orbit-app
python3 <skill>/scripts/ledger.py --hours example-orbit-app 2.5 --note "filming"
python3 <skill>/scripts/ledger.py --report --min-hourly 15

The ledger follows each video: submitted, approved, rejected with the campaign's reason, appealed (--appeal-draft writes an appeal that quotes the rule), paid. It refuses a link already submitted or made for another campaign. Views come from the per-post view counts WoopSocial returns (collected about once a day) or numbers you read in the app, each with its source. Nothing is estimated. From the example report, with made-up numbers:

  received $33.12; open estimate $10.98 (nothing settled yet; $10.98 still counting); 1 of 2 decided rejected
  4 h logged, 1.3 h per video: $8.28 per hour received, $11.02 per hour if every open estimate pays
  under $15.00 per hour so far, but views are still counting (the newest video is 33% into its window). Judge it once 60% of the window has passed

That last line is the point of the ledger. Most videos earn little, and the campaign's own count decides what is paid. Logging your hours tells you which campaigns are worth your time and which to drop.

What stays manual, on purpose

  • Filming. Always you, always your own face and voice, or your own voiceover for a faceless brief.
  • Your choices: TikTok's audience, comments, duets and stitches, YouTube's made-for-kids answer, marketplace verification.
  • Instagram's paid partnership label and YouTube's paid promotion box, in the apps, inside the 30 minutes.
  • Submitting each link to the campaign.
  • Make-money campaigns run with the FTC's and TikTok's rules: nothing that promises quick or easy money, and a money figure only next to the brand's true typical-earnings line.

Posting, and what it costs

Disclosure: the skill is made by the team behind WoopSocial, and posting goes through WoopSocial, a TikTok posting MCP and API that also posts to Instagram and YouTube. It posts and schedules through its own approved apps, so you need no developer app or app review of your own, returns each post's link for your submission, and returns per-post view counts for the ledger. It doesn't write captions or edit video: the agent and the skill's scripts do that. At the time of writing (2026-11-19) the free plan covers 2 social accounts with no card, and TikTok, Reels and Shorts are 3, so on the free plan the agent posts 2 and gives you a ready page for the third, or all 3 go out on a paid plan. You can start on the free plan, or post everything by hand with no account at all. The connect step for your agent is on the setup page for each agent.

The skill doesn't promise income. It keeps the things that get a paid video rejected (a skipped rule, a claim you can't stand behind, a late disclosure, a late link) from happening, and tells you honestly what each campaign paid per hour.

Try it

npx skills add social-media-skills/ugc-creator-skill -g -a claude-code -y

Then tell your agent "I got a UGC campaign" and paste the brief. It needs Python 3.9+, ffmpeg and Pillow; scripts/check_setup.py prints the install command for anything missing. The full example, with every file the skill made, is in examples/orbit-ugc/. Cutting other people's footage into clips for campaigns is a different job: that's ai-clipping-skill. Issues and pull requests are welcome, and there are our other free social media skills from the same team.

來源:DEV Community · dev.to