Most social analytics tools hand you a dashboard full of numbers. None of them let your AI agent actually look at a video and reason about why it underperformed. Developer Usama Latif has changed that by exposing the Reelyze analysis engine as a plain HTTP API with a self-contained SKILL.md file that plugs directly into Claude or Cursor, giving agents eyes on Instagram Reels, TikToks, and YouTube Shorts.

What the API Actually Does

Give it a public short-form video URL and you get frame-by-frame hook analysis, first-3-second retention scoring, scene-by-scene signals with exact drop-off timestamps, plus transcribed audio. The engine can also download clean MP4 files, extract audio as MP3, generate full timed scripts (hooks → body → CTA), or batch out content ideas from a topic. The base URL is https://api.getreelyze.com and authentication uses a single Bearer key (rk_live_...) created in the Reelyze dashboard. CORS is open by default, so it works equally well from a backend service, browser extension, or agent environment—no SDK required.

Async Video Tools Use Submit-Then-Poll Pattern

The video analysis tools aren't instant because they're pulling frames, running OCR on on-screen text, transcribing audio, and scoring retention. You submit a job via POST to /v1/analyze, get back a job_id with status "queued", then poll /v1/jobs/{job_id} every few seconds until status flips to "completed". The completed response includes report_markdown with verdict, hook assessment, scene-by-scene retention data, and the exact drop-off timestamps. Transcript, downloader, and audio extraction use the identical submit-then-poll shape—you just read transcript_text or artifact_url instead of report_markdown when done. Content generation endpoints (/v1/script and /v1/ideas) return synchronously with no polling needed.

The Agent Skill Eliminates Boilerplate

Instead of writing your own poll loop, you drop in a single SKILL.md file (MIT-0 licensed) that teaches any compatible agent the full endpoint list, async pattern, and how to parse the analysis report. Set one environment variable—REELYZE_API_KEY—and the agent handles everything: it submits jobs, polls until completion, and delivers actionable verdicts. A prompt like "audit this reel and tell me why it flopped" returns hook strength scores, specific drop-off timestamps with explanations (like "3s of B-roll with no narration at 0:11"), and concrete fixes ("cold-open on the result; cut the intro"). The same skill works across Claude, Cursor, browser tools, or your own backend since it's plain HTTP behind a Bearer key.

Workflow Chaining Is the Real Win

The author makes a compelling point about why this matters more than single API calls: chaining. An agent with Reelyze access can transcribe → analyze → rewrite the hook → generate the next script in one flow, compare two competing reels' hooks side-by-side, or audit a competitor's winner and draft an informed response—all without human tab-switching. The analysis becomes a step in a workflow instead of a dashboard you check separately. Free tier gets you 50 calls per day on transcript, downloader, and audio extraction; full video analysis requires paid plans but the first run is free.

Key Takeaways

  • Video analytics API supports Instagram Reels, TikToks, and YouTube Shorts via single URL input
  • Frame-by-frame hook scoring includes exact drop-off timestamps with specific fix recommendations
  • Async jobs (analyze/transcript/download) use submit-then-poll pattern; sync endpoints for script generation
  • Single SKILL.md file enables Claude and Cursor agents to handle full video audit workflows
  • Open CORS means same skill works across agents, backends, or browser tools—no SDK dependency

The Bottom Line

This is exactly the kind of infrastructure glue the AI agent ecosystem needs—well-scoped APIs with zero-friction auth that slot into existing tooling without ceremony. If you're building anything in the creator economy space and not thinking about how agents will need to reason about video performance, you're already behind.