LinkedIn Analytics — describe honestly, then refuse to over-conclude
Analyzes your own exported LinkedIn post data to describe performance honestly and test patterns without over-concluding. It reports medians and outlier bands, runs permutation tests against a null, and sizes real experiments, refusing to draw conclusions from fewer than 10 posts.
✨ What it does
- Reports median and MAD instead of mean and standard deviation, with Tukey percentile bands and IQR breakout thresholds.
- Runs a four-gate permutation test (5 posts in/out, 15% median difference, 90% of 2000 shuffles, multiple-comparisons acc
- Sizes a two-arm experiment, names confounds to hold constant, and writes a falsification condition.
- Refuses to conclude anything below 10 posts; exits with descriptive-only mode.
- Classifies posts as BREAKOUT or DUD based on statistical thresholds.
- Reports confidence levels for any external claims (official, third-party, folklore).
🎯 When to use it
- When you want to understand which of your LinkedIn posts performed well and why.
- When you suspect a pattern (e.g., carousels do better) and need to test if it's real.
- When your reach dropped and you want to investigate without jumping to conclusions.
- When you're planning a posting experiment and need to determine feasibility and sample size.
🚀 How to use
Trigger by asking to analyze your LinkedIn posts or mentioning 'why did my reach drop' or 'do carousels do better for me'. Provide your exported post data (CSV or JSON) from LinkedIn Analytics or your data copy. Run the scripts in order: describe, test, plan. Example prompts:
Why did my reach drop last month? Analyze my posts.csv.
I think carousels perform better for me. Test that pattern on my post data.
📄 Output: Produces descriptive statistics, pattern test results, and an experiment plan (or a refusal) as human-readable reports.
📦 Add this skill to Claude Code
# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills
# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/marketing/linkedin/skills/linkedin-analytics .claude/skills/linkedin-analyticsSkill source: marketing/linkedin/skills/linkedin-analytics/SKILL.md
⚠️ Good to know
Requires your own exported LinkedIn data; does not fetch or scrape data, and cannot conclude from fewer than 10 posts.
❓ FAQ
What does the skill do if I have fewer than 10 posts?
It runs in descriptive-only mode, reporting medians and outliers but refusing to draw any conclusions about patterns.
Can I benchmark my performance against someone else's numbers?
No, the skill explicitly prohibits benchmarking against others' numbers because denominators and audiences differ.
What is the most common honest result from the pattern test?
The most common honest answer is that nothing survived the statistical gates, and that is a real finding to report as is.
🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.