CDCStream/skills

Creator vetting with transcripts

Build a one-page brief on a creator from public data: who they are on each platform, how often they post, how their recent posts perform, which brands they already work with, and what they say on camera.

What this can and cannot tell you

• Only public, logged-out data. Private accounts, audience demographics, follower authenticity scores and DMs are not available — say so; never invent them.
• Engagement rate here is (likes + comments) / followers over the posts pulled. It is a rough signal from a small sample, not an audited metric.
• Some counts are rounded by the platform (e.g. "1.2M"); keep the ...IsApproximate flags from the response and show "≈" for them.
• Transcripts come from speech in the video (TikTok, Instagram) or published captions (YouTube). A missing transcript means no speech or no captions, not that the creator said nothing worth checking.

Inputs to collect first

1. The creator's profile URLs or handles on TikTok, Instagram and/or YouTube. Do not guess handles across platforms — ask, or mark the platform as "not provided".
2. How many recent posts per platform (default 10) and how many transcripts (default 3 most recent videos per platform).
3. What the brand cares about: competitors to look for, topics to avoid.

Credit estimate — show before running

Per creator, with defaults (10 posts, 3 transcripts per platform):

Platform	Calls	Credits
TikTok	/v1/tiktok/channel-details 1 · /v1/tiktok/channel-posts 2 · 3 × /v1/tiktok/transcript 2	9
Instagram	/v1/instagram/channel-details 1 · /v1/instagram/channel-posts (0.3 per post, 10 posts = 3) · 3 × /v1/instagram/transcript 2	10
YouTube	/v1/youtube/channel-details 1 · /v1/youtube/channel-videos 2 · 3 × /v1/youtube/transcript 1 · /v1/youtube/video-sponsors 0	6

All three platforms: about 25 credits per creator; up to ~35 with more posts or transcripts. Print the estimate for the requested creators, check GET /v1/account/balance (free), and ask the user to confirm.

How to call Captapi

• MCP: tiktok_channel_details, tiktok_channel_posts, tiktok_transcript, instagram_channel_details, instagram_channel_posts, instagram_transcript, youtube_channel_details, youtube_channel_videos, youtube_transcript, youtube_video_sponsors.
• REST: GET https://api.captapi.com<path>?url=<profile or video URL> with header Authorization: Bearer $CAPTAPI_API_KEY.

Failed calls cost 0 credits. A 403 such as profile_restricted or a 404 is a final answer for that platform — record it and move on.

Steps

For each platform the user provided:

1. Profile — channel-details with the profile URL. Keep followers, following, post count, bio, bio links, verified flag, account type.
2. Recent posts — tiktok/channel-posts, instagram/channel-posts or youtube/channel-videos with limit=<N>. From the rows compute: posts per week over the window, median views (when present), median likes and comments, engagement rate, and the best and worst post.
3. Sponsor signals from the same rows:
   • Instagram rows carry isPaidPartnership; list those posts and the brands they mention.
   • TikTok and YouTube: scan captions/titles for #ad, #sponsored, "paid partnership", discount codes and brand @mentions.
   • YouTube: youtube/video-sponsors?url=<video URL> for each transcribed video returns community-submitted sponsor segments (free). Label them as community-reported.
4. Transcripts — for the most recent videos, tiktok/transcript, instagram/transcript or youtube/transcript with the video URL. Read them for: products and brands named, competitor mentions, claims that would need substantiation (health, finance), and language or topics the brand listed as off-limits.

Brief format

1. Snapshot table: one row per platform — handle, followers, posts/week, median views, engagement rate, verified. Each cell traceable to its endpoint.
2. Past sponsors: brand, platform, post date, link, and how it was detected (isPaidPartnership, caption tag, community segment).
3. What they say: 3–5 bullets drawn from transcripts, each with a short quote and the video link.
4. Flags [needs human review]: competitor work, risky claims, off-limits topics, sudden follower jumps between profile and post data.
5. Gaps: platforms not provided, restricted, or with no transcripts.
6. Credit total for the run.

Where a human must review

• Every flag — transcripts can be misheard and context matters.
• Whether a detected "sponsor" was paid or just a mention.
• Fit with the brand: the brief gives evidence, not a hire / no-hire verdict.