Spot Review Trends

Spot review trends

Show how a batch of reviews changes over time: how many arrive per period, whether ratings are climbing or slipping, how the sentiment mix shifts, and which themes are rising, steady, or fading. Every trend is backed by the real per-period counts - no invented movement, no guessed causes. The reviews can come from a Feedspace account through the Feedspace MCP, a CSV, or pasted text.

The full method lives in [references/trends-method.md](references/trends-method.md). Follow it exactly. It runs entirely in this session and needs no API key; pulling reviews from the Feedspace MCP is a separate, read-only connection.

Steps

1. Get the reviews - with their dates. Trends need a date on every review.

• Feedspace MCP: first check whether it is already connected (are Feedspace tools like

list_workspaces available?). If yes, use them. The MCP has no date filter, so to build a timeline you fetch the reviews and read each one's own date (reviewed_at, falling back to created_at). See [references/mcp-source.md](references/mcp-source.md) for the fetching strategy and the import-date caveat. Only if the tools are not available, point the user to setup there.

• Otherwise read a CSV that has a date column (see [references/csv-format.md](references/csv-format.md))

or ask the user to paste reviews with dates. Reviews with no date cannot be placed on a timeline.

1. Snapshot, then choose the period. Give a 1 to 3 line snapshot: how many dated reviews (and how

many undated ones were dropped), the date span covered, and whether it is one business or several mixed. Pick a bucket size that fits the span (weekly, monthly, or quarterly) and say which. If the span is too short or there are too few dated reviews for a reliable trend, say so and offer a plain overview instead.

1. Run the trends by following [references/trends-method.md](references/trends-method.md): volume

over time, average rating over time, sentiment mix over time, and theme movements - each with a per-period count you can see. Or run just the one the user asked for.

1. Offer one sensible next step, then stop. Zoom into a single period or a single theme's

trajectory, or narrow by review type or rating and re-run. If the reviews came from the Feedspace MCP, offer to pull a specific segment over the live connection.

Guardrails

• Every trend claim must be backed by the actual per-period numbers. Show the small period table so

the movement is visible, not asserted.

• Do not guess causes. A volume spike from an import batch is visible (many reviews sharing one import

date) - flag that. A marketing or seasonal cause is not visible in the data - do not claim it.

• Small samples are noise, not trends. If a period has only a handful of reviews, label the movement

as noise and say so.

• Prefer the review's own date (reviewed_at) over its import date (created_at), and say which you

used. Imports can cluster on one date and distort volume - flag it when it happens.

• If the workspace mixes several businesses, cluster the trends or say so. Detect the reviews'

language and write in it.

• The Feedspace MCP path is read-only. Never modify or delete anything.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Feedspace
• Source: Feedspace/Feedspace-Cookbook
• License: MIT
• Homepage: https://feedspace.io

Install and usage instructions live in the source repository linked above.