google-workspace

Router skill. Resolve topic, then Read EXACTLY ONE playbook from routes:. Never load the whole cluster — one route per invocation. [DOC]

When to use

A request touches a Google API/Workspace surface (Sheets, Docs, Slides, Drive, Calendar, Maps, GA4/Analytics) or spans several Google services and needs an auth/scope/quota/retry plan. If no Google surface is involved, decline. [INFERENCE]

Routing

• One service named → that topic (e.g. "GA4 events" → analytics-implementation; "read a spreadsheet" → google-sheets-mcp). [INFERENCE]
• Two+ services or "integrate/connect Google APIs" → google-apis-integration. [DOC]
• analytics-implementation = GA4/Firebase/BigQuery setup; google-analytics = reporting/Data API querying. Don't conflate. [DOC]
• Ambiguous between two topics → ask one disambiguating question, don't guess. [INFERENCE]

Depth

• quick (default): essentials only — the minimal correct path. [DOC]
• deep: apply the playbook exhaustively with verification at each step. [DOC]

Spine

Discover → Analyze → Execute → Validate. Quality gates: constitution v6.0.0 (enforcement), evidence tags (Alfa set: [DOC]/[CONFIG]/[CÓDIGO]/ [INFERENCE]/[ASSUMPTION]), script-first rule. [DOC]

Acceptance gate (before "done")

• Exactly one playbook was loaded; topic matches the user's actual surface. [DOC]
• Credentials/scopes confirmed present; no secrets pasted inline. [ASSUMPTION]
• Output carries one-family evidence tags; assumptions pair with a verify step. [DOC]

Assets

Routing gate is backed by assets/ — assets/routing-checklist.md (resolve → load → plan → gate) and assets/quality-rubric.json (8-criterion acceptance gate). See assets/README.md. [DOC]

Anti-patterns

• Loading multiple routes "to be safe" — defeats the router. [INFERENCE]
• Answering Google-API questions from memory instead of the playbook. [INFERENCE]
• Guessing topic when the request is ambiguous, or inventing quotas/prices. [DOC]