---
name: showpilot-weekly-review
description: Read the week on a ShowPilot workspace — what moved, what visitors asked that nobody could answer, and which companies are warm. Use when someone asks how the site agent is doing, for a weekly or monthly review, or before a pipeline meeting.
---

# Weekly review

Five tool calls and a short written answer. Do not pad it.

## Steps

1. `whoami`. Confirm the workspace. If it warns the workspace was defaulted, stop and ask which one before reading anything else — the numbers belong to one tenant.
2. `weekly_digest`. Default is the last seven days; pass `from` and `to` for any other window.
3. `find_knowledge_gaps` with `status: "open"`. These are the questions the agent could not answer.
4. `find_sales_ready_accounts`. Companies with recent engagement and real intent.
5. Write the review.

## Writing the review

Lead with the single largest mover from `metrics`, by size, not by whichever metric you happen to like. Say the number and the direction. If `change` is `null` the previous window was zero, so say "up from none" rather than inventing a percentage.

Then three sections, each at most four lines:

- **What moved.** The two or three metrics that actually changed. Ignore the flat ones.
- **What we could not answer.** Group the open questions yourself — they arrive unclustered because you group them better than a pattern does. Name the two or three themes and how many questions sit under each. Quote one real question per theme.
- **Who to call.** The top three accounts by intent. For each: the company, how many people, and whether the CRM already shows an open deal. An account with `hasOpenDeal: false` and high intent is the interesting one.

## Rules

- If `missingDays` is not empty, say so. The figures are partial and a reader who does not know that will read a gap as a decline.
- Never compare against a window the tool did not return. `comparedWith` is the only valid baseline.
- Questions come back with the people masked. Do not speculate about who asked.
- No recommendations the data does not support. "Nobody asked about pricing this week" is a finding; "we should change our pricing page" is not.