persona-detection

Persona Detection Skill

Inheritance: heir:vscode

Purpose

Detect the most appropriate user persona for a workspace to personalize the Alex experience — sidebar branding, skill suggestions, and working memory P5-P7 configuration.

Priority Chain (Highest to Lowest)

Detection stops at the first confident match. Lower priorities only run if higher ones return null.

Available Personas (as of v5.7.1)

developer, academic, researcher, technical-writer, architect, data-engineer, devops, content-creator, fiction-writer, game-developer, project-manager, security, student, job-seeker, presenter, power-user

Workspace Scoring (Priority 6 Heuristics)

Three pattern types with distinct matching logic:

Critical rule: Never use bidirectional substring matching for workspace patterns. It causes false positives (e.g., .github/workflows/ matching any .github/ entry).

LLM Detection Path

When detectAndUpdateProjectPersona() is called (initialize/upgrade):

1. Detect technologies from file structure
2. Gather project context (directory tree, README, package.json, copilot-instructions.md)
3. Send structured prompt to LLM (prefers GPT-4o > Claude Sonnet > any)
4. Parse JSON response → match to existing persona or create dynamic one
5. Save to user-profile.json as projectPersona
6. Update Active Context persona field via updatePersona()

Active Context Integration

Persona detection updates the Active Context section in copilot-instructions.md:

## Active Context
Persona: Developer (90% confidence)
Objective: v5.7.1 — Avatar UI Enhanced + Architecture Validation Complete
Focus Trifectas: release-management, quality-assurance, documentation-quality-assurance
Principles: KISS, DRY, Quality-First
Last Assessed: 2026-02-15 — v5.7.1

What gets updated:

• Persona: Name + confidence (e.g., "Developer (90% confidence)")
• Focus Trifectas: NOT auto-assigned during persona detection — managed independently via updateFocusTrifectas()

Why separate?

• Persona = WHO the user is (developer, writer, researcher)
• Focus Trifectas = WHAT I'm working on NOW (3 skills for current task)
• Focus changes frequently (per-session), persona changes rarely (per-project)

The welcome view displays the persona with confidence % and detection source (profile vs. project detection).

Adding New Personas

1. Add to PERSONAS array in personaDetection.ts with: id, name, bannerNoun, hook, skill, icon, accentColor, keywords, techStack, projectPatterns
2. Add persona ID to the LLM prompt's personaId enum
3. Add skill to skillNameMap in welcomeView.ts
4. Ensure matching skill exists in .github/skills/

Note: Focus Trifectas are NOT defined per-persona. They're managed independently based on current work context.

Known Pitfalls