User Personas (from real people, not archetypes)

Produce 3 personas that a skeptical stakeholder can't dismiss as made up — because every trait, pain, and goal traces to a real person who appears in the Evermuse corpus, and each persona carries a block of that person's own words. A persona nobody in the corpus resembles is a failure of this skill.

Step 0 — Relevance & availability

Confirm this is user/customer profiling work and Evermuse is connected (see using-evermuse Step 0). If the MCP isn't connected, produce framework-only personas labeled ⚠ ungrounded — Evermuse not connected and tell the user to authorize the MCP. Never present invented archetypes as research-backed.

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.
• one find_sources browse (no query, limit up to 50) to map who actually exists in the corpus — the participants, attendees, accounts, and domains across sources. To pull everything one person or one account said, re-run find_sources with participant (name or email) or attendee_domain. Personas are composites of these real people; this call is the raw material for Instructions step 1.

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

Build 3 refined personas for $ARGUMENTS, grounded entirely in the corpus.

1. Map who's really there. From find_sources, list the distinct people/accounts. This is your raw material — personas are composites of these real users, not personas you'd expect the market to have.
2. Cluster by job and behavior. Group the real people by shared jobs-to-be-done, workflows, and motivations. Distinct needs, not distinct age brackets. Aim for 3 non-overlapping personas; if the corpus only supports 2, say so rather than padding.
3. Enrich each persona from evidence, using the structure below. Where the source method would reach for a survey or "if the user provides data", reach for evidence searches and a quote-angled search instead.
4. Validate. Every attribute must trace to something a real person said or did. Flag any trait you're inferring vs. one that's directly evidenced.

Persona structure (each of the 3)

Name & snapshot — a memorable label + role/context (the composite of who this represents in the corpus).

Primary job-to-be-done — the core outcome they're trying to achieve, with context and frequency, drawn from what they described.

Top 3 pains — each a real obstacle, each badged with the account(s) it came from and severity.

Top 3 gains — the outcomes they seek and how they'd measure success.

One unexpected insight — a counterintuitive pattern the evidence reveals, and why it matters for the product.

In their own words (required) — a block of 2–4 verbatim quotes from real people in this cluster:

> "[verbatim quote]" — [Name], [Meeting], [Date] [`1`](URL)
> "[verbatim quote]" — [Name], [Meeting], [Date] [`2`](URL)

Product fit — how $ARGUMENTS serves (or fails) this persona, with the friction points evidenced.

End with a one-line honesty note if any persona rests on thin evidence (e.g. "Persona C: only 2 accounts — validate with more interviews"). Offer /evermuse:customer-research to go deeper on any one persona.

Further reading

• Persona method adapted from phuryn/pm-skills (MIT); grounding is Evermuse-native. See also the JTBD Masterclass.