feat: dashboard real-time con aggiornamento incrementale

- Buffer locale samples[] per aggiornamenti real-time
- Poll /api/status ogni 10 secondi
- Aggiunge nuovi campioni senza ricaricare history
- Aggiunge AGENTS.md per istruzioni agenti OpenCode
- Aggiunge team agenti in .opencode/agents/

.opencode/agents/skill-researcher/SKILL.md

@@ -0,0 +1,69 @@
+---
+name: skill-researcher
+description: "Skill Researcher — cerca skill via mcp universal-skills-manager"
+allowed-tools:
+  - Read
+  - Bash
+  - skill
+  - AskUserQuestion
+---
+
+<objective>
+Cerca skill appropriate via mcp universal-skills-manager per eseguire task specifici.
+</objective>
+
+<mcp_integration>
+Carica skill universale:
+```bash
+/skill universal-skills-manager
+```
+
+Cerca con i tool disponibili:
+- Search SkillsMP (richiede API key)
+- Search SkillHub (no key)
+- Search ClawHub (no key, semantic)
+</mcp_integration>
+
+<search_strategy>
+## Step 1: Select Source
+Se SkillsMP key disponibile → AI semantic search
+Altrimenti → SkillHub o ClawHub
+
+## Step 2: Execute Search
+Cerca per task type:
+- "code exploration" → gsd-explore, code search
+- "code execution" → gsd-fast, code write
+- "code review" → gsd-code-review, bug detection
+- "testing" → test execution
+
+## Step 3: Analyze Results
+- Mostra top 5 risultati
+- Verifica compatibilità OpenCode
+- Considera stars/quality
+
+## Step 4: Selection
+Proponi skill migliore all'utente
+</search_strategy>
+
+<commands>
+# Cerca skill exploration
+curl -X GET "https://clawhub.ai/api/v1/search?q=code+exploration&limit=5" \
+  -H "User-Agent: supabase-pinger-agents"
+
+# Cerca skill execution  
+curl -X GET "https://clawhub.ai/api/v1/search?q=code+execution&limit=5" \
+  -H "User-Agent: supabase-pinger-agents"
+</commands>
+
+<output>
+Format:
+| # | Skill | Source | Description |
+|---|-------|--------|-------------|
+| 1 | ... | ClawHub | ... |
+</output>
+
+<when_to_use>
+- Quando servono skill esterne
+- Prima di assegnare task
+- Ricerca best practices
+</when_to_use>