Your LinkedIn analytics, offline and yours. A single file dashboard, built with Claude, that merges the two halves of your data:

• Quantitative, straight from LinkedIn's official .xlsx export (impressions, interactions, follower growth, daily series, demographics).
• Qualitative, the part LinkedIn does not export (per post hook, pillar, format, CTA, saves, profile visits, followers gained, and your own reading of what worked), harvested with Claude in the browser.

You drop in two inputs, run one Python script, and reopen the HTML. No server, no build step, no framework. It opens straight from file:// and works fully offline.

Live demo (real numbers): https://marcogalluccio.com/claude-linkedin-dashboard/demo/ Landing page: https://marcogalluccio.com/claude-linkedin-dashboard/

Paste this to Claude (Claude Code, Cowork, or claude.ai with web access). It reads the instructions and builds the dashboard with you, no download:

read https://marcogalluccio.com/claude-linkedin-dashboard/skill.md and help me create my LinkedIn dashboard

There is no zero-install path here: building the dashboard runs a small Python script, so you need Claude that can run code. Two ways:

1. Claude Code (terminal, recommended). Clone this repo, drop your .xlsx in data/, and ask Claude to harvest the qualitative layer and run ingest.py. Full power, including the qualitative browser harvest. Start here if you can.
2. Cowork (Claude on claude.ai, no terminal). Install the packaged skill (download the zip from Releases) as a Skill, upload your .xlsx, and Claude builds the dashboard in the sandbox and hands you the finished file. Quantitative dashboard; the qualitative layer is a Claude Code extra.

In italiano: serve far girare un piccolo script Python, quindi si usa Claude Code (consigliato) o Cowork, non la chat semplice. Con Claude Code cloni il repo; con Cowork installi la skill dalle Releases e carichi l'.xlsx.

• Headline KPIs (impressions, reach, interactions, followers, top save rate).
• Growth over time, with a shared horizon zoom (All, 180d, 90d, 30d, 7d, custom dates) and Monthly / Weekly (Mon to Sun) / Daily granularity, for both the impressions+interactions chart and the follower chart (cumulative or new per period).
• Post ranking (bars), impressions by pillar (donut), reach vs engagement (scatter, bubble size = saves).
• An enriched top 10 (saves, profile visits, followers, sends, format, CTA).
• The full post table, sortable and searchable.
• Follower demographics (location, seniority, company size, industry).
• Your "quick reads", the qualitative insights of the period.

data/*.xlsx          (quantitative, official LinkedIn export, dated)
qualitative/*.md     (qualitative, one ```json block, dated)
        │
        ▼  python3 ingest.py   (joins the two by post URL)
data.js              (generated: window.DATA = { meta, kpis, daily, posts, demographics, insights })
        │
        ▼  <script src="data.js">
dashboard.html       (stable template + render code; never edited on update)

The HTML is a template: it holds the structure and all the rendering. It never changes when you update. Everything dynamic comes from data.js, which ingest.py regenerates. The two data layers are joined by the post URL, so the join is exact even when several posts share a date.

1. Quantitative. On LinkedIn, export your content performance to .xlsx and drop it in data/YYYY-MM-DD-linkedin-export.xlsx.
2. Qualitative. Run python3 cowork_delta.py prompt, paste the printed prompt into Claude in the browser (Cowork) on your LinkedIn analytics page; it writes the result to qualitative/incoming/.
3. Merge. Run python3 cowork_delta.py merge. It splices the new posts into the qualitative file and runs ingest.py.
4. Open dashboard.html.

You can also run python3 ingest.py directly when only the .xlsx changed (unlabeled posts show as "da etichettare"). The two skills in skills/ document the data contracts and the harvest prompt.

• Python 3 with openpyxl (pip install openpyxl).
• A modern browser. That is all.

.
├── README.md
├── LICENSE
├── dashboard.html         # template + render code, reads window.DATA
├── data.js                # GENERATED, do not edit by hand (ships with a real demo dataset)
├── ingest.py              # the engine: xlsx + qualitative -> data.js
├── cowork_delta.py        # delta+splice helper: `prompt` / `merge`
├── demo/                  # the live demo served on Vercel
├── skills/
│   ├── ingest/SKILL.md
│   └── qualitative-cowork/SKILL.md
└── cowork/                # the packaged skill for claude.ai / Cowork (also in Releases as a zip)

data.js ships with a real dataset (one year of LinkedIn activity) so the dashboard is alive the moment you open it. Replace it with your own by running the pipeline on your export.

MIT. Built with Claude by Marco Galluccio.