Lean and assumes Claude's competence — it does not over-explain what a dashboard is, and every line earns its place with a named tool, parameter, or concrete rule. Not level 2 because there is no unnecessary explanation to tighten.

Gives concrete, executable guidance — named MCP tools (dashboards-get-all, dashboard-templates-retrieve, dashboard-update, dashboard-insights-run, etc.) with specific params (search, scope) and precise instructions such as "3–7 word name" and "insights you omit are removed". As an instruction-only skill, the absence of code is not penalized.

A clearly sequenced flow (create vs update → templates → select insights → assemble → verify) with an explicit validation checkpoint ("Verify with dashboard-insights-run to confirm the tiles return data") and a destructive-op warning, so it is not capped at 2.

No bundle files exist (references/, scripts/, assets/ absent) and the body is self-contained and well-sectioned with only single-level pointers to other skills; per the simple-skills note, well-organized sections earn a 3 without external file references.

Lists multiple specific concrete actions — "Build a new dashboard, or update an existing one", "put several metrics/charts together on one page", "add/remove/replace insights on a dashboard" — beyond the level-2 bar of naming domain plus only some actions.

Explicitly answers both what (build/update a dashboard from a set of insights) and when ("Use when a user asks to create a dashboard..."); the explicit Use-when clause clears the level-2 cap.

Covers natural phrasings a user would actually say — "create a dashboard", "assemble a dashboard for a topic (product analytics, retention, revenue, activation, etc.)", "add/remove/replace insights" — with good variation coverage rather than jargon.

A clear niche (PostHog dashboards over MCP) with distinct triggers and explicit scoping away from single insights and widget tiles, making conflict with other skills unlikely.