Analyze Feature Requests (theme, dedupe, find the real need)

Turn a messy pile of customer asks into named themes with the duplicates merged and the underlying need surfaced behind each one — "never allow customers to design solutions; prioritize the problem." The requests don't come from a spreadsheet the user pastes; they come straight from Evermuse, so the triage reflects the whole corpus, not just what someone remembered to forward.

Step 0 — Relevance & availability

Confirm this is request triage and Evermuse is connected (see using-evermuse Step 0). If disconnected, ask the user to paste the requests and label the output ⚠ ungrounded — but the point of this skill is pulling them from the corpus, so prefer to fix the connection.

Evermuse Grounding (required)

Follow ${CLAUDE_PLUGIN_ROOT}/skills/using-evermuse/SKILL.md.

Required — one parallel batch of searches. Verify the product (Rule 1), then fire in a single parallel batch:

• 3–4 evidence searches, each worded from a different angle (limit up to 50). Evidence comes back rich and varied — expect large, useful result sets.
• one guidance search and one context search (limit up to 50). These are usually sparse or empty; run them anyway and note when they're thin.

Read each response's digest — it reports how many more results exist. Use judgment on whether a query is worth pulling deeper (raise limit toward the 100 max and/or page with next_offset to avoid repeats), weighing payload size, remaining context, task complexity, and the value of the data. For deep pulls, consider spawning sub-agents — instruct them to return every citation with the same metadata the tools return (url, who_said_it, meeting_name, created_at) so you can still cite.

Optional — considered use. Once grounded, reach for the other tools only when they add value: a quote-angled search for verbatim voice, read_source for a single deep dive, inline citations (references/citations.md), and add_source to save the deliverable. Sources, citations, and saving are optional — not required.

Instructions

1. Confirm the product goal the triage serves — it decides which themes matter, and strategic alignment can't be judged without it.
2. Cluster into themes. Group the notes into named themes. Collapse duplicates and near-duplicates (different words, same ask) into one theme with a combined count. Note when the same customer asked repeatedly vs. many distinct accounts — distinct accounts is the stronger signal.
3. Surface the underlying need. For each theme, look past the requested solution to the job/pain driving it. "Add a CSV button" and "let me email this to my boss" may both be "I need to get this data out to share it." Name that need — it's what you'll actually prioritize and build against.
4. Assess each theme on demand strength (distinct accounts), sentiment, strategic alignment, and rough effort/risk. Open hot items with view_item to get specifics.
5. Recommend the top 3 themes to act on, each with: the underlying need, the customer evidence (counts + quotes), alternative solutions worth considering, the highest-risk assumption, and a cheap way to test it (hand off to /evermuse:assumptions).

Deliverable

**Goal:** [product objective the triage serves]

### Themes
| Theme | Underlying need | Requests (distinct accounts) | Dupes merged | Sentiment | Aligned? |
|-------|-----------------|------------------------------|--------------|-----------|----------|
| Data export | "get my data out to share it" | 9 (6 accts) | 4 | frustrated | ✓ |

### Top 3 to act on
1. **[Theme]** — need: […]. Evidence: > "[verbatim]" — [Name], [Meeting], [Date] [`1`](LINK) (6 accounts). Alt solutions: … Riskiest assumption: … Test: …

Be honest when a "hot" request is really one loud account repeating itself — say so rather than inflating it into a trend.

Further reading

• Adapted from phuryn/pm-skills (MIT) — analyze-feature-requests. Opportunity Score for problem-ranking: Dan Olsen, The Lean Product Playbook.