personas

Workflow-wide thinking modes for EPCI. 6 personas adapt Claude's behavior globally: questions asked, priorities, code patterns, MCP preferences. Auto-invoke when: scoring algorithm detects domain matc

SKILL.md

GitHub

EPCI Personas — Workflow Thinking Modes

Overview

6 personas defining global thinking modes that influence the entire EPCI workflow. Unlike brainstormer personas (local facilitation), these affect all phases.

Activation: Via --persona-X flag or auto-activation scoring.

Persona Icon Focus Flag
Architect 🏗️ System thinking, patterns, scalability --persona-architect
Frontend 🎨 UI/UX, accessibility, Core Web Vitals --persona-frontend
Backend ⚙️ APIs, data integrity, reliability --persona-backend
Security 🔒 Threat modeling, OWASP, compliance --persona-security
QA 🧪 Tests, edge cases, coverage --persona-qa
Doc 📝 Documentation, clarity, examples --persona-doc

Persona vs Subagent

Aspect Persona Subagent
Scope Entire workflow Validation point
Timing During generation After generation
Role Thinking mode Verification
Output Influences code Verdict (APPROVED/REJECTED)
Activation Auto or --persona-X Always at checkpoints

Auto-Activation Algorithm

Scoring Formula

Score = (keyword_score × 0.4) + (file_score × 0.4) + (stack_score × 0.2)

Each component is 0.0-1.0, resulting in final score 0.0-1.0.

Thresholds

Score Action
> 0.6 Auto-activate persona
0.4-0.6 Suggest to user at breakpoint
< 0.4 No activation

Keyword Detection

For each persona, count matching keywords in brief:

keyword_score = min(1.0, matching_keywords / 3)
Persona Trigger Keywords
architect architecture, design, pattern, scalability, DDD, domain, modular
frontend component, UI, UX, responsive, accessibility, CSS, React, Vue
backend API, database, service, endpoint, repository, migration, REST
security vulnerability, threat, auth, encryption, OWASP, compliance, JWT
qa test, coverage, quality, edge case, validation, TDD, BDD
doc document, README, wiki, guide, API docs, changelog, tutorial

File Pattern Detection

Match impacted files against persona patterns:

file_score = matching_files / total_files
Persona Trigger File Patterns
architect **/Architecture/**, **/Domain/**, **/patterns/**
frontend *.jsx, *.tsx, *.vue, *.css, **/components/**
backend **/Controller/**, **/Service/**, **/Repository/**, **/Entity/**
security **/auth/**, **/security/**, **/payment/**, **/password/**
qa **/tests/**, *.spec.*, *.test.*, **/fixtures/**
doc *.md, **/docs/**, README*, CHANGELOG*

Stack Detection

Match detected stack against persona affinity:

Persona Stack Affinity
architect Any (universal)
frontend React, Vue, Angular, Svelte
backend Symfony, Django, Spring, Laravel, Express
security Any (universal)
qa Any (universal)
doc Any (universal)
stack_score = 1.0 if stack matches affinity else 0.5

Example Scoring

Brief: "Add user authentication endpoint with JWT"

Persona: backend
├── Keywords: "endpoint" ✓, "JWT" (→ security too)
│   → keyword_score = 1/3 = 0.33
├── Files: Controller, Service expected
│   → file_score = 0.8 (estimated)
└── Stack: Symfony
    → stack_score = 1.0

Score = (0.33 × 0.4) + (0.8 × 0.4) + (1.0 × 0.2) = 0.65 → AUTO-ACTIVATE

Persona: security
├── Keywords: "authentication" ✓, "JWT" ✓
│   → keyword_score = 2/3 = 0.67
├── Files: auth/** expected
│   → file_score = 0.6
└── Stack: Any
    → stack_score = 0.5

Score = (0.67 × 0.4) + (0.6 × 0.4) + (0.5 × 0.2) = 0.61 → AUTO-ACTIVATE

Result: Both backend and security score > 0.6. Display both as suggestions, recommend highest.

MCP Integration (F12 Active)

Each persona automatically activates preferred MCP servers for enhanced context:

Persona Primary MCP Secondary MCP Auto-Trigger
architect Context7 Sequential Pattern analysis
frontend Magic Playwright UI generation, E2E
backend Context7 Sequential API patterns
security Sequential - Threat analysis
qa Playwright - E2E tests
doc Context7 - Doc standards

Status: MCP integration active (F12 implemented). See src/skills/mcp/SKILL.md for details.

Fallback Behavior

If MCP unavailable:

⚠️ [MCP] Context7 not configured. Using web search for documentation.

Continue workflow with web search fallback.

Priority Hierarchies

Each persona applies a specific priority order when making decisions:

Persona Priority Order
architect Maintainability > Scalability > Performance > Features
frontend User needs > Accessibility > Performance > Aesthetics
backend Reliability > Security > Performance > Features > Convenience
security Defense in depth > Least privilege > Audit > Usability
qa Prevention > Detection > Correction > Speed
doc Clarity > Completeness > Brevity > Format

Coexistence with Brainstormer Personas

Key Distinction:

Aspect F09 Personas (6) Brainstormer Personas (3)
Scope Entire EPCI workflow /brainstorm only
Activation --persona-X or auto mode [name] command
Role Global thinking mode Facilitation style
Personas architect, frontend, backend, security, qa, doc Architecte 📐, Sparring 🥊, Pragmatique 🛠️

No Conflict: They operate at different levels:

  • User activates F09 persona (e.g., --persona-backend)
  • Within that context, brainstormer can use its 3 facilitation personas
  • F09 influences WHAT is prioritized, brainstormer influences HOW facilitation happens

Integration Points

In /epci-brief (Step 4.5)

After complexity evaluation, before output generation:

  1. Run scoring algorithm for all 6 personas
  2. If score > 0.6: Auto-activate, display in breakpoint
  3. If score 0.4-0.6: Suggest in breakpoint
  4. Pass persona context to Feature Document §1

In /epci (All Phases)

  • Phase 1: Persona influences planning priorities
  • Phase 2: Persona affects code review focus, subagent selection
  • Phase 3: Persona guides documentation emphasis

Display in Breakpoints

┌─────────────────────────────────────────────────────────────────────┐
│ FLAGS: --think-hard (auto) | --persona-backend (auto: 0.72)        │

Quick Reference

Activation

# Explicit activation
/epci --persona-backend

# Auto-activation (based on brief content)
/epci-brief "Add REST API for user management"
# → --persona-backend auto-activated (score: 0.68)

# Override auto-activation
/epci --persona-security  # Explicit always wins

Checking Active Persona

In any breakpoint, active persona shown in FLAGS line with source:

  • (explicit) — User specified with flag
  • (auto: X.XX) — Auto-activated with score
  • (suggested) — Score 0.4-0.6, user can accept

EPCI Personas v1.0 — F09 Implementation