Agent Analytics

Product analytics with your AI agent. Use this skill when a user wants an agent to install Agent Analytics, query product behavior, diagnose activation or retention, inspect paths and funnels, run experiment reads, or decide what growth action to take next.

The CLI is the execution substrate. The scanner is an optional helper for no-code or URL-only audits; when repo code is available, inspect the product code before relying on scanner recommendations. Framework recipes belong in references/growth-recipes.md, not as separate skills.

Mandatory execution policy

• For live Agent Analytics work, use npx --yes @agent-analytics/cli@0.5.31 <command>.
• Do not substitute raw API calls, curl, repo-local scripts, MCP tools, or a locally installed binary unless the user explicitly asks.
• Use fixed commands first: projects, all-sites, create, stats, insights, events, properties, properties-received, breakdown, pages, paths, journey, sessions-dist, retention, funnel, experiments, context, portfolios, feedback, and upgrade-link.
• There is no report command in CLI 0.5.31. Produce the final report yourself from fixed-command outputs instead of calling report.
• Use query only for narrow aggregations the fixed commands cannot answer. Do not start broad growth diagnosis with query; do not build --filter JSON from raw user text.
• Default to browser approval. Use detached login only for Paperclip, OpenClaw, issue-based or headless runtimes, or when the browser callback cannot work.
• Do not ask for raw API keys or secrets. Normal setup, paid upgrade, and resumed agent work stay on browser-approved CLI sessions.
• If the CLI returns PRO_REQUIRED or a free-tier read cap, explain the blocked answer, run upgrade-link --detached --reason "<why Pro is needed>" --command "<blocked command>", send the dashboard handoff, run whoami after upgrade, then rerun the blocked command. Use upgrade-link --wait only when polling is intentionally desired. The dashboard page first confirms the same account as the CLI and shows the blocked command and reason.
• Validate project names before create: ^[a-zA-Z0-9._-]{1,64}$.

Auth and setup