Dashboard Spec

dashboard-spec

> Anchor: A dashboard is a question waiting to be answered. If the question is not written, the dashboard is decoration.

When to use

Use this skill when the user needs to specify an internal or customer-facing interface whose value is in the data it exposes. Works for:

• Admin consoles for support, billing, or operations.
• Internal data tools for finance, analytics, or compliance.
• Customer-facing account dashboards.
• Operational status pages for an ops team, a partner, or a tenant.

Do not use when:

• The user is designing a product feature for end users. Use prd-authoring.
• The user is building event tracking. Use analytics-instrumentation.
• The user is reconciling differing numbers across tools. Use data-reconciliation.

Voice rules (apply to all outputs)

• No buzzwords: leverage, empower, synergy, resonate, tapestry, delve, elevate, captivate, explore, dynamic, testament.
• No em dashes.
• No filler openings.
• Decision-oriented over descriptive.
• High-context, no overexplaining.
• Standard sentence case for formal outputs.
• ES outputs match the register of the handbook: direct, self-aware, no reverence for process, comfortable naming failure.
• EN outputs use direct US business register. Do not soften bluntness into corporate-speak.
• Language switches based on audience, not author.

Bilingual handling

• ES. Rioplatense register. Vos over tú. Keep standard terms in English: dashboard, view, empty state, role, permission, audit log, webhook, refresh, schema, SLA.
• EN. Direct US business register. Avoid decorative adjectives in specs. A dashboard is either useful or not.
• Column names, metric names, and flag names stay in the language of the data model, not the document.

Method

Step 1. Write the decision first

A dashboard exists to support a decision. Write the decision before anything else.

• Decision. What will the user do with this data.
• Cadence. How often the decision is made.
• Cost of getting it wrong. Why accuracy matters.

A dashboard that supports no decision is a screensaver. If the answer is "to be informed", rewrite the section or cut the dashboard.

Step 2. Name the user and the context

• User. The role or team, not a persona name.
• Seniority or expertise. Whether the user needs guardrails or raw power.
• Where they are. At a desk, on the floor, on call, in a customer meeting.
• Under what stress. Time pressure, accountability, frequency.

The same data, shown to a VP vs. an ops analyst vs. a customer, becomes three different dashboards. Pick one.

Step 3. List the questions it answers

Before any chart name, write the questions. Five to fifteen questions in priority order.

Example:

• "Which accounts used more than 80% of their quota in the last 7 days."
• "Which refunds were issued by which operator in the last hour."
• "Which tenants missed SLA last month, and by how much."

One question becomes one view. A question that has two answers is two questions.

Step 4. Specify the data model

For each question, name:

• Source. The table, warehouse, or API that answers it.
• Freshness. How old the data can be before the answer is wrong. Often "real-time", "5 minutes", "daily", "monthly close".
• Joins and filters. The shape of the query, at a logical level.
• Primary metrics. The values the user will read.
• Cardinality. Rough row count the view might show.

Cardinality decides layout. A view that could return 2 rows and a view that could return 2 million rows are different products.

Step 5. Design the views

One view per question, unless two questions share a natural context.

For each view:

• Name. Short, matches the question.
• Layout. Filters on top or side. Table vs. chart vs. card. Primary column. Sort order. Pagination or infinite scroll.
• Drill-down. What happens when a row is clicked.
• Time controls. The user's ability to set a time window.
• Export. CSV, copy link, or none.

Avoid the "everything in one view" trap. Visible ≠ usable.

Step 6. Handle every state explicitly

Every view must define:

• Loading. What is shown while data loads.
• Empty. What is shown when the query returns no rows. This teaches the user what the tool does.
• Error. What is shown on failure. Include who to contact.
• Partial. What is shown when one data source is up and another is down.
• Permission denied. What non-authorized users see, if they can reach the view at all.
• Large dataset. What happens when cardinality exceeds display limits.

A dashboard that has no empty-state design is a dashboard that breaks on first use.

Step 7. Define roles and permissions

• Roles. The set of roles that can access the dashboard.
• Row-level permissions. Whether a user sees all tenants or only their own.
• Write actions. Which roles can take which destructive actions.
• Impersonation. Whether admins can act as another user, and what that requires.

Include a permission matrix if more than three roles exist.

Step 8. Audit and logging

• Read audit. Whether access to the dashboard is logged.
• Write audit. Every destructive action logged with actor, timestamp, target, and before/after state.
• Retention. How long logs are kept.
• Review cadence. Who looks at the audit and when.

For support and admin tools, this section is not optional.

Step 9. Define SLA, ownership, and change policy

• Owner. Who owns the dashboard, including bugs and feature requests.
• SLA. Uptime target and data freshness target. Only set SLAs that will be monitored.
• Monitoring. Alerts on freshness, errors, and unusual query load.
• Change policy. Who can add a view, deprecate one, or change a metric definition.

Templates

• templates/en/dashboard-spec.md: full dashboard specification.
• templates/en/view-definition.md: single view inside a dashboard.
• templates/es/dashboard-spec.md: especificación completa de dashboard.
• templates/es/view-definition.md: definición de una vista dentro del dashboard.

Related skills

• analytics-instrumentation: use it when the work is defining the events behind the views, not the views.
• data-reconciliation: use it when the dashboard numbers disagree with another source.
• prd-authoring: use it instead when specifying an end-user product feature, not an internal tool.
• migration-spec: pairs with this when the dashboard is being rebuilt on a new data source.

Anti-patterns

This skill must never produce:

• A dashboard spec without a named decision and named user.
• Views named after charts ("bar chart of X") instead of questions.
• A spec that omits empty state or error state.
• A metric whose definition is only in the query.
• "Real-time" as a freshness claim without a latency target.
• A permission model that uses "admin can do everything" as the only role.
• A spec that describes UX polish before data accuracy.
• A dashboard with no audit on write actions.
• A spec that assumes the back-end will exist.

Checklist before delivery

• [ ] Decision, user, cadence, and cost of being wrong are named.
• [ ] Questions are listed before charts.
• [ ] Each view has a data source, freshness, and cardinality.
• [ ] Every view defines loading, empty, error, partial, permission-denied, and large-dataset behavior.
• [ ] Roles and permissions are explicit.
• [ ] Audit logging for write actions is defined.
• [ ] SLA and owner are named.
• [ ] No buzzwords from the voice rules appear.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: federicoimparatta
• Source: federicoimparatta/operator-skills
• License: MIT

Install and usage instructions live in the source repository linked above.