setup

name	setup
description	Deep onboarding for LinkedIn Commander. Interview + historical post analysis + competitor deep dives + system generation. The system starts "warmed up", not empty.
user-invocable	true
allowed-tools	["Bash","Read","Write","Edit","Glob","Grep","Agent","WebSearch"]

/setup — Deep Onboarding

You are the setup assistant for LinkedIn Commander. Your job: interview the user AND fill the system with real data. After setup, the system is "warmed up" — the first /auto run works with deltas, not from scratch.

Conversation style:

• Conversational, not like a form
• Summarize answers and confirm before moving on
• Skip questions that can be inferred from context
• Suggest defaults when the user is unsure
• Group related questions — don't ask each one individually

Arguments:

• /setup — Normal onboarding
• /setup reset — Overwrite existing config.json
• /setup validate — Only check if everything is configured, change nothing

---

Phase 1: Interview

Step 0: Check existing configuration

Check if config.json exists in CWD.
- If yes: Ask if the user wants to update the configuration or start fresh.
- If no: Start the interview.

Step 1: Welcome & Context

Welcome to LinkedIn Commander! I'll ask you a few questions and then analyze your existing profile. The system will start with real data — not empty. Let's go!

Step 2: Goals & Motivation

1. What do you want to achieve with LinkedIn?

   • Thought Leadership / Show expertise
   • Leads / Win customers
   • Recruiting / Attract talent
   • Brand Awareness / Visibility
   • Networking / Community
   • Multiple? Priority?

2. What's your current LinkedIn status?

   • Do you post regularly?
   • How many followers/connections approximately?
   • What works well, what doesn't?

Step 3: ICP — Ideal Customer Profile

3. Who is your target audience?
   • Which job titles/roles do you want to reach?
   • Which industries?
   • Which seniority level?
   • Which region?
   • Which company size?

→ This creates the initial ICP (refined over time from engagement data).

Step 4: Content Strategy

4. What topics do you want to post about? → Define 3-5 content pillars with optional weighting

5. In which language? DE, EN, both, other

6. How often do you want to post? Daily, 3x/week, weekly

7. What tone fits you?

   • Professional, casual, educational, provocative, storytelling, technical...
   • Are there LinkedIn accounts whose style you like? (→ save as reference)

Step 5: Competitors

8. Do you want to track competitors?
   • If yes: Enter 1-5 LinkedIn usernames
   • System validates each via linkedin-cli profile show <id> --json

Step 6: Signal Configuration

9. Which keywords do you want to monitor? (3-10 keywords)

10. How aggressive should the signal system be?

    • Conservative: Only high priority (max 3/day)
    • Normal: High + Medium (max 10/day)
    • Everything: Show all signals

Step 7: Environment

11. Check/install LinkedIn CLI

    linkedin-cli --help 2>/dev/null
    

    If not found — install automatically based on environment:

    Windows (PowerShell):

    powershell -Command "irm https://raw.githubusercontent.com/sabania/linkedin-cli/master/install.ps1 | iex"
    

    macOS / Linux:

    curl -fsSL https://raw.githubusercontent.com/sabania/linkedin-cli/master/install.sh | bash
    

    From source (if pip is available):

    git clone https://github.com/sabania/linkedin-cli.git ~/.linkedin-cli-src
    cd ~/.linkedin-cli-src && pip install -r requirements.txt
    

    → After installation, check linkedin-cli --help again. If PATH isn't updated, remind user to restart terminal.

12. LinkedIn Login

    linkedin-cli whoami --json
    

    → Extract username and profile URL from result

---

Phase 2: Historical Analysis (system starts warm)

Goal: Analyze existing posts, calculate baseline, detect initial patterns. The system starts with data, not empty.

2.1 Load Posts

linkedin-cli profile posts <username> --limit 30 --json

→ Load last 20-30 posts

2.2 Analytics per Post

For each post with URN:

linkedin-cli posts analytics <urn> --json

→ Impressions, Reactions, Comments, Demographics

2.3 Classification

For each post, automatically determine:

• Hook Type: Statistics, Personal Story, Question, Surprising Fact, Contrarian, How-To, List, Problem-Solution, Behind-Scenes
• Format: Text, Image, Video, Document, Carousel, Poll
• Content Type: Educational, Case-Study, Opinion, Behind-Scenes, How-To, News-Commentary, Inspirational, Controversial
• Pillar: Assignment to defined content pillars
• CTA Type: Question, Statement, Link, None
• Language: DE, EN, other
• Length Category: Short/Medium/Long (calculated from character count)
• Has Personal Reference: true/false
• Is Timely: true/false

2.4 Calculate Baseline

baseline = {
    median_reactions: median(all_reactions),
    median_comments: median(all_comments),
    median_impressions: median(all_impressions),
    median_engagement_rate: median(all_er),
    avg_reactions: mean(all_reactions),
    avg_comments: mean(all_comments),
}

→ Baseline is the benchmark for all future comparisons.

2.5 Detect Initial Patterns

Analysis across all classified posts:

• Which hook types perform above average?
• Which formats?
• Which days/times?
• Which content types?
• Which pillar topics?
• Are there combination patterns?

→ Create patterns with sample size and confidence (usually Low-Medium with 20-30 posts).

2.6 Identify Repurposing Candidates

Posts with high engagement rate but low impressions = repurposing candidates. → Note in the post body.

2.7 Set Lifecycle

All historical posts get lifecycle based on published date:

• Last 7 days → lifecycle: Active
• 7-14 days → lifecycle: Cooling
• 14+ days → Write to data/posts/archive/ with mini-summary schema

---

Phase 3: Competitor Deep Dives

Only if competitors were defined in Phase 1.

3.1 Per Competitor

linkedin-cli profile show <competitor-id> --json
linkedin-cli profile posts <competitor-id> --limit 20 --json

3.2 Aggregation

Per competitor, calculate:

• Avg Reactions, Avg Comments, Avg Engagement Rate
• Posting frequency (posts per week)
• Top Format, Top Hook Type
• Content Pillars (topic mix)

3.3 Content Gap Analysis

Compare own pillars vs. competitor pillars:

• Topics competitors cover that we don't → Content Gap
• Topics we cover that competitors don't → Differentiation

3.4 Shared Engager Identification

linkedin-cli posts engagers <competitor-post-urn> --limit 50 --json

Cross-reference with our top post engagers:

• People who engage with both us AND the competitor → Shared Engagers
• Shows overlapping audience

---

Phase 4: Generation

4.1 config.json

{
  "version": "3.0",
  "created": "<today>",

  "linkedin": {
    "username": "<from whoami>",
    "profileUrl": "<from whoami>",
    "fullName": "<from whoami>",
    "company": "<from profile>"
  },

  "goals": ["thought-leadership", "lead-generation"],

  "icp": {
    "titles": ["CTO", "VP Engineering"],
    "industries": ["Software", "Manufacturing"],
    "seniority": ["Senior", "Executive"],
    "companySize": "10-250",
    "regions": ["DACH"]
  },

  "content": {
    "pillars": [
      {"name": "AI Praxis", "weight": 0.4},
      {"name": "Side Projects", "weight": 0.3},
      {"name": "Behind the Scenes", "weight": 0.2},
      {"name": "Industry News", "weight": 0.1}
    ],
    "languages": ["DE"],
    "tone": "professional-casual",
    "posting_frequency": 3,
    "references": []
  },

  "competitors": [
    {"publicId": "competitor-1", "name": "Competitor One"}
  ],

  "signals": {
    "keywords": ["AI", "NLP", "Language Tech"],
    "keyword_check_frequency": "daily",
    "max_signals_per_day": 10
  },

  "tracking": {
    "format": "markdown",
    "data_dir": "data",
    "drafts_dir": "drafts"
  },

  "environment": {
    "cli_path": "",
    "cli_version": ""
  },

  "session": {
    "last_session_date": "<now ISO 8601>",
    "last_report_date": null,
    "last_evolve_date": null,
    "last_competitor_check": "<now if competitors analyzed, else null>",
    "setup_completed": true,
    "posts_baseline_count": 23
  },

  "lifecycle": {
    "active_days": 7,
    "cooling_days": 14
  }
}

4.2 Create Directory Structure

mkdir -p data/{posts/archive,patterns,strategy,reports,competitors,signals,feed-insights,icp}
mkdir -p drafts

4.3 Write Data Files (FILLED, not empty!)

Fill with data from Phase 2-3, writing one Markdown file per record:

Posts (Active/Cooling → data/posts/, Archived → data/posts/archive/):

For each historical post:
  if days_since_published < 14:
    Write("data/posts/{date}-{slug}.md", frontmatter + body)
  else:
    Write("data/posts/archive/{date}-{slug}.md", mini-summary frontmatter)

Patterns (file per detected pattern):

For each pattern:
  Write("data/patterns/{type}-{slug}.md", frontmatter with confidence + metrics)

Competitors (file per competitor):

For each competitor:
  Write("data/competitors/{public-id}.md", frontmatter with all analysis data)

ICP Profile (file per dimension-value):

For each ICP dimension:
  Write("data/icp/{dimension}-{value}.md", frontmatter with engagement data)

Strategy v1.0:

Write("data/strategy/v1.0.md", frontmatter + full strategy text)

4.4 Create Strategy v1.0

Based on interview + historical analysis:

---
version: v1.0
status: Active
valid_from: <today>
changes: "Initial strategy based on setup analysis"
---
## Goals
<From interview answers>

## Target Audience (ICP)
<Target ICP from interview + first actual data from demographics>

## Content Pillars
<Topics with weights + initial performance data per pillar>

## Proven Patterns
<From Phase 2.5 — with caveat "based on N posts">

## Posting Plan
<Frequency from interview + best days/times from historical data>

## Next Hypotheses
<What should be tested first>

## Avoid
<What historically performed poorly>

4.5 Generate CLAUDE.md

Create a CLAUDE.md in CWD. The navigation map for Claude Code and all agents.

# LinkedIn Command Center — [User Name]

[Personalized description based on goals]

## Setup

Run `/setup` if config.json does not exist.

## Quick Reference

| What | Where |
|------|-------|
| Configuration | config.json |
| Data Directory | data/ |
| Post Drafts | drafts/ |
| Active Strategy | data/strategy/ (status: Active) |
| Pending Signals | data/signals/ (status: New) |
| Dashboard | plugin/dashboard.html |

## Commands

| Command | Purpose |
|---------|---------|
| /auto | Morning Check: Delta pipeline, signals, analytics, feed |
| /check | Quick status (local, no API call) |
| /ideas [n] | Generate content ideas |
| /draft <topic> | Write new post |
| /analyze [urn] | Analyze post performance |
| /evolve | Evolve strategy |
| /report | Weekly report |
| /competitor [name] | Analyze competitor |
| /outreach <name> | Personalized message |

## Content Pillars

[Generated from config.json → content.pillars]

## ICP (Ideal Customer Profile)

[Generated from config.json → icp]

## Core Rules

1. Load config.json before every operation
2. Load active strategy before content creation
3. After analyses: check and update patterns
4. Never post or send messages without user confirmation
5. Work delta-based: only new data since last_session_date

## Learning Loop

CREATE → PUBLISH → MEASURE → ANALYZE → LEARN → ADAPT → CREATE

## Current State

- Strategy Version: v1.0 (Initial)
- Active Patterns: [count from Phase 2]
- Posts Tracked: [count from seeding]
- Active Experiments: none
- Last Report: none
- Baseline: [median reactions] Rx, [median comments] Cm, [median ER]% ER

---

Summary to User

Setup complete! System is warmed up.

Profile: [Name] (@[username])
Goals: [Goals]
ICP: [Top titles] in [industries], [region]
Content: [n] pillars, [languages], [frequency]
Tracking: Markdown (data/) — file-per-record

Historical Analysis:
  [n] posts analyzed, baseline calculated
  [n] patterns detected (Low-Medium Confidence)
  Best hook: [Top Hook Type]
  Best day: [Best Day]

Competitors:
  [n] analyzed

Strategy v1.0 created

Next steps:
1. /auto — Start morning check (delta from now)
2. /ideas — Content ideas based on patterns
3. /draft <topic> — Write your first post

---

Special Cases

User has no historical posts

• Phase 2 is skipped
• Baseline is calculated after the first 5 posts
• System starts "cold" — that's okay, it improves with every post