This is a Pro feature — part of the Pro engine (v2.4.0+). The free toolkit checks the llms.txt basics via site_audit; the full 8-check scoring lives in Pro.
The problem
A growing share of your traffic isn’t human and isn’t a search crawler — it’s autonomous agents. Agents read what curl reads: no JavaScript patience, no HTML error-page parsing, no guessing your API surface. A site that returns app-shell 200s, HTML error pages, and JS-only content is invisible to them.
Vercel’s is-agentic.com made this measurable — public scores for how well agents can discover, retrieve, and use a site. For reference: vercel.com scores 85, linkedin.com 47, and the reference implementation of this exact checklist took zens.ink from 64 to 98/100.
The 8 checks
| # | Check | Tier | Points |
|---|---|---|---|
| 1 | Agent-friendly 404s (real status + markdown recovery body) | essential | 15 |
| 2 | OpenAPI spec (operationIds, descriptions, typed schemas) | essential | 15 |
| 3 | JSON error responses (structured, never HTML) | essential | 15 |
| 4 | Markdown negotiation (Accept: text/markdown + Vary) | essential | 15 |
| 5 | Developer discoverability (/developers/ + llms.txt links) | recommended | 10 |
| 6 | Agent instructions (when-to-use in llms.txt) | recommended | 10 |
| 7 | Trust anchors (about/contact/privacy + Organization schema) | recommended | 10 |
| 8 | CLI / MCP surface documented | recommended | 10 |
Usage
python3 -m zens_ink_pro.agent_readiness https://yoursite.com
# Save report + JSON
python3 -m zens_ink_pro.agent_readiness https://yoursite.com --report agent-report.md --json agent-data.jsonDeterministic scoring — no LLM calls, no API keys, ~10 HTTP requests per run. Every finding names the exact change needed and its point value, bucketed P0 (critical) / P1 (high) / P2 (polish).
After fixing
Rescan with the official public scorer to confirm:
npx is-agentic yoursite.comWhat good looks like
The zens.ink implementation (August 2026) is the reference: OpenAPI 3.1 at /openapi.json, markdown on Accept: text/markdown with Vary: Accept, JSON 404s with hints, trust-anchor pages in four languages, and a llms.txt that opens with agent guidance. 98/100, grade “Strong technical baseline” — every fix pattern is documented in the Pro package’s references/ directory.