Agent Skills › leamsigc/ShortsGenerator › ig-research-analyze

ig-research-analyze

GitHub

该技能用于社交媒体内容心理学分析，读取原始帖子和转录数据，识别触发模式并生成内容角度建议。同时执行视频创意流水线，基于热门视频生成脚本、标题及元数据报告。

skills/ig-research-analyze/SKILL.md leamsigc/ShortsGenerator

Trigger Scenarios

需要分析社交媒体帖子的心理触发因素请求生成基于数据的视频创意和脚本

Install

npx skills add leamsigc/ShortsGenerator --skill ig-research-analyze -g -y

More Options

Use without installing

npx skills use leamsigc/ShortsGenerator@ig-research-analyze

指定 Agent (Claude Code)

npx skills add leamsigc/ShortsGenerator --skill ig-research-analyze -a claude-code -g -y

安装 repo 全部 skill

npx skills add leamsigc/ShortsGenerator --all -g -y

预览 repo 内 skill

npx skills add leamsigc/ShortsGenerator --list

SKILL.md

Frontmatter

{
    "name": "ig-research-analyze",
    "version": "2.0.0",
    "description": "Content psychology analysis state. Reads raw-posts.json and transcripts,\nanalyzes top 6 posts through a psychological lens, identifies trigger\npatterns, and generates 5 tailored content angle recommendations.\nAlso runs the video-ideas pipeline (shared video-idea-research playbook):\ntop 10 reels\/videos by views in the window → outlines → 5 video ideas × 3\nscripts (viral \/ storytelling \/ inspiration) + YouTube metadata\n(title, description, tags, keywords) → ideas-report.md + REPORT.html.\n",
    "allowed-tools": [
        "Bash",
        "Read",
        "Write",
        "Skill"
    ]
}

Social Research — Analyze State

After scraping and transcription, this state reads the data and provides content psychology analysis, then the video-ideas pipeline.

Preconditions

Raw data exists: .ig-research/projects/<project-name>/raw-posts.json
Transcripts exist in .ig-research/projects/<project-name>/transcripts/
Report exists in .ig-research/projects/<project-name>/report.html

Steps

1. Read report data

cat .ig-research/projects/<project-name>/report.html

2. Read raw posts

node -e "const d=require('./.ig-research/projects/<project-name>/raw-posts.json'); console.log(JSON.stringify(d.posts.slice(0,6),null,2))"

3. For each of the top 6 posts, identify psychological triggers

Analyze against these frameworks:

Trigger	Description	Signal
Curiosity gaps / open loops	Creates info gap the brain wants to close	"Here's why...", "The one thing..."
Identity signaling	Makes viewer feel like "that person"	"This is for people who..."
Social proof / tribal belonging	Shows community validation	High engagement numbers, "join"
Loss aversion / FOMO	Fear of missing out	"Don't make this mistake", "before it's too late"
Authority / credibility	Positions creator as expert	Named frameworks, credentials
Pattern interrupt / contrarian	Goes against expectation	"Everything you know about X is wrong"
Reciprocity / value-first	Gives before asking	Free frameworks, actionable tips
Anchoring / specificity	Specific numbers feel researched	"3 things", "7 strategies"

4. Give 5 content angle recommendations

For each angle: psychological trigger, hook example, why it works in the niche.

5. Summarize findings

Write a brief summary: how many posts scraped, top 3-5 performers, common patterns.

6. Open the report

xdg-open .ig-research/projects/<project-name>/report.html

(Use open on Mac, start on Windows)

7. Video ideas pipeline (`skill video-idea-research`)

When the user wants video ideas for a topic, run the shared playbook:

Top 10 by views in the window — rank the scraped reels (plus a fresh hashtag/explore scrape for the topic when needed). Real play counts only; views: null + the proxy metric when the count isn't visible.
Outlines — hook = first caption line or transcript opening; beats and CTA from the Whisper transcript (scripts/transcribe.sh); format/duration from the reel itself.
5 ideas, each grounded in ≥2 top reels with their real numbers.

3 scripts per idea — viral (viral_shorts), storytelling (storytelling), inspiration (motivational); TTS-ready, [HOOK]/[CTA] labels only. Optional backend seeding:

curl -s -X POST http://localhost:8080/api/script \
  -H "Content-Type: application/json" \
  -d '{"videoSubject":"<idea title>","extraPrompt":"<angle + outline context>","aiModel":"g4f","scriptTemplate":"viral_shorts","scriptLength":"standard"}'

YouTube metadata per idea (title via the youtube-title rubric, description, tags, keywords).

Save video-ideas.json and render the report:

node "$CLAUDE_SKILL_ROOT/scripts/build-ideas-report.js" <path-to-video-ideas.json>
# fallback: node ./skills/video-idea-research/scripts/build-ideas-report.js <path>