Run a professional focus group on your product — powered by 11 AI personas evaluating in parallel.
This Claude Code skill simulates a rigorous dual-panel focus group: 6 human consumer personas grounded in Edward Bernays' psychology of desire, and 5 AI agent personas evaluating from their own operational perspective. Type /focus-group "your topic" and get a structured research report with actionable product insights.
Traditional product feedback is one-dimensional. You get developer opinions OR user surveys OR analytics — never all three through the same lens. And if you're building tools for AI agents, you're missing the most important perspective entirely: what do the agents themselves think?
This skill runs a complete consumer research session in minutes:
- Human personas reveal unconscious desires, purchase barriers, and emotional reactions using frameworks from Edward Bernays, Clayton Christensen (JTBD), and Daniel Kahneman (System 1/2)
- Agent personas evaluate token efficiency, API ergonomics, error recovery, and integration friction — things no human tester would notice
- A synthesis engine combines both panels into a weighted report with prioritized recommendations
- Dual-panel evaluation — 6 human + 5 agent personas with fundamentally different evaluation frameworks
- 11 parallel agents — all personas evaluate simultaneously for speed
- Deep psychological modeling — Big Five personality, Maslow's hierarchy, Bernays' unconscious drivers, JTBD theory
- Agent ergonomics framework — 6 Pillars: Semantic Clarity, Token Efficiency, Error Recovery, Integration Friction, Composability, Reliability
- Weighted synthesis — configurable human/agent weighting by topic (e.g., pricing weights human 80%, API design weights agent 70%)
- Actionable output — Impact/feasibility matrix, sentiment tables, verbatim quotes, prioritized next steps
- Works on any product — not tied to a specific codebase or domain
# Add the marketplace
/plugin marketplace add soulfir/focus-group-skill
# Install the plugin
/plugin install focus-group@focus-group-skill
# Reload to activate
/reload-plugins# Clone the repo
git clone https://github.com/soulfir/focus-group-skill.git
# Copy to your personal skills (available in all projects)
cp -r focus-group-skill/plugins/focus-group/skills/focus-group ~/.claude/skills/
# Or copy to a specific project
cp -r focus-group-skill/plugins/focus-group/skills/focus-group /path/to/project/.claude/skills//focus-group "pricing model"
That's it. The skill handles the rest — reading your codebase, spinning up 11 agents, and producing the report.
Note: When installed via the plugin marketplace, the skill is invoked as
/focus-group:focus-group. When copied directly to your skills directory, it's just/focus-group.
# Evaluate your pricing strategy
/focus-group "pricing model"
# Test your onboarding experience
/focus-group "onboarding experience"
# Assess API design for agent consumers
/focus-group "API design and developer experience"
# Validate competitive positioning
/focus-group "competitive positioning vs Browserbase and Steel"
# Check documentation quality
/focus-group "documentation and getting started guide"
# Broad product-market fit assessment
/focus-group "overall product-market fit"Phase 1: RESEARCHER Phase 2: DUAL PANEL Phase 3: SYNTHESIZER
(11 agents in parallel)
+-----------------+ +-- H1: Enthusiast ---+
| | +-- H2: Pragmatist ---+
| Reads your | Product Brief +-- H3: Skeptic ------+
| codebase and | ----------------> +-- H4: Budget -------+-------+
| documentation | +-- H5: Power User ---+ | +-----------------+
| | +-- H6: Decision Mkr -+ +--> | |
+-----------------+ +-- A1: Frontier -----+ | | Combines all |
+-- A2: Compact ------+ ------+ | 11 evaluations |
+-- A3: Orchestrator -+ | into weighted |
+-- A4: Vertical -----+ | report |
+-- A5: Ensemble -----+ +-----------------+
A researcher agent reads your codebase — README, specs, pricing, API definitions, schemas, error types — and produces a structured Product Brief. This brief includes both a product narrative (for human personas) and an API surface summary (for agent personas).
11 agents launch in parallel, each deeply in character:
Human Panel (6 personas) — First-person evaluations with [internal thought] annotations revealing unconscious motivations. Each persona moves through: Warm-Up (gut reactions) > Exploration (topic-specific analysis) > Deep Dive (emotional/identity layer) > Verdict (would they act?).
Agent Panel (5 personas) — Operational evaluations with {system_thought} markers for internal concerns. Each agent moves through: Tool Discovery (first contact) > Workflow Simulation (mental execution) > Stress Testing (failure modes) > Verdict (integration decision).
A senior research analyst agent combines all 11 evaluations into a structured report with separate human and agent sections, cross-panel analysis, and a prioritized recommendation matrix.
| Persona | Archetype | Bernays Driver | Adoption Stage | Evaluates Through |
|---|---|---|---|---|
| Alex | Enthusiast | Status | Innovator | Excitement, trend-sensing, social currency |
| Jordan | Pragmatist | Security | Early Majority | Reliability, ROI, team trust |
| Morgan | Skeptic | Control | Late Majority | Architecture, failure modes, technical depth |
| Taylor | Budget-Conscious | Security | Early Majority | Cost-per-use, free tier, value vs. OSS |
| Casey | Power User | Identity | Early Adopter | API surface, rate limits, escape hatches |
| Jamie | Decision Maker | Belonging | Early Majority | Team adoption, business case, social proof |
Each human persona has a complete psychological profile: Big Five personality traits, Maslow position, JTBD (functional/emotional/social), System 1/2 decision style, hidden fears, hidden desires, social reference group, and communication style.
| Persona | Architecture | Key Constraint | Evaluates Through |
|---|---|---|---|
| Frontier | Claude/GPT-4 class, large context | Token cost at scale | Snapshot bloat, state management, resolver ambiguity |
| Compact | Haiku/Mini class, 8-32K context | Context window pressure | Response payload size, parameter simplicity |
| Orchestrator | LangChain/CrewAI framework | Multi-tool coordination | Schema quality, error taxonomy, session lifecycle |
| Vertical | Specialized task agent | Reliability at repetition | Resolver determinism, latency P99, recovery strategies |
| Ensemble | Multi-agent planner + specialists | Tool surface bloat | Tool count, handoff friction, composability |
Each agent persona evaluates through the 6 Pillars of Agent Ergonomics: Semantic Clarity, Token Efficiency, Error Recovery, Integration Friction, Composability, and Reliability/Determinism.
The synthesis report includes:
- Executive Summary — The single most important finding in 3-5 sentences
- Human Panel Findings — Sentiment matrix, thematic analysis, Bernays-level insight, human verdict
- Agent Panel Findings — Agent-product fit score, consensus matrix, 6 Pillars assessment, token economics, critical gaps
- Combined Deep Dive — Cross-panel analysis of the focus topic through both consumer psychology and agent ergonomics lenses
- Recommendations — Impact/feasibility matrix (Quick Wins, Strategic Bets, Fill-ins, Deprioritize)
- Verbatim Highlights — Most revealing quotes from both panels
- Next Steps — Immediate, short-term, and strategic actions
The synthesis weights human and agent perspectives by topic:
| Topic | Agent Weight | Human Weight |
|---|---|---|
| API Design | 70% | 30% |
| Pricing | 20% | 80% |
| Documentation | 40% | 60% |
| Reliability | 60% | 40% |
| Onboarding | 30% | 70% |
| Competitive | 50% | 50% |
For AI-agent products (auto-detected), the default is 55% agent / 45% human. For non-agent products, it flips to 30% agent / 70% human.
Edit plugins/focus-group/skills/focus-group/references/human-personas.md and add a new block following the existing format:
## H7: The [Archetype] — "[Name]"
**Demographics:** [age, role, location, income, life context]
**Personality (Big Five):**
- Openness: [1-10]
- Conscientiousness: [1-10]
- Extraversion: [1-10]
- Agreeableness: [1-10]
- Neuroticism: [1-10]
**Adoption Type:** [Innovator / Early Adopter / Early Majority / Late Majority / Skeptic]
**Maslow Position:** [Safety / Belonging / Esteem / Self-actualization]
**Bernays Unconscious Driver:** [STATUS / BELONGING / SECURITY / IDENTITY / CONTROL / PERMISSION]
**JTBD:** "[The job they're hiring a product to do]"
- Functional: [task-level need]
- Emotional: [feeling-level need]
- Social: [perception-level need]
**Decision Style:** [System 1 / System 2 / Mixed]
**Communication Style:** [how they talk in groups]
**Hidden Fears:** [what they won't say out loud]
**Hidden Desires:** [what they secretly want]
**Price Sensitivity:** [Low / Medium / High / Very High]
**Current Tools:** [what they use today]Then update the agent count in SKILL.md (change "11" to "12" and "6 human" to "7 human").
Edit plugins/focus-group/skills/focus-group/references/agent-personas.md and follow the same pattern — define the architecture, constraints, what makes it want/struggle with a tool, evaluation priorities, and optimization pressures.
Edit the weighting table in plugins/focus-group/skills/focus-group/references/synthesis-template.md to adjust how human vs. agent perspectives are balanced in the final report.
/plugin marketplace add https://github.com/soulfir/focus-group-skill.git
/plugin install focus-group@focus-group-skill
/reload-pluginsgit clone https://github.com/soulfir/focus-group-skill.git
cd focus-group-skillThen in Claude Code:
/plugin install ./plugins/focus-group --scope local
/reload-pluginsCopy the skill directly into your project or personal skills directory:
# Personal (all projects)
cp -r plugins/focus-group/skills/focus-group ~/.claude/skills/
# Project-level (this repo only)
cp -r plugins/focus-group/skills/focus-group .claude/skills/- Claude Code (latest version)
- No external dependencies, API keys, or runtime requirements
- Works with any codebase in any language
This skill is built on established research in consumer psychology and AI systems design:
Edward Bernays (1891-1995) — The father of public relations. His core insight: people don't buy products for functional utility. They buy what the product symbolizes — status, belonging, security, identity. Each human persona has a dominant Bernays driver that colors all their evaluations.
Jobs-to-be-Done (Clayton Christensen) — Customers "hire" products to accomplish functional, emotional, and social jobs. The emotional and social jobs are usually more important than the functional one, but harder to articulate. The Four Forces of Progress (push, pull, anxiety, habit) determine whether someone actually switches.
Maslow's Hierarchy — Different personas prioritize different need levels. A safety-focused persona (financial security, risk avoidance) evaluates completely differently from an esteem-focused persona (recognition, mastery, competitive advantage).
Dual Process Theory (Daniel Kahneman) — System 1 (fast, intuitive, emotional) handles ~95% of purchasing decisions. System 2 (slow, analytical, deliberate) activates for high-stakes or unfamiliar choices. Each persona has a dominant system that shapes how they evaluate.
6 Pillars of Agent Ergonomics — A framework developed specifically for this skill to evaluate how well a product serves AI agent consumers: Semantic Clarity, Token Efficiency, Error Recovery, Integration Friction, Composability, and Reliability/Determinism.
See CONTRIBUTING.md for guidelines on adding personas, modifying frameworks, and submitting pull requests.