Summarize a Conversation (interview or meeting)

Turn one specific conversation into a structured, accessible summary — pulling needs, decisions, and verbatim quotes with timestamps out of the actual transcript. This is the sanctioned place to use read_source: a single-conversation deep dive, not a corpus search. The summary can then be saved back so one call's insights compound into the evidence base.

Step 0 — Relevance & availability

Confirm the user means a specific conversation (not "what do customers think about X" across many — that's /evermuse:customer-research) and Evermuse is connected (see using-evermuse Step 0). If disconnected but the user pasted a transcript, summarize that directly and label it ⚠ ungrounded — not from Evermuse (no source links).

Evermuse Grounding (required)

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

Required — locate, then read (this skill is serial, not a corpus batch). Verify the product (Rule 1), then:

1. find_sources to locate the one conversation (filter by title_keyword, participant (a person's name or email), attendee_email/attendee_domain, or date; or query if the user described it by topic). If more than one plausible match, confirm which meeting with the user before reading.
2. read_source on that source and read it fully before writing — page with offset/next_offset until has_more is false. The transcript IS the grounding for this skill.

Do not run the multi-search corpus batch here — that's for research skills across many calls (/evermuse:customer-research). One conversation in, one summary out.

Optional — considered use. After reading, reach for other tools only when they add value: a single evidence search to check whether a need heard here echoes across the corpus, inline citations (references/citations.md), and add_source to save the deliverable. Citations and saving are optional — not required.

Two flavors (pick by conversation type)

A — Customer interview → discovery summary (JTBD-focused)

Tag interview-summary. Template:

**Date**: […]   **Participants**: [names + roles]
**Background**: [about the customer]
**Current Solution**: [what they use today]

**What They Like** (JTBD · desired outcome · importance · satisfaction):
- …

**Problems With Current Solution** (JTBD · desired outcome · importance · satisfaction):
- …

**Key Insights / Notable Quotes**:
- > "[verbatim]" — [Speaker] [12:340] [`1`](URL)

**Action Items**: [Date · Owner · Action]

B — Team / stakeholder meeting → minutes (decision-focused)

Tag meeting-summary. Template:

## Meeting Summary
**Date & Time**: […]   **Participants**: [names + roles]   **Topic**: […]

**Summary**
- **Point 1**: [key discussion point or decision]
- **Point 2**: …

**Action Items**
| Due Date | Owner | Action |
|----------|-------|--------|

**Decisions Made**
- …
**Open Questions**
- …

Instructions

1. Locate, then read the one conversation (never dump the whole corpus). If ambiguous, confirm which meeting.
2. Read the full transcript before writing.
3. Extract to the right template. Use "-" where info is missing. Preserve timestamps on quotes (they're the transcript's citation). Pull needs and pains as JTBD (interview) or decisions and owners (meeting).
4. Write for clarity — plain language a non-attendee understands; be objective, summarize what was said not opinions; use "we" for team meetings.
5. Offer to save so the needs/quotes rejoin the evidence base, then offer next steps (/evermuse:opportunity-solution-tree, /evermuse:analyze-feature-requests).

Further reading

• Adapted from phuryn/pm-skills (MIT) — summarize-interview (pm-product-discovery) + summarize-meeting (pm-execution).