Skill · introduced Week 5
Weekly Report
Generates your recurring report from real data — same structure, same voice, every week — so a 2-hour job becomes a 5-minute review.
Difficulty: intermediate ReportingStatus updatesTurning data into narrative
Recurring reports are the perfect Skill: the format never changes, only the data does. Build it once and Claude writes the draft from your spreadsheet/connector while you keep judgement for the “so what”.
What you’ll build
A Skill that takes this week’s data (a file, or a live connector) and returns a finished draft:
- Headline — the one number that matters and whether it’s good.
- What changed — week-on-week movement, explained in plain language.
- Why — the likely drivers, flagged as fact vs. hypothesis.
- What we’re doing — next actions, pulled from context.
How to build it
- Decide the data source: a dropped-in
.xlsx/.csv, or a connector (e.g. a CRM via MCP). - Capture your house format in
report-template.mdand a realexample-report.md. - Encode the rules in
SKILL.md— especially how to label certainty.
The SKILL.md
---
name: weekly-report
description: Draft the recurring weekly report from the latest data, in our house format.
---
When asked to draft the weekly report:
1. Read the latest data (attached file, or the configured connector).
2. Compute week-on-week deltas for each headline metric.
3. Write to report-template.md. Lead with the single most important number.
4. Separate FACT (in the data) from HYPOTHESIS (your inference) — label inferences explicitly.
5. Pull "next actions" from recent emails/notes if available; otherwise leave a clear placeholder.
6. Keep it skimmable: headline, then bullets. Match the tone of example-report.md.
Watch-outs
- Label inference. The fastest way to lose trust is a confident “why” that’s a guess. The fact/hypothesis split is the Discernment habit that makes the report safe to forward.
- Don’t let it invent numbers — if the data’s missing, it says so.
- Keep the template in version control; when the format changes, the Skill changes with it.
Become the AI-powered one in your office
Eight weeks. A real project every week. A personal AI assistant that does real work for you — taught live.