Anti-bot — defense in depth (CF WAF + Turnstile + Hono rate-limit)

Bot defense: CF WAF + Turnstile + Hono rate-limit. All free

Anti-bot — defense in depth (CF WAF + Turnstile + Hono rate-limit)

Decision

The family runs three anti-bot layers, each at a different stage of the request lifecycle. A request must pass all three to reach a route handler. All three are free, no card, and run on infrastructure the family is already using.

Why three layers (not one)

The user’s direction was ”+ Turnstile (already locked)”. The two sibling layers (WAF + rate-limit) ride alongside because:

• Each catches what the other misses — WAF can’t see per-route intent (everything looks like the same zone); Turnstile only fires on form-submit (most API endpoints aren’t form-submits); Hono rate-limit can’t see traffic the WAF already dropped at the edge.
• Each is on different substrate — a misconfiguration / outage / quota cliff on one layer is recoverable from the other two: WAF down ? Turnstile + rate-limit still gate; Turnstile rejected by region ? hCaptcha takes over; Worker over-quota ? WAF still drops the floor.
• All three are free on the family’s existing Cloudflare account — zero incremental cost, zero new vendors.

This is the same defense-in-depth pattern as the double security-headers audit (securityheaders.com + Mozilla Observatory) and the two-captcha pair (Turnstile primary + hCaptcha fallback).

Layer detail

1. Cloudflare WAF + Bot Fight Mode (edge)

• Configured per zone (oriz.in, *.oriz.in).
• Free managed ruleset; auto-updated by Cloudflare.
• Bot Fight Mode flags + challenges signature-known bots.
• Coarse zone-wide rate-limit rule (10K req/10min) as DDoS safety net.
• Runs before any Worker / Pages serve; zero CPU on origin.

2. Turnstile + hCaptcha fallback (form-submit)

• The shared <Captcha> component in gates every unauthenticated POST surface — <ContactForm>, sign-up, comment boxes — per the locked Turnstile + hCaptcha pair decision.
• The Worker verifies the token server-side via challenges.cloudflare.com/turnstile/v0/siteverify (or hCaptcha’s equivalent) before reading the form payload.
• App Check + reCAPTCHA Enterprise still front Firestore writes — different attack surface, different layer; not affected by this decision.

3. Hono rate-limit middleware (API per-route)

• Fine-grained per-route, per-IP, sliding-window throttle.
• Lives in @chirag127/oriz-kit/server so every Worker route imports the same middleware.
• Backed by Workers KV — runs on existing Worker + KV free tier per the worker quota mitigation playbook.
• Each route declares its own budget (10/min for contact, 100/min for OG, 1000/min for feed reads).

Implications

• Every site inherits all three layers automatically — they ride on shared infrastructure (Cloudflare zone, oriz-kit, api.oriz.in Worker), no per-site configuration beyond mounting <Captcha> in forms.
• CSP allow-list in _headers preset already permits challenges.cloudflare.com (Turnstile) and hCaptcha origin; no per-site CSP exception.
• Observability — the api.oriz.in Worker logs rate-limit trip events to Axiom; Cloudflare WAF event log is queryable in the dashboard. PR-time alarms not yet wired (TODO).
• Incident playbook — if a flood breaks through (bypasses all three layers), the response is:
  1. Tighten Hono rate-limit on the affected route in oriz-kit, ship.
  2. Add a Cloudflare custom WAF rule (free plan: up to 5) for the observed pattern.
  3. Flip the affected form’s <Captcha> to interactive mode (Turnstile interactive challenge) instead of invisible.
• No card-on-file — none of the three layers requires the Cloudflare Workers Paid plan, Cloudflare Pro plan, or any other paid Cloudflare product. The family stays on the free-plan posture.

Cross-refs

• Cloudflare WAF service
• Cloudflare Turnstile service
• hCaptcha service — Turnstile fallback
• Hono rate-limit service
• Captcha pair decision (Turnstile + hCaptcha) — sibling
• Security headers strategy decision — adjacent layer
• Umbrella Hono Worker decision
• CF Worker quota mitigation playbook
• No card-on-file rule

Edit on GitHub · Back to index