---
name: weekly-client-report-drafter
description: "Draft a recurring client performance report from raw platform data — led by the client's actual success metric, reconciled against their source of truth, with anomalies flagged instead of smoothed over. Read-only, draft only; a named human reviews and sends every report. Use on the client's reporting cadence, for ad-hoc updates, or as a first draft before a client call."
metadata:
  version: "1.1.0"
  scope: "standalone-template"
---

> Standalone draft/audit template. Not validated for OLLIE execution.


# Weekly Client Report Drafter

## Who it's for

Any agency that owes clients a recurring performance update — weekly, biweekly, or monthly — and is tired of the same few hours every cycle spent turning raw exports into something a client can actually read.

## What problem it solves

Reporting is repetitive, time-consuming, and usually the first thing that slips when the team gets busy — which is exactly when clients start to worry. This skill gives an agent the job of turning raw platform data into a plain-language draft report, so a human's only job is to review, correct anything that needs it, and hit send.

## When to use it

- On a fixed recurring schedule (weekly or monthly, matching whatever cadence you've committed to a client)
- Whenever a client asks for an ad-hoc update outside the normal cycle
- As a first draft before a client call, so you're not building the deck from scratch the morning of

## Required inputs

- Raw exports or read-only access to whichever platform(s) the client's performance lives in (ad platform, analytics, CRM — whatever applies)
- The client's actual success metric (cost per lead, ROAS, purchase value, whatever the client actually cares about — not just whatever the platform shows by default)
- The prior period's report or key numbers, so the agent can show a real trend, not just a snapshot

## Agent instructions

You are drafting a recurring client performance report. You have READ-ONLY
access to the data sources listed. You may not send anything, post anything,
or contact the client — you are producing a draft for a human to review.

1. Pull the raw performance data for the reporting period, plus the prior
   period's numbers for comparison.
2. Identify the client's actual success metric and lead the report with that
   number, not a generic platform metric that doesn't map to what the client
   cares about.
3. Reconcile the numbers against the client's known source of truth if one
   exists (e.g., their CRM or revenue system) rather than reporting the
   platform's self-reported numbers uncritically. If you find a meaningful
   mismatch between sources, flag it clearly instead of picking one silently.
4. Write the report in plain language a non-technical client stakeholder
   could read: what happened, why (in one or two sentences, not a wall of
   jargon), and what's planned for next period.
5. If anything in the data looks anomalous, broken, or suspicious (a metric
   dropping to zero, a tracking gap, numbers that don't reconcile), call it
   out explicitly in the draft rather than smoothing over it or omitting it.
6. Do not send, publish, or share this draft anywhere. Produce the draft and
   stop — a human reviews and sends it.

## Expected output format

A structured draft: executive summary (2-3 sentences), a KPI table (this period vs. prior period), a plain-language explanation of what changed and why, a short section on any data anomalies found, and a brief next-period plan.

## Human approval checkpoint

No report reaches a client without a named human reading the exact draft and explicitly approving it. This is not a one-time setup step — every single report, every cycle, waits for that same explicit sign-off before it goes anywhere.

## Safety rules

- Read-only against every data source — no write access needed for this task at all
- Anomalies and data mismatches are always surfaced, never silently smoothed over or hidden
- The draft is written for the client's actual success metric, not whatever the platform happens to surface first

## What the agent must not do

- Send an email, message, or any client-facing communication
- Publish or share the draft anywhere outside the immediate review process
- Pick a number when two data sources disagree without flagging the disagreement
- Invent or estimate a figure it doesn't have real data for

---

Part of the Outloop Agency Skill Pack ([outloop.co](https://outloop.co)). To turn this from a draft-only workflow into a safe runtime workflow across client workspaces, use Outloop to approve the right access, block wrong-client use, and keep secrets out of the agent context.