short-form-research

これはレビュー時に保存された抜粋です。完全で最新の内容は外部ソースを確認してください。

---
name: short-form-research
description: "Discovers what's working right now on short-form video platforms (TikTok, Instagram Reels, YouTube Shorts; X video and LinkedIn video by opt-in) for a given topic and market. Produces a per-platform best-practice catalog at .agents/skill-artifacts/research/short-form-research/[slug].md that short-form-brief consumes. Not for long-form video (parked) or static visual (use design-brief). For audience research, see icp-research; for campaign planning, see campaign-plan."
argument-hint: "[topic or angle]"
allowed-tools: Read Grep Glob Bash WebSearch WebFetch Write
license: MIT
metadata:
  author: hungv47
  version: "1.0.0"
  budget: deep
  estimated-cost: "$3-6 (default 3 platforms) / $5-10 (--all)"
promptSignals:
  phrases:
    - "what's working on tiktok"
    - "short-form trends"
    - "reels research"
    - "shorts research"
    - "tiktok patterns"
    - "viral hook research"
  allOf:
    - [short-form, research]
    - [tiktok, hook]
    - [reels, pattern]
  anyOf:
    - "trending sound"
    - "hook archetype"
    - "platform research"
    - "what's hitting"
  noneOf:
    - "long-form"
    - "youtube long"
    - "podcast"
    - "blog post"
  minScore: 6
routing:
  intent-tags:
    - short-form-research
    - platform-pattern-mining
    - hook-research
    - content-research
  position: pipeline
  lifecycle: pipeline
  produces:
    - .agents/skill-artifacts/research/short-form-research/[slug].md
  consumes:
    - product-context.md
    - icp-research.md
  requires: []
  defers-to:
    - skill: icp-research
      when: "no audience context exists yet — research without ICP underperforms"
    - skill: market-research
      when: "user wants competitive landscape, not platform pattern mining"
  parallel-with:
    - market-research
  interactive: false
  estimated-complexity: heavy
---

# Short-Form Research — Orchestrator

*Pipeline skill — produces the per-platform best-practice catalog that `short-form-brief` consumes per asset.*

**Core Question:** "What's working right now on short-form for our topic and market — and which patterns should the next 30 days of briefs bet on?"

---

## Critical Gates — Read First

Non-negotiable constraints before dispatching any agent:

1. **No fabricated data.** Every claim, number, and pattern must trace to a source URL, video ID, or cited platform doc. Orphan claims fail critic rubric #1.
2. **Single market per artifact.** Multi-market campaigns re-run research per market — never mix VN and US findings in one artifact. Cultural patterns are not averageable.
3. **Hard cap on platforms.** Default 3 (TikTok + Reels + Shorts). X video and LinkedIn video are explicit opt-in via `--all` or `--platforms`. Maximum ever is 5. Cost discipline.
4. **Sample-size honesty.** Every per-platform section declares OK (n≥8), LOW_SAMPLE (n=3-7), or INSUFFICIENT_DATA (n<3). LOW_SAMPLE flags carry through to brief skill warnings. INSUFFICIENT_DATA means no pattern claims at all — only observed examples.
5. **Two freshness windows.** Trend signals refresh every 14d, warn at 30d. Platform mechanics refresh every 90d, warn at 180d. Frontmatter records `mechanics_sources_verified[]` — the actual doc URLs and their last-updated dates, not just the run timestamp.

---

## Philosophy

Research yields a **catalog of bets**, not a survey. The brief skill consumes specific recommendations ("TikTok hook should be credential-flash archetype in 0–1.5s — 8/12 in this sample") not generic advice ("strong hooks matter"). Specificity is the contract.

Trend signals decay fast — 14-day windows are deliberate. Platform mechanics change quarterly — 90-day windows match algorithm-update cadence. The two-window split prevents the "fresh date, stale truth" failure mode where one timestamp masks the other.

## Inputs / Output

**Inputs:** Topic (required); target platforms (default 3); market (single per artifact); audience hint or `research/icp-research.md` (warm-start); optional competitor h