Verihubs is Poor to agents.
Discry independently scored how well an AI agent can discover and understand the Verihubs API from what’s public — not whether it’s usable. Below: every signal we checked, what’s costing the score, and what to change.
SCORED UNDER RUBRIC 1.2 · A full re-launch under Discry Score 2.5 — a new behavioral instrument, not comparable to these scores — is in progress.
Discovery
45% of score · 50/100Comprehension
55% of score · 66/100What we found
- An agent landing anywhere on the docs gets actively steered to machine-readable content: llms.txt exists on both the marketing site and the docs, and even the docs 404 page tells AI agents to visit llms.txt — a deliberate agent-readiness investment that is rare in this corpus.
- Every documentation page is retrievable as clean markdown via a .md suffix (ReadMe content negotiation), so an agent can read the entire guide and reference set without HTML parsing.
- There is no downloadable OpenAPI spec: openapi.json 404s on both domains and no spec link exists anywhere, so an agent must stitch the API surface together from per-endpoint OpenAPI fragments embedded in ~45 separate reference pages.
- Error codes are richly enumerated with descriptions, but no error includes recovery guidance — an agent hitting INSUFFICIENT_QUOTA or a 2 RPS Face-service rate limit is told only to contact Customer Success, not what to do programmatically.
- Field naming is inconsistent within single endpoints — snake_case body fields (is_quality, error_code) sit next to PascalCase coordinates (TopLeftX, BottomRightY) — and duplicate v1/v2 reference pages with '-1' suffixes make it easy for an agent to land on the wrong version.
What to change
Prioritized by impact on discoverability. You (or your docs platform) deploy these — Discry never touches your API.
- 01Publish a downloadable OpenAPI spec (one file per service is fine) and link it from docs.verihubs.com and both llms.txt files — this is the single largest discovery-score gap (weight 5).
- 02Add an llms-full.txt or a compact API overview page: the docs llms.txt is 171KB, well past agent-consumable size, and there is no page where an agent can grasp the full API scope in under 5,000 tokens.
- 03Replace schema placeholder values on reference pages with realistic, copy-pasteable request examples in at least 2-3 languages (cURL, Python, Node).
- 04Add recovery guidance to error tables: what to retry, back-off behavior for the per-service RPS limits, and how to resolve quota and payload-format failures without contacting support.
- 05Normalize field casing across endpoints (pick snake_case), remove or clearly label the duplicate v1/v2 '-1' reference pages, and add an AGENTS.md plus an MCP server listing to reach agents where they discover tools.
Execution coverage · INFORMATIONAL, UNSCORED
Whether an agent can actually complete a call and recover from errors is the deeper Audit layer — documented here, but not part of the Discry Score.
Auth via per-application App-ID/API-Key headers is documented in the reference pages. Errors are machine-parseable JSON with rich per-endpoint code enums (e.g. INVALID_PAYLOAD_IMAGE_FORMAT, INSUFFICIENT_QUOTA, FAIL_ON_1). Per-service rate limits are documented with concrete numbers (50 RPS SMS, 2 RPS Face) but no rate-limit response headers are named. Pagination and idempotency are not mentioned.