Auditing Endpoints

This skill produces a project-wide audit of the Endpoints product. Use it when the user wants to find what to clean up — unused endpoints, failing materialisations, materialised versions that nobody calls any more. It does not modify anything; it reports.

The deeper investigation per endpoint is diagnosing-endpoint-performance. The audit’s job is to find candidates and hand off.

When to use this skill

• “Audit my endpoints” / “What endpoints can I clean up?”
• The user is taking over a project and wants to know what they’ve inherited
• A periodic review (monthly / quarterly) of endpoint sprawl
• The user is over a materialisation cost budget and wants to know what to disable

The dedicated tools give a fast endpoint-level view. For call frequency, recency, and cost over time, query the query_log table with execute-sql (endpoint-level). Per-version recency comes from endpoint-versions — each version carries its own last_executed_at.

Available tools

Prefer reading from the system tables over the endpoints-get-all / endpoint-get tools — one SQL query returns the whole inventory and lets you join metadata to usage in query_log.

What counts as an issue

Usage counts only personal-API-key calls — an endpoint exercised solely from the Playground tab or the app will look unused. Per-version last_executed_at is recorded only for runs since that tracking was added, so a version can read null while still being used; always confirm before removing.

Workflow

1. List endpoints and their metadata

One execute-sql query gets the whole inventory from system.data_modeling_endpoints:

SELECT name, is_active, current_version, derived_from_insight, last_executed_at
FROM system.data_modeling_endpoints
ORDER BY name

No rows → the project has no endpoints; say so and stop. Don’t invent issues. (The last_executed_at column here is a convenience endpoint-level timestamp; for call frequency and cost, use query_log in the next step.)

2. Pull usage from query_log

query_log records every personal-API-key call, tagged with the endpoint name. One query gives recency and call counts across all endpoints:

SELECT name, count() AS calls, max(query_start_time) AS last_called
FROM query_log
WHERE endpoint LIKE '%/endpoints/%' AND is_personal_api_key_request
GROUP BY name
ORDER BY name

Cross-reference with step 1:

• In metadata, absent from query_log → never called via API key
• Last call more than 30 days ago → stale

query_log also exposes query_duration_ms, read_rows, and read_bytes per call — useful to flag expensive endpoints in the same pass. This is endpoint-level; per-version recency comes from endpoint-versions (step 3).

3. Check materialisation health and unused versions

For each materialised endpoint, call endpoint-materialization-status (this isn’t in the system tables). Surface any with status: "Failed" separately — these are active failures, not staleness.

Then call endpoint-versions and read each version’s last_executed_at: a materialised version that’s null or long stale is an unused-materialised-version candidate. Treat this as a lead, not proof — per-version recency only counts API-key runs since tracking was added, so confirm with the user before unmaterialising.

4. Present the audit

Render a prioritised report grouped by category. Don’t dump raw JSON; use a readable table per section:

## Endpoints audit — 9 issues

### 🔴 Failing materialisations (1)
- weekly_revenue (v3) — Failed 2h ago, "Column 'event_date' does not exist"
  → hand off to diagnosing-endpoint-performance

### 🟠 Never called via API key (3)
- internal_admin_query — created 5 months ago
- legacy_signup_funnel — created 1 year ago, materialised
- experiment_arm_lookup — created 9 months ago

### 🟠 Unused materialised versions (2)  [from endpoint-versions]
- monthly_active_users — v3 materialised, last_executed_at null (currently on v4 — unmaterialise v3)
- order_summary — v1 materialised, last_executed_at null

### 🟡 Stale (3)
- holiday_promo_2024 — last called 4 months ago
- ab_test_phase_1 — last called 2 months ago
- daily_revenue_cohort — last called 45 days ago

Recommended order:

The exact format is less important than: prioritised, grouped, actionable, and hand-off clear.

5. Offer the next step

End with a clear question, not a decision:

• “Want me to unmaterialise the unused versions?” — needs endpoint-update with is_materialized: false per version
• “Want me to disable the never-called endpoints?” — needs endpoint-update with is_active: false
• “Want me to dig into the failing materialisation?” — hands off to diagnosing-endpoint-performance

Never act from the audit alone. Disabling or unmaterialising affects external API consumers; always confirm before modifying.

Example interaction

User: "Audit my endpoints"

Agent steps:
- execute-sql on system.data_modeling_endpoints → 12 endpoints (4 materialised)
- execute-sql on query_log grouped by name → 8 endpoints have calls
- endpoint-materialization-status on each materialised endpoint → 1 failing
- endpoint-versions on the materialised ones → order_summary v2's last_executed_at is null

- Report:
  "Found 6 things worth looking at across 12 endpoints:

   🔴 Failing materialisations (1)
   - monthly_active_users (v2) — materialisation failed yesterday with
     'memory limit exceeded'. The endpoint is still callable inline but
     responses will be slower.

   🟠 Never called via API key (3)

Important notes

• The audit is read-only. Never call destructive tools from this flow. Hand off or confirm before any modification.
• Empty = healthy. Don’t pad an empty report with theoretical issues. “Nothing to clean up” is a good answer.
• Read with SQL, drill in with the version tool. system.data_modeling_endpoints (metadata) and query_log (endpoint-level call counts, recency, cost) via execute-sql answer most of the audit. Per-version recency comes from endpoint-versions (each version’s last_executed_at).
• API-key-only scope. Usage only counts personal-API-key calls. An endpoint exercised only from the Playground tab or the app will look unused. Always confirm before acting.
• Materialisation costs storage and compute. When an endpoint no longer needs materialisation, the cheapest fix is endpoint-update with is_materialized: false — not deleting the endpoint.
• Inactive ≠ stale. An endpoint with is_active: false was deliberately turned off. Don’t recommend deletion unless the user confirms it’s truly abandoned.