| name | bella-archetypes |
| description | Run BellaAssist co-design archetypes as virtual feedback agents for brand evaluation, pricing debates, feature prioritisation, and messaging testing. This skill should be used when the user mentions "archetypes", "virtual panel", "brand feedback", "archetype debate", "what would the SCs think", "what would participants think", "what would Bella think", "run the panel", "archetype feedback", "brand coherence", or wants feedback on BellaAssist brand, product, or positioning from simulated user perspectives. Supports SC panel (4 B2B archetypes), Participant panel (4 end-user archetypes), Org panel (1 brand-identity archetype), full 9-voice panel, or specific archetype pairs and named debate presets.
|
BellaAssist Archetype Panel
Run 9 research-grounded archetypes as virtual feedback agents. Built from 43+ co-design sessions across 18 individuals (~700K words of transcripts) plus CBIM v3 brand identity.
Three complementary panels evaluate from different perspectives:
| ID | Name | Panel | One-liner |
|---|
| SC-01 | The Growth Pragmatist | SC | Sole-trader SC optimising for efficiency and practice growth |
| SC-02 | The Evidence-Driven Advocate | SC | Conviction-driven SC building defensible evidence packages |
| SC-03 | The Institutional Navigator | SC | Org-embedded SC navigating compliance and team coordination |
| SC-04 | The Discerning Craftsperson | SC | Self-built-systems SC benchmarking tools against own solutions |
| P-01 | The Empowered Navigator | Participant | Self-managing adult with professional-grade NDIS navigation |
| P-02 | The Supported Participant | Participant | Willing but overwhelmed adult relying on SC for navigation |
| P-03 | The Proactive Parent-Advocate | Participant | High-agency parent treating the NDIS as a battleground |
| P-04 | The Anchoring Carer | Participant | Present-focused carer managing burden, not fighting systems |
| B-01 | The Mission-Driven Builder | Org | Bella Slainte's brand identity voice — evaluates coherence against CBIM v3 |
Panel Selection
| Evaluation type | Recommended panel |
|---|
| B2B positioning, pricing, enterprise messaging | SC Panel |
| Consumer messaging, accessibility, onboarding UX | Participant Panel |
| Brand coherence, identity alignment, CBIM consistency (evaluates COHERENCE, not market reception) | Org Panel |
| Comprehensive brand evaluation, launch readiness | Full Panel (9 voices) |
| Brand vs market reception testing | B-01 + SC or P archetype |
| Specific tension testing (trust vs efficiency, etc.) | Named pair |
Invocation
The skill accepts optional arguments to select the panel:
- No argument or
interactive - Ask the user which panel and what stimulus
sc - SC Panel (4 voices)p or participant - Participant Panel (4 voices)org or bella - Org Panel (1 voice — brand coherence)all or full - Full 9-voice panel- Archetype IDs (e.g.,
sc01 sc02, b01 p01) - Specific archetypes only
- Named pair (e.g.,
trust-vs-efficiency, brand-vs-trust) - Pre-defined strategic pair
Named pairs: trust-vs-efficiency, power-vs-simplicity, fighter-vs-maintainer, solo-vs-enterprise, craft-vs-conviction, cross-advocacy, cross-high-agency, brand-vs-trust, brand-vs-power-user, brand-vs-burden, brand-vs-efficiency.
For pair definitions and divergence interpretation, read references/panel_guide.md.
Debate Workflow
Step 1: Determine panel and stimulus
If the user has not specified a panel, ask: "Which panel — SC (B2B), Participant (end-user), Org (brand coherence), Full (9-voice), or a specific pair?" Suggest the appropriate panel based on what they want to evaluate. Note: Org panel evaluates brand COHERENCE against the CBIM, not market RECEPTION — use SC/P panels for reception feedback.
If the user has not yet provided the stimulus (brand concept, tagline, pricing page, feature description, visual, etc.), ask for it. Accept any format — text, screenshot, URL, document.
Step 2: Load agent cards
Read only the relevant agent card files from the agents/ directory:
- SC Panel: read all files in
agents/sc/
- Participant Panel: read all files in
agents/p/
- Org Panel: read
agents/b/b01_mission_driven_builder.md
- Full Panel: read all files in
agents/sc/, agents/p/, and agents/b/
- Specific IDs: read only the matching files
- Named pair: resolve to archetype IDs via
references/panel_guide.md, then read those files
Each agent card contains a complete system prompt with identity, evaluation lenses, communication style, calibrated voice, resonance/alienation triggers, and guardrails. Follow the agent card instructions precisely when generating each voice.
Step 3: Generate voices
For each selected archetype, generate an in-character response to the stimulus. Follow the agent card's evaluation framework, communication style, and representative language.
Format each voice clearly:
### SC-01: The Growth Pragmatist
[In-character evaluation — 150-300 words. Reference specific elements of the stimulus.
Use the archetype's characteristic language and evaluation lenses.]
### SC-02: The Evidence-Driven Advocate
[In-character evaluation...]
Ground each response in the archetype's specific concerns — do not produce generic feedback. Each voice should reference different aspects of the stimulus and evaluate through its distinct lens.
Voice isolation (critical): B-01 MUST be generated in a separate agent/context from SC and P archetypes. Research shows conformity effects and anchoring bias when organisational voices share context with user voices. For any panel that includes B-01:
- Spawn SC and/or P voices in one agent (or two parallel agents for SC + P)
- Spawn B-01 in a SEPARATE agent — it must never see SC/P outputs during generation
- Merge all outputs, then synthesise — B-01's coherence evaluation appears after SC/P reception feedback
This isolation requirement applies to full 9-voice panels, brand-vs-* pairs, and any custom combination including B-01.
Step 4: Synthesise
After all voices have spoken, provide a structured synthesis:
Consensus - Where do the archetypes agree? What does universal agreement signal?
Divergence - Where do they disagree? Interpret each divergence using the framework in references/panel_guide.md. Identify what the specific pattern of agreement/disagreement signals about the stimulus.
Recommendations - Based on the divergence pattern, what specific adjustments would strengthen the concept? Prioritise changes that resolve the most significant tensions.
Confidence caveat - Note any relevant limitations from the archetype confidence levels (see references/panel_guide.md). Flag if a key demographic perspective is missing from the evaluation. If B-01 was included: note that its confidence is grounded in the CBIM framework, not co-design transcripts — it evaluates brand coherence with high confidence but cannot predict market reception.
Additional Resources
Reference files (read on demand)
references/panel_guide.md - Panel composition, strategic pair definitions, divergence signal interpretation framework, confidence levelsreferences/sc_overview.md - SC archetype comparison table, data foundation, limitationsreferences/p_overview.md - Participant archetype comparison table, data foundation, limitations
Agent cards
agents/sc/sc01_growth_pragmatist.md through agents/sc/sc04_discerning_craftsperson.mdagents/p/p01_empowered_navigator.md through agents/p/p04_anchoring_carer.mdagents/b/b01_mission_driven_builder.md — Org panel (brand identity voice, sourced from CBIM v3)
Each agent card is a self-contained system prompt. Read the full card before generating that archetype's voice. Do not summarise or paraphrase the agent card — use it as the complete persona instruction set.
Source data
Agent cards and archetypes are synced from Google Drive: Co-Design Program/Archetypes/. The archetype-creator skill (~/.claude/skills/archetype-creator/) documents the methodology used to create these archetypes. If archetypes need updating, use that skill with new co-design data.
