---
name: feedback-synthesis-weekly
description: Pull feedback from Slack, Zendesk, and Intercom, cluster it into themes, and track the trend against last week. Use this skill when: Every Monday, 6:00 AM — Weekly sweep before the planning meeting.
---

# Synthesize the week's feedback into themes

Goal: synthesize the week's feedback into themes with counts and trend deltas.

Connected tools this skill uses: slack, zendesk, intercom, notion.

## When to use

- Every Monday, 6:00 AM — Weekly sweep before the planning meeting.
- Or when the user asks directly to synthesize the week's feedback into themes with counts and trend deltas.

## Method

Run the weekly feedback synthesis.

1. Pull every message from {{feedback_channel}} in Slack, plus new tickets tagged
   {{feedback_tag}} in Zendesk and Intercom, opened since the last run
   ({{last_run_date}}).
2. Cluster the raw feedback into named themes by topic, not by keyword; count
   how many distinct customers raised each theme.
3. For each theme, select one representative quote — exact wording, with
   speaker and source link. Never paraphrase a quote; if none fits cleanly,
   drop the quote and keep only the theme count.
4. Diff this week's theme counts against last week's snapshot in
   `reports/feedback/{{week}}.md`; note what's rising, falling, or new.
5. Validation: confirm every theme's quote is verbatim against its source
   and the link resolves; confirm every count reflects distinct customers,
   not raw messages.
6. Patch the feedback-themes view with this week's themes — don't rebuild
   it — and post the top 5, with counts and trend arrows, to
   {{delivery_channel}}.

Never file a Linear issue from a theme without my approval — surface
candidate opportunities for review instead.

## Theme clustering

Group raw feedback messages into named, topic-based themes instead of leaving them as an unsorted list.

## Method
1. Collect every raw feedback item pulled this run from slack, zendesk, and intercom.
2. Read each item and identify the underlying topic, not just the surface keyword — two messages using different words can be the same theme.
3. Group items into named themes; use a short, specific theme name (e.g. "CSV export missing columns", not "exports").
4. Check existing theme names in memory before creating a new one; reuse an existing name if the topic matches.
5. Count distinct customers per theme, not raw message count — one customer posting five times is one vote.
6. Merge near-duplicate themes; split a theme that actually covers two distinct topics.
7. Store the final theme list with customer counts for this run.

## Rules
- A theme needs at least (set on first run, then stored in memory) distinct customers to surface in the digest.
- Never invent a theme from a single ambiguous message — leave it unclustered instead.
- Theme names stay stable across weeks; renaming a theme requires a deliberate merge decision, not a fresh label each run.
- Keep the split between "feedback channel" and "support issue" as defined on first run.

## Output
A list of themes, each with: theme name, distinct customer count, and the list of source message/ticket IDs it was built from.

## Verbatim-quote policy

Attach one exact, sourced quote per theme — never a paraphrase standing in as a quote.

## Method
1. For each theme, scan its source messages for one line that best represents the theme.
2. Copy the quote exactly as written — same words, same spelling, same punctuation.
3. Attach the speaker's name (or handle) and a link back to the source message or ticket.
4. Confirm the link resolves before attaching it to the theme.
5. If no message in the theme reads cleanly as a standalone quote, do not force one — leave the theme without a quote.
6. Re-check the quote against the source text once more before delivery; a single altered word fails this check.

## Rules
- A quote is verbatim or it does not ship — no summarizing, no cleaning up grammar, no combining two messages into one quote.
- Every quote carries a source link; a quote without a resolvable link is dropped.
- One quote per theme maximum for the weekly digest, chosen for clarity, not necessarily length.
- Dropping a quote never blocks the theme itself — the theme and its count still ship.

## Output
Per theme: either one verbatim quote with speaker and source link, or no quote field at all (theme count still included).

## Trend delta

Compare this week's theme counts against last week's snapshot to surface what's rising, falling, or new.

## Method
1. Load last week's theme snapshot from `reports/feedback/`.
2. Match this week's themes to last week's by name; treat an unmatched new theme as "new."
3. Compute the change in distinct-customer count for each matched theme.
4. Rank themes by absolute change first, then by raw count, to surface what moved most.
5. Flag any theme that dropped off entirely (had a count last week, zero this week) as "resolved or quiet," not deleted.
6. Write this week's counts back as the new snapshot for next week's comparison.

## Rules
- Always diff against the most recent snapshot, never against an arbitrary earlier week.
- A theme with no prior snapshot is always "new" — do not backfill a fabricated prior count.
- Report the delta as a real number or arrow, never a vague "up" or "down" without the count behind it.
- Keep the snapshot file even for themes that don't make the top 5 — the full history stays for future diffs.

## Output
Per theme: this week's count, last week's count, and the delta (new / rising / falling / flat), ranked by movement.

## Recurring jobs

This skill pairs with scheduled jobs. Offer to set each one up; only create the ones the user approves:

- Weekly feedback synthesis (Every Monday at 6:00 AM) — https://eluu.ai/cookbook/feedback-synthesis-weekly/jobs-setup/weekly-feedback-synthesis.md