Skip to content

106 social media skills for AI agents - strategy, writing, video, design, platform growth, publishing, and analytics. Works with Claude, Cursor, OpenClaw, Hermes & 40+ agents.

3.6 KB
Raw Blame

OpusClip — integration & connection guide

What it is: the AI clip-finder (opus.pro) this stack uses to turn long video into short-clip candidates. OpusClip finds + cuts candidates; the human approves; WoopSocial publishes. Skill: skills/opus-clip/; the general craft lives in skills/captions-and-clipping/ (clipping.md covers the multi-tool clipping layer; this file goes deep on OpusClip itself). Tiers/gates shift and sources conflict — verify-quarterly, confirm in-app.

Connection layers

  1. API — reported Business-tier (custom pricing) only. Below that, no direct agent automation of OpusClip itself; never claim it. Zapier/Make integrations exist for parts of the flow (e.g. new-clip triggers) — usable on lower tiers, always with human review in the loop (no unreviewed auto-post pipelines on any tier; a ~40% discard rate makes review non-negotiable).
  2. In-app (human executes): the agent supplies the pipeline plan — genre/length settings, negative prompts ("exclude intros, audio checks, sponsor reads"), moment-search phrasing, the triage plan, the per-clip QA checklist, caption fixes, the drip schedule — and the human runs it.
  3. Publish handoff: approved exports → scheduling-and-queue → WoopSocial (upload media → attach → validate → create post; per-platform required fields validate atomically — see woopsocial.md). OpusClip's own scheduler exists; the WoopSocial split (10 platforms, validation, one queue; measurement stays in the platforms' native analytics) is a stated stack choice.

Credit mechanics (the billing trap — attribute)

1 credit = 1 minute of SOURCE video processed, regardless of clips out. +1 credit per direct X post (refunded on failure). Trim sources before upload; don't process what you won't clip; map monthly source-minutes to the tier first (weekly 60-min show ≈ 240 min/mo). Reported patterns (Trustpilot, attributed): failed processing eating credits before refunds; projects vanishing after subscription ends — download exports promptly; cancellation friction — manage billing deliberately.

Tiers (≈, conflicts noted — verify in-app)

Free = evaluation only: 60 min/mo, watermark, 3-day clip expiry, 9:16 only, no score, no editor. Starter ≈ $15/mo: 150 min, score, basic posting. Pro ≈ $29/mo (~$14.50 annual): 300 min, all aspect ratios, scheduler, AI B-roll, XML export to Premiere/Resolve, team workspace — the real production tier. Business: custom; API access. Editor gating conflicts across 2026 sources (Starter-included vs Pro-gated) — confirm in-app before subscribing.

Feature + reliability cautions

The Virality Score (0–99) is a proprietary prediction: the one independent test (BIGVU) found ~40% of clips discarded and the score regularly mispredicting — triage only, never truth, never quoted as a metric. Captions ~97% vendor-claimed — QA names/jargon. No burned-in-subtitle sources (overlap). Prefer auto-music OFF → licensed or platform-native audio at publish. Reported processing hangs ("stuck at 96%"), no public status page or SLA — build slack into deadlines. SOC 2 Type II.

Hard lines

Human reviews every clip (stands alone / fair to the speaker / serves the audience) before publish · no unreviewed auto-post pipelines · no out-of-context clips or manufactured endorsements (deception/defamation; consent + likeness rules) · the score is never a guarantee or an auto-delete threshold · never state a tier/gate/capability as fact without in-app verification · WoopSocial does not clip, edit, or score video.