Claude Code drafts client reports from live data

John Aspinall runs weekly client-performance reporting through Claude Code skills linked to business context, formatting conventions and connected tracking data. A report skill applies his analysis framework and prepares a narrative in the style each client expects. He reviews and edits the draft before relying on it.

The implementation grows through iterations rather than one long instruction file. Aspinall says his first oversized skill broke repeatedly and that a simpler version remained usable. He also identifies a concrete failure: a report skill produced good output for top accounts but nonsense for accounts with incomplete data, which he discovered only after a client saw it.

That failure explains why his later process tests messy inputs and revisits scheduled outputs instead of assuming that a successful clean example proves the workflow. Report timings and weekly savings are his own estimates, not independently measured results. The useful copyable pieces are a linked output template, an explicit analysis framework and a missing-data policy checked against weak source data. The source does not name a client, expose the tracking system or establish an audited accuracy rate; the entry therefore keeps the outcome at recurring report drafting with documented limitations.

Prompt
Suggested prompt (written by Agents In Use, not the author's)

Source: Blog · John Aspinall