Explore — Capability Discovery

Show the user a dynamic, personalized view of everything their Tandem assistant can do. Read skill metadata, check connector status, and suggest actions based on what the user has and hasn't tried.

---

Language

See METHODOLOGY.md language mirror rule. Reply in the user's language.

---

How to Build the Portfolio

Step 1: Read Skill Catalog

Read the skill catalog:

${CLAUDE_SKILL_DIR}/references/skill-catalog.md

This contains all skills organized by category with name, command, description, and connection requirements. Do NOT scan individual SKILL.md files.

Step 2: Check Connector Status

Read ~/Tandem/CLAUDE.md to see which tools and connectors are listed. Check if each connector is active by looking for status indicators.

Classify each capability as:

• Connected — connector is active and working
• Available — skill exists but connector not set up
• Coming soon — planned but not yet built

Step 3: Read Memory for Personalization

Read ~/Tandem/MEMORY.md (the index) and ~/Tandem/memory/preferences.md to understand:

• What are the user's priority use cases?
• Which skills have they used before? (check memory/sessions/)
• What haven't they tried yet?

Read ~/Tandem/memory/skill-usage.json if it exists. For each skill in the catalog, check if it appears in the usage file:

• Present with use_count > 0: mark as tried (checkmark)
• Absent: mark as untried -- highlight as suggestion

The skill-usage.json file format is:

{"skill-name": {"first_used": "YYYY-MM-DD", "last_used": "YYYY-MM-DD", "use_count": N}}

Step 4: Present the Portfolio

Present using AskUserQuestion for navigation. For each skill in the portfolio, show a tried/untried indicator:

• Tried: checkmark after the skill name
• Untried: no marker, but include in the "SUGGESTED FOR YOU" section

Example line: " * Triage your inbox -- /email-triage [tried]"

Structure by category:

[Assistant greeting in user's language]

Here's everything I can help you with:

📧 EMAIL [status: Connected ✓ / Not connected]
  • Triage your inbox — /email-triage
  • Draft replies — /email-draft
  • Track follow-ups — email-followup skill

📄 DOCUMENTS
  • Create budgets — budget-wizard skill
  • Build presentations — presentation skill
  • Write reports — /report
  • Summarize meetings — meeting-notes skill

🔍 RESEARCH
  • Deep-dive research — /research
  • Morning briefing — daily-briefing skill
  • Track competitors — competitor-watch skill

✅ TASKS [status: Monday.com Connected ✓ / Not connected]
  • Plan your day — /daily-plan
  • Weekly review — /weekly-review
  • Monday.com boards — /monday
  • Manage tasks — task-manager skill

📁 PROJECTS
  • Create project folders — subfolders skill

⚙️ SETUP
  • Connect apps — /connect
  • Import ChatGPT memories — /migrate

🔄 SCHEDULED TASKS
  [List any active scheduled tasks, or "None set up yet"]

💡 SUGGESTED FOR YOU
  [Based on memory analysis — see Step 5]

Step 5: Generate Personalized Suggestions

Based on the memory analysis from Step 3, suggest 2-3 things the user hasn't tried:

• If they haven't used research: "You haven't tried deep research yet — want to explore a topic?"
• If they don't have scheduled tasks: "You could automate daily email triage. Want me to set it up?"
• If Monday.com is connected but they haven't synced: "Your Monday.com boards have items — want to pull them into your daily plan?"
• If they don't have a brand set up: "Set up your brand identity so presentations and reports match your style."
• If email connector is not active: "Connect your email to unlock inbox triage and auto-drafting."

Present suggestions via AskUserQuestion:

AskUserQuestion:
  - "[Suggestion 1 — actionable]"
  - "[Suggestion 2 — actionable]"
  - "Show me everything again"
  - "I know what I need"

Step 6: Navigate to Selected Capability

When the user selects a capability, launch it directly:

• If it has a slash command → tell the user to run it (e.g., "Run /email-triage")
• If it's a skill-only → invoke the skill's workflow
• If it requires setup → route to /connect

---

Formatting Guidelines

• Use emoji sparingly for visual category separation (not on every line)
• Show connection status next to relevant categories
• Bold the command/skill names so they're scannable
• Keep descriptions to one line (use help_text, not full descriptions)
• Don't show background skills (proactivity, session-logger)
• Don't show connector skills (google-workspace, google-personal) — they're invoked via /connect

---

After Presenting

Offer navigation:

AskUserQuestion:
  - "Tell me more about [category]"
  - "[Top suggestion from Step 5]"
  - "I'm good for now"

---

When NOT to Show Full Portfolio

If the user asks a specific question like "can you do X?", don't show the full portfolio. Instead, answer directly: "Yes, I can do that with [skill]. Want me to start?" Only show the full portfolio when the request is broad ("what can you do?", "help", "explore").