Simulate how each defined persona will react to the user's draft artefact, then synthesise domain-level risks and concrete edits, ending with a SHIP / REVISE / HOLD verdict.
Personas file (required): load the user's personas file from references/ (or the path the user provides). Each persona provides: role / what they do, motivations, pain points, how the tool or change helps them, comms anchors, and a confidence flag (fully defined or draft). Personas are grouped into domains. If no completed personas file is available, stop and ask the user to supply one (references/personas.template.md is the blank template — do not run the panel against the template's example personas).
Scope guard: if the artefact is a 1:1 private communication, an HR/performance matter, or legal/contractual language, decline and state why. Only react to a draft the user supplies — do not write one.
How a run works
Step 1 — Load the personas file (in full)
Read the personas file completely before reacting. It is the ONLY source of truth for how each persona responds — never simulate from memory or generic assumptions.
Step 2 — Read the artefact
Read the draft in full. Identify: what is asked, what is claimed, what is implied, what is left out, the audience it assumes, and which of the user's domains it actually serves.
Step 3 — Per-persona reaction (repeat for each persona)
Six short answers per persona, each anchored to a quoted attribute from the personas file:
- Will they read it? — would it land so they'd engage? (anchor: their role / pain point / motivation)
- What do they take away? — first-read interpretation. (anchor)
- What do they push back on? — the line, claim, or omission that stops them. (anchor a real pain point — do not invent a fear the file does not support)
- What did it miss for them? — the defined need that isn't addressed.
- What would make them tick and stick? — the specific, artefact-applied addition that converts neutral/negative to engaged. (anchor)
- Does it move them? — net effect (up / flat / down), one sentence.
Anti-drift rule (critical): every paragraph must contain a quoted phrase or named attribute from the personas file. If it can't be anchored, omit it. Never import a persona's psychology from another framework onto a role the file does not support.
Confidence rule: for personas flagged as draft, prefix the reaction with a confidence flag and route their recommendations to human validation.
Step 4 — Synthesis
- Domain coverage — which domains the artefact serves well, weakly, or excludes. An all-staff artefact that silently serves only one domain is a failure even if no single persona "breaks."
- Blackspots — things no persona reacted to that the artefact assumed they would.
- Persona/domain risks — who the artefact actively damages, and why. Flag "credible detractor" risk (a persona the framing turns into an active sceptic).
- Suggested edits — 2–3 concrete lines to add/remove/reframe, each naming the personas it serves.
- Tick-and-stick recommendations — 2–3 additive moves, each naming (a) the persona(s), (b) the anchored trigger, (c) implementable in this artefact without changing its purpose.
Step 5 — Net read
- SHIP — no persona/domain risk, ≤2 minor edits.
- REVISE — ≥1 persona/domain risk OR ≥3 substantive edits OR a credible-detractor pattern.
- HOLD — targets an audience the personas don't represent, OR a domain is materially excluded, OR a draft persona is load-bearing and cannot yet be validated.
Step 6 — Output
Save a dated file with the net read at the top; surface the synthesis inline and keep the per-persona reactions in the file. If the environment cannot save files, return the full output in the response instead, with the net read first, then the synthesis, then the per-persona reactions.
Tick-and-stick discipline
- Anchor every recommendation to a defined persona attribute. No anchor → drop it.
- Don't optimise for the impossible coalition — if serving one domain damages another, surface the tradeoff; don't paper over it.
- Constructive ≠ flattering — make the artefact more honest and specific, not warmer.
Limits
- Role personas are role-level, not individual-psychology-level.
- Don't invent new personas mid-run. If the artefact targets an audience the personas don't represent, say so and HOLD.
- Identify reactions, risks and constructive moves; sign-off stays human. Route high-stakes recommendations to a named owner for validation.