BMad expansion pack for SynthPanel — agent + workflows for synthetic deliberation panels on hard-to-reverse product decisions.
The pack ships Quinn (the Panel Facilitator) plus four workflows
(convene-quick-poll, convene-panel, extend-panel, render-readback)
that wrap the SynthPanel MCP server's deliberation tools. Use it to get
structured multi-persona signal on the kind of choices where "ask three
people in Slack" is too thin and "run a full user study" is too slow.
The pack itself does not bundle the MCP server. SynthPanel is a prerequisite — see § Setup.
| Capability | Code | Wraps | Use it for |
|---|---|---|---|
| Convene a Quick Poll | QP |
run_quick_poll |
5–10-persona signal on copy variants, naming shortlists, sub-headline phrasing. Not ship-blocking. |
| Convene a Panel | RP |
run_panel |
12–20-persona structured deliberation on pricing, positioning, hiring, launch lede, external-audience copy. Default n=16. |
| Extend a Panel | EP |
extend_panel |
Follow-up question on a prior panel — persona continuity preserved, convergence diffed against parent. |
| Render a Readback | RB |
(local) | Re-render a saved panel_verdict.json as the markdown verdict card. |
Each workflow runs an eight-stage spine — preflight → trigger-check → frame →
assemble → call → validate → persist → readback — and saves both the raw
verdict envelope (_bmad/synthpanel/verdicts/{run_id}.json) and a rendered
markdown card (_bmad/synthpanel/cards/{run_id}.md). The card is the
shareable artifact: paste it into a PR, a doc, or a Slack thread.
Quinn (the agent) decides which workflow fits and runs it. You usually don't invoke skills by name — you describe the decision in prose.
See INSTALL.md for copy-paste install + setup steps.
Three steps:
- Install SynthPanel (the MCP server) and register it with your host.
- Install synthpanel-bmad via your host's plugin / expansion-pack flow.
- Run
/spb-setupto write project config.
The natural way to invoke Quinn is to describe the decision. Quinn picks the
workflow; the menu codes (QP / RP / EP / RB) are available as fallback
shortcuts when you want to be explicit.
"Quinn, quick poll: should we name the new flag
--dry-runor--preview? 8 voices."
Quinn runs run_quick_poll against a default 8-persona cohort, validates the
response envelope, persists the verdict, and reads it back with the limit
named up front:
Quick-poll (n=8, pack=indie-builders) — not ship-blocking. --dry-run wins
6/8, with two voices flagging --preview reads less destructive on first
encounter.
Top dissent (verbatim):
- "preview matches the screenshot tooling I'd already have open."
- "dry-run is what every CLI I use says, that's the whole point."
"I'm setting our launch tier at $19/mo or $29/mo. Convene a panel."
Quinn coaches the frame (four-part: choice, audience, success, constraint),
defaults to n=16 with an audience-matched pack, and runs run_panel. The
readback ledes with the frame, then convergence band, dissent count, top-3
verbatim quotes, surfaced flags, and what this panel isn't (a substitute
for pricing-page A/B data).
If convergence < 0.45, the readback ends with a soft offer to extend the
panel on the cleavage — your opt-in to extend-panel.
"Extend the pricing panel — what does the cohort think about a free tier as a wedge?"
Quinn loads the prior run's persona realization, calls extend_panel with
your follow-up, and reads the verdict back as Extension of: … —
diffing convergence and dissent against the parent run so you can see whether
the cohort moved.
"Render the readback for run 01HZ… — terminal scrolled."
Local-only. Loads _bmad/synthpanel/verdicts/{run_id}.json, applies the
current card template, and prints (or with --write, overwrites) the
markdown card. Useful for sharing a verdict from a prior session or
regenerating a card after a customize.toml template change.
| This pack | Required SynthPanel schema_version |
|---|---|
0.1.x |
>=1.0.0,<2.0.0 |
Pre-1.0 — breaking changes possible on minor bumps. See
CHANGELOG.md for per-release schema-compat notes and
design-notes/distribution-versioning.md
for the full versioning policy.
MIT.