SituationalMITOriginal

AI Endpoint Smoke Test

Fires a Python urllib smoke test at the AI API endpoints on Vendors.ie, ReadingCurator and FluentAtDesk, flagging 500s and timeouts.

Entry verified April 21, 2026

The short answer

Hits seven endpoints across three sites with minimal valid payloads — vendors.ie /api/recommend, /api/smart-compare, /api/true-cost; readingcurator /api/what-to-read-after, /api/check-availability; fluentatdesk /.netlify/functions/submission-created. 400/401/405 are treated as live-but-rejecting-test-payload, not failures.

When to use it

Run as a scheduled check or after a Claude SDK update. Useful when you need confidence the Claude-powered tools are still up without triggering real user traffic — 500s or timeouts are the only results worth paging on; 400s just mean the endpoint is live and rejecting the stub payload.

Setup

  1. 1

    Save the file as ~/.claude/commands/ai-smoke-test.md.

  2. 2

    Invoke: /ai-smoke-test — no arguments.

  3. 3

    Uses urllib only, with a 15-second timeout per endpoint.

  4. 4

    OK = 200, WARN = 400/401/405 (endpoint live, payload rejected), FAIL = 500 or timeout.

  5. 5

    Do NOT flag 400s as failures — the endpoints require real auth or data to return 200.

Example

You: /ai-smoke-test
Claude: 3 endpoints tested. ✓ Vendors.ie /api/explain. ✓ FluentAtDesk /api/rewrite. 🔴 ReadingCurator /api/recommend — 500, Anthropic API key expired.

Source & attribution

Author
Bryan Collins
Licence
MIT (author's own work)
Type
Original

Original pattern published under MIT — attribution preserved by convention, not licence requirement.

Caveats

Don't include sensitive test inputs in the skill — inject from a separate secrets file.

New skills, explained plainly.

One short email when a new Claude Code skill is worth installing — and when to skip the hype. No spam.

56 skills across 10 categories, all licence-verified.