m29

Sisi data & orkestrasi dari konten sosial: riset tren/kompetitor, insight audiens, social listening, analitik performa, A/B variant, best-time-to-post, engagement handler, dan *full content pipeline

SKILL.md

GitHub

m29 — Content Research, Analytics & Pipeline (NEW in v4.1)

Sisi data & orkestrasi dari konten sosial: riset tren/kompetitor, insight audiens, social listening, analitik performa, A/B variant, best-time-to-post, engagement handler, dan full content pipeline (research → draft → visual → schedule → analyze). Opsional. Otak strategi di belakang m27 (produksi) & m28 (tulisan).

Jujur soal data: analitik & "best time" butuh data nyata dari platform/akun operator (API/export). Tanpa itu = heuristik & benchmark umum, ditandai sebagai estimasi — bukan angka palsu.


1. Trend & competitor research

- Trending topic: cari di X/IG/LinkedIn/TikTok + Google Trends per niche & region
- Competitor teardown: format yang nempel, frekuensi, hook, gaya, gap yang bisa diisi
- Pain points: scrape komentar/forum/review → tema keluhan berulang = bahan konten

Sumber via web fetch operator / API. Untuk riset mendalam bersitasi → handoff m22. Output: daftar peluang konten ter-ranking by relevansi × momentum.

2. Audience insight & social listening

# analisis komentar/engagement → tema + sentimen (real data dari export/API operator)
from content import analyze_comments
insight = analyze_comments(comments)   # → {themes, sentiment, questions, objections, quotes}

Real-time listening: pantau keyword/brand/kompetitor di sosial → peluang (pertanyaan belum terjawab, tren naik, krisis). Sambung alert (m4/m14). Pertanyaan audiens = ide konten + objection buat di-address.

3. Performance analyzer

from content import analyze_performance
report = analyze_performance(posts)    # butuh metrik real: impressions, engagement, CTR, saves
# → top/bottom performer, pola (format/jam/hook yang menang), rekomendasi konkret

Tanpa data real → kasih kerangka apa yang harus diukur, jangan ngarang angka. Pola menang → reinforce; yang flat → drop. Kaitkan ke self-improving loop (x4): pola kebukti → instinct konten.

4. A/B testing & best-time

from content import ab_variants, best_time
variants = ab_variants("caption asli", n=3, vary=["hook","cta","length"])  # 1 variabel/uji
slots = best_time(audience_tz="Asia/Jakarta", history=post_history)         # data > tebakan

A/B: ubah satu elemen per uji (hook ATAU CTA, jangan dua) — biar tau mana yang ngefek (prinsip sama x5 eval). Best-time dari data audiens sendiri > benchmark generik; tandai kalau cuma benchmark.

5. Engagement response handler

- Draft balasan komentar (match brand voice m20) — POSITIF cepat, KRITIK hati-hati
- Eskalasi: keluhan serius / legal / krisis → flag ke operator, JANGAN auto-reply
- Human-in-the-loop: draft dulu → approval → kirim (jangan auto-post balasan sensitif)

6. Full content pipeline (orkestrasi)

RESEARCH (m29) → DRAFT (m28+m27) → HUMANIZE (m20) → VISUAL (m18) →
SCHEDULE (m4/m17 cron) → PUBLISH (cross-platform, R9 gate) → ANALYZE (m29) → loop balik
from content import repurpose
pieces = repurpose(long_form_text, into=["x_thread","ig_carousel","reels_script","linkedin_post"])
# 1 long-form → banyak platform, tiap format di-adapt (bukan copy-paste)
  • Cross-platform publisher: adapt (m27) lalu publish ke banyak platform via API (m6).
  • Human-in-the-loop: generate draft → minta approval operator → refine. Default buat outward content.
  • Scheduled agent: konten rutin via cron (m4/m17). Publish = aksi outward → R9 gate.
  • Self-improving loop: performa (§3) → pelajaran (x4) → strategi konten makin tajam.
  • Multi-agent: riset / tulis / desain bisa di-split jadi lane swarm (m17) — executor yang publish, bukan lane riset.

Catatan

  • Analitik butuh data real — tanpa itu, heuristik bertanda "estimasi".
  • Publish & auto-reply = outward → human-in-the-loop / R9 gate. Jangan auto-reply hal sensitif.
  • Combo: m29 + m27 (produksi) + m28 (copy) + m20 (humanize) + m18 (visual) + m22 (riset dalam) + m6/m4/m17 (publish/schedule) + x4/x5 (improve/eval).