Cloudflare security settings, rule order and origin exposure
Cloudflare acts only on traffic that passes through it, and the security features built on its Ruleset Engine run in a fixed order.
This guide explains the zone settings that decide both: proxied DNS, the encryption mode to your origin, managed WAF rules, bot controls, rate limits, cache rules, origin authentication and the headers your origin trusts. It comes with a worksheet for your own zone, including owner-run checks of whether your origin answers a request sent around Cloudflare and which visitor address it then believes.
In this guide
Cloudflare zone configuration and rule-order worksheet
Use one spreadsheet copy of this table for each zone. Each row names an area, where to look, what to write down, what Cloudflare's documentation says about it and the evidence to file.
How to use itHide the instructions
Rows 4, 5, 15 and 16 send requests from your own machine, each sent once, to a hostname in your own zone or to your own origin: one per hostname in rows 4 and 5, one or two in row 15 and one in row 16. Row 12 is a page view in your own browser, and every other row reads a setting, list, log, file on your own server or policy page. Record that a token, key, certificate or tunnel credential exists, its scope and who holds it, never its value. Each plan condition is what Cloudflare's documentation said when read on 14 November 2026, and this page calls the plan Cloudflare lists after Business the fourth plan.
20 rows
Area
Where to look
1. Proxied DNS records
- Where to look
- The DNS Records page. Cloudflare's exposed IP addresses page says it shows a banner when the zone has records that are not proxied.
- What to record
- Each A, AAAA and CNAME record with its name, its proxy status and whether it serves web traffic. Mark every web record that is DNS-only, and note names that have several A or AAAA records.
- What Cloudflare's documentation says
- Cloudflare's proxy status page: only A, AAAA and CNAME records can be proxied, Cloudflare recommends proxying all that serve web traffic, and a DNS-only record is answered with the origin's actual address. If one A or AAAA record on a name is proxied, all on that name are treated as proxied.
- Evidence to keep
- A dated export or screenshot of the DNS table with the proxy status visible.
2. DNS-only records that share the origin
- Where to look
- The same DNS table, read for DNS-only records such as mail, SPF and TXT records.
- What to record
- Any DNS-only record whose content is, or names, the origin's address, any mail service on the same server as the website, and the date the origin's address last changed.
- What Cloudflare's documentation says
- Cloudflare's exposed IP addresses page: a DNS-only record pointing to the same origin server as a proxied record reveals its address to a dig query. Cloudflare's origin protection guide: review existing DNS-only records, SPF and TXT among them, for origin address information; if possible, keep mail off the server that hosts the website; and rotate origin addresses once onboarded, since historical records would still hold the addresses used before.
- Evidence to keep
- Record names and types, and the date of the last address change.
3. Encryption mode, zone-wide and per hostname
- Where to look
- The SSL/TLS Overview page, which Cloudflare's encryption modes page names as where the mode is chosen, and the zone's configuration rules.
- What to record
- The mode in effect for the zone, whether it was chosen or set by Automatic SSL/TLS, and every configuration rule that sets a different mode for a hostname.
- What Cloudflare's documentation says
- Cloudflare's encryption modes page strongly recommends Full or Full (strict) where possible and says a configuration rule can set the mode for specific hostnames. Flexible uses HTTP to the origin, Full does not validate the origin certificate, and Full (strict) does. Cloudflare's Strict (SSL-Only Origin Pull) page lists that mode for fourth plan zones only.
- Evidence to keep
- A dated screenshot of the mode and a list of hostname-level overrides.
4. Visitor-side TLS (sends one request per hostname)
- Where to look
- The Edge Certificates page, which Cloudflare's pages for HSTS, Always Use HTTPS and Minimum TLS Version each name. Then, once from your own machine for each proxied web hostname, Cloudflare's own example: curl https://www.example.com -svo /dev/null --tls-max 1.1, with your hostname in place of www.example.com. Only hostnames in your own zone, on port 443 as the command sends, one request each, no other ports, no address ranges, no repeats and no load.
- What to record
- Whether Always Use HTTPS is on, whether HSTS is on with its max-age and includeSubDomains values, and the minimum TLS version. For each hostname, whether the TLS 1.1 handshake completed.
- What Cloudflare's documentation says
- Cloudflare's page for each of these settings lists it on every plan in its availability table, with a per-hostname minimum TLS version included with Advanced Certificate Manager. Its Minimum TLS Version page says a blocked version does not complete the handshake, and notes that a local VPN or device security client may prevent connections that use legacy protocols. Its HSTS page says subdomains that do not support HTTPS become inaccessible with includeSubDomains on.
- Evidence to keep
- The settings screenshot and, per hostname, the line of curl output showing whether the handshake completed.
5. Response headers through Cloudflare (sends one request per hostname)
- Where to look
- Once from your own machine for each proxied web hostname: curl -I https://www.example.com, with your own hostname. curl's man page describes -I as fetching the headers only. Only hostnames in your own zone, on port 443 as the command sends, one request each, no other ports, no address ranges, no repeats and no load.
- What to record
- Whether the response carries Strict-Transport-Security, its cf-cache-status value, and the other security headers you expect. Whether the Add security headers Managed Transform is on.
- What Cloudflare's documentation says
- Cloudflare's HSTS page says the header is served on HTTPS requests only. Its Managed Transforms page lists the headers Add security headers adds, a Content Security Policy not among them, and says adding them may have an impact on the website, such as blocking resources or triggering certificate errors.
- Evidence to keep
- The response headers saved as text with the time the request was sent, with any Set-Cookie value removed.
6. Origin CA certificates
- Where to look
- The Origin Certificates tab of the Origin Server page, which Cloudflare's origin CA page names as where certificates are created, and the certificate file on each origin server.
- What to record
- Each origin certificate's hostnames, expiry date, the server that uses it and who renews it. Where the private key is stored, never the key.
- What Cloudflare's documentation says
- Cloudflare's origin CA page lists these certificates on every plan in its availability table and, as read on 14 November 2026, says Cloudflare does not send expiration notifications for them. The private key cannot be seen after the creation screen is closed.
- Evidence to keep
- Hostnames, expiry dates and server names, and the renewal reminder you set.
7. Custom rules and Skip rules, in order
- Where to look
- The zone's security rules, filtered to custom rules. Cloudflare's custom rules page says custom rules are one of the types of security rules in its security dashboard.
- What to record
- Each custom rule in list order with its action and what it matches. Flag every Skip rule and the phases it skips. Note any account-level ruleset, which runs before zone-level rulesets in each phase.
- What Cloudflare's documentation says
- Cloudflare's interoperability page gives the phase order: HTTP DDoS Attack Protection, custom rules, rate limiting rules, Managed Rules, Super Bot Fight Mode. A terminating action stops later phases, and Skip can bypass rate limiting, Managed Rules and Super Bot Fight Mode. Its custom rules page lists 5 rules on Free, 20 on Pro, 100 on Business and 1,000 on the fourth plan, with Log only on the fourth plan; the interoperability page's e-commerce scenario says a Pro or Business plan adds custom rules.
- Evidence to keep
- The ordered rule list, with the owner and reason for each Skip rule.
8. Controls outside the phase order
- Where to look
- The zone's IP Access Rules, Zone Lockdown, User Agent Blocking, Browser Integrity Check, Hotlink Protection and Security Level settings.
- What to record
- Each setting's state, each rule it holds, and who added it.
- What Cloudflare's documentation says
- Cloudflare's interoperability page says these are not powered by the Ruleset Engine, are evaluated independently and do not follow the phase order. Its Bot Fight Mode page says Bot Fight Mode will not trigger if an IP Access rule matches the request first.
- Evidence to keep
- A screenshot or export of each list.
9. Managed WAF rules and exceptions
- Where to look
- The Security rules page filtered by Managed rules, which Cloudflare's Managed Ruleset page names as where a deployment is configured.
- What to record
- Which managed rulesets are deployed, the scope of each, any ruleset, tag or rule overrides, and each WAF exception with what it skips and its place in the list.
- What Cloudflare's documentation says
- Cloudflare's managed rules page lists the Free Managed Ruleset on every plan and the Cloudflare Managed Ruleset and OWASP Core Ruleset on Pro, Business and the fourth plan. Its Managed Ruleset page says some rules are disabled by default. Its managed rules page says exceptions have priority over overrides, and that a skip rule only bypasses the execute rules listed after it.
- Evidence to keep
- Dated screenshots of each deployment's settings and of the exceptions list.
10. Bot controls
- Where to look
- The Security Settings page filtered by Bot traffic, which Cloudflare's Bot Fight Mode and Super Bot Fight Mode pages name as where each is turned on.
- What to record
- Which bot product is on; where Super Bot Fight Mode is used, the action for Definitely automated, Likely automated and Verified bots; each Skip rule that exempts traffic from it; and which hostnames serve APIs, mobile apps or tunnels.
- What Cloudflare's documentation says
- Cloudflare's interoperability page lists Bot Fight Mode on Free and Super Bot Fight Mode on Pro, Business and the fourth plan without Bot Management. It says custom rules cannot skip Bot Fight Mode and can skip Super Bot Fight Mode with the Skip action, while the Super Bot Fight Mode page's considerations say both products cannot be customised, adjusted or reconfigured through WAF custom rules. Cloudflare says both may challenge API or mobile app traffic, and to keep Definitely Automated set to Allow where Cloudflare Tunnel is used.
- Evidence to keep
- A screenshot of the bot settings and the list of exempted traffic with reasons.
11. Rate limiting rules
- Where to look
- The zone's rate limiting rules.
- What to record
- Each rule's expression, what it counts by, its period, threshold, action and duration, and the endpoints it covers, such as sign-in and password reset.
- What Cloudflare's documentation says
- Cloudflare's rate limiting page says the rules are not designed to let a precise number of requests reach the origin, and lists counting by IP on Free and Pro, with IP with NAT support added on Business. Its request rate page says counters are not shared across data centres, except between data centres associated with one geographical location, and its parameters page gives 429 as the default response when the action is Block.
- Evidence to keep
- The rule list with its settings. No test traffic in this row; the rate limiting guide covers how limits are tested.
12. Cache rules and signed-in pages (one page view)
- Where to look
- The zone's cache rules. Then, signed in to your own site as yourself, one page that shows account details, opened once, with your browser's developer tools open on the network view.
- What to record
- Each cache rule that makes HTML or JSON eligible for cache or sets an Edge Cache TTL, the hostnames and paths it covers, and whether Cache Deception Armor is on in it. For the signed-in page, its Cache-Control and cf-cache-status values.
- What Cloudflare's documentation says
- Cloudflare's default cache behaviour page says it does not cache HTML or JSON by default, and that an Edge Cache TTL cache rule can override the origin's cache headers. Its Cache Deception Armor page says the protection depends upon Origin Cache Control, and warns that a Cache-Control header sent by the origin, or an Edge Cache TTL cache rule, may take precedence over it. Its cache responses page defines HIT, MISS, BYPASS and DYNAMIC, among other values.
- Evidence to keep
- The rule list, and the page path with its two header values. Never the cookie or the page content.
13. Origin authentication
- Where to look
- The zone's Authenticated Origin Pulls setting at each level, the origin web server's TLS configuration, the origin firewall's inbound rules, and Networking, then Tunnels, which Cloudflare's tunnel permissions page names as where tunnels are listed.
- What to record
- Which method each origin uses: an allow list of Cloudflare's addresses, Authenticated Origin Pulls at global, zone or per-hostname level and whose certificate, or Cloudflare Tunnel. For each, whether both halves are in place: the Cloudflare setting and the origin's own check.
- What Cloudflare's documentation says
- Cloudflare's AOP page lists AOP on every plan in its availability table, says it does not apply with Off or Flexible, and says the shared global certificate proves a request came from Cloudflare's network, not from your account. Its origin protection guide rates an IP allow list moderately secure and vulnerable to spoofing. Its Tunnel page lists Tunnel on all plans, and its AOP page says AOP settings have no effect on a hostname reached through Cloudflare Tunnel.
- Evidence to keep
- The method per hostname, the client certificate's expiry, firewall rule names, and tunnel names with the hostnames they serve. If a configuration file read here holds a secret, record only that it is present.
14. Trusted headers
- Where to look
- The origin web server's visitor address settings, such as RemoteIPTrustedProxy on Apache or set_real_ip_from on nginx as in Cloudflare's instructions, the application's setting for which header it reads, and the zone's Managed Transforms.
- What to record
- Which header the web server and application take the visitor address from, which source addresses they trust it from, when that list was last compared with Cloudflare's published ranges, and whether "Remove visitor IP headers" or the True-Client-IP transform is on. If the configuration file holds a secret, record only that it is present.
- What Cloudflare's documentation says
- Cloudflare's restoring visitor IPs page shows CF-Connecting-IP read only from trusted proxy addresses filled with Cloudflare's ranges, and says the nginx list needs to be updated regularly. Its HTTP headers page recommends CF-Connecting-IP or True-Client-IP over X-Forwarded-For, and lists True-Client-IP for the fourth plan only.
- Evidence to keep
- The relevant configuration lines with any secret removed, and the date of the range comparison.
15. Owner-run check: does the origin answer directly? (one or two requests)
- Where to look
- From your own machine, one request to your own origin's address naming your own hostname: curl -I --resolve www.example.com:443:ORIGIN-ADDRESS https://www.example.com, with your hostname and your origin's address in place of the examples. curl's man page says --resolve makes the request use the address you give for that host and port. Only if the origin also listens on port 80, one more request with www.example.com:80:ORIGIN-ADDRESS and an http:// address. Only your own origin's address and your own hostname, from your own machine, no other ports, no address ranges, no repeats and no load.
- What to record
- Whether each request was refused, timed out, stopped at a certificate error, or returned your site. If the origin is a hosted platform and you do not know its address, write that and skip the row.
- What Cloudflare's documentation says
- Cloudflare's IP addresses page says someone who discovers the origin's address could send traffic directly, bypassing its protections, and recommends blocking traffic that does not come from Cloudflare's addresses or those of trusted partners, vendors or applications. curl's man page says that without -k curl verifies the server's TLS certificate before it continues. Cloudflare's scans and penetration testing policy limits testing to customer-owned IPs, Cloudflare's designated public IPs and the customer's registered DNS entries.
- Evidence to keep
- The command, the time, and the first response line or the error. No other ports, no address ranges, no repeats. If your origin is with a hosting provider, a note that you read its terms first.
16. Owner-run check: does the origin trust a header it should not? (one request, only if row 15 returned your site)
- Where to look
- One more request to the same address with an extra header naming an invented visitor address: curl -I --resolve www.example.com:443:ORIGIN-ADDRESS -H "CF-Connecting-IP: 203.0.113.50" https://www.example.com. Then the origin's own access log for that request. curl's man page says -H adds the header to the request. Sent once, from your own machine, to the same origin address and hostname as row 15, with no other addresses or ports, no repeats and no load.
- What to record
- Which address the log line for that request shows: your own machine's, or the invented one. If it shows the invented address, the origin took the header from a sender that was not Cloudflare.
- What Cloudflare's documentation says
- Cloudflare's restoring visitor IPs page says the original visitor address arrives in CF-Connecting-IP, and shows web servers configured to trust it only from Cloudflare's addresses. Its Caddy instructions advise accepting traffic only from Cloudflare's IP addresses, because otherwise the header could be spoofed.
- Evidence to keep
- The log line's address field and time, and nothing else from the log. If the line holds a secret, such as a token in a URL, record only that it is present.
17. Members, roles and two-factor sign-in
- Where to look
- The account's member list with each member's roles, and My Profile, then Authentication, which Cloudflare's two-factor page names for each user's own setting.
- What to record
- Each member with their roles and scope and whether they still need access, how many hold Super Administrator - All Privileges, and whether 2FA Enforcement is on.
- What Cloudflare's documentation says
- Cloudflare's roles page says Super Administrator - All Privileges can manage members and all account-owned API tokens, and that Administrator cannot manage members. Its two-factor page says a Super Administrator can turn on 2FA Enforcement for all members.
- Evidence to keep
- The dated member list with roles.
18. API tokens, the Global API key and tunnel credentials
- Where to look
- My Profile, then API Tokens, for user tokens and the Global API key; Manage account, then Account API tokens; and Networking, then Tunnels. Cloudflare's pages on creating tokens, account API tokens, the Global API key and tunnel permissions name these places.
- What to record
- Each token by name, owner, policy (accounts or zones, Read or Edit), expiry, client IP filter and what uses it. Whether anything still authenticates with a Global API key. Each tunnel by name, its hostnames, and who holds a role that can manage it. Do not press the Global API key's View or Change button. On a tunnel's Overview tab, do not select Add a replica, which Cloudflare's tunnel permissions page says reveals the installation command holding the token, or Refresh token, which replaces it.
- What Cloudflare's documentation says
- Cloudflare's create token page says a token secret is shown once and anyone with it can perform the actions it allows. Its Global API key page says the key has the same permissions as the user and recommends API tokens instead. Its tunnel permissions page says anyone with a remotely-managed tunnel's token can run the tunnel.
- Evidence to keep
- A dated list of token names, scopes, expiry dates and owners, and tunnel names with their hostnames. Never a secret.
19. Audit log
- Where to look
- The Audit Logs page, which Cloudflare's audit log page names.
- What to record
- Changes since the last review to DNS records, encryption mode, rules, bot settings, members and tokens, with who made each and when.
- What Cloudflare's documentation says
- Cloudflare's version 1 audit log page says audit logs are available on all plan types, include account actions such as login and zone configuration changes, can be filtered by date and downloaded as CSV, and are kept for 18 months. It adds that most beta features do not appear until they leave beta.
- Evidence to keep
- The CSV for the review period, stored where only the account's administrators can read it. If an entry shows a secret, record only that it is present.
20. Testing policy before any active test
- Where to look
- Cloudflare's scans and penetration testing policy, your origin host's terms, and your plan.
- What to record
- The date you read the policy, the zones and hostnames in scope, the origin addresses involved, and each configuration change the policy asks for before a penetration test, with who approved it and whether it was kept afterwards.
- What Cloudflare's documentation says
- Cloudflare's policy limits scans and tests to customer-owned IPs, Cloudflare's designated public IPs and the customer's registered DNS entries, asks for scans throttled to a reasonable rate, and lists configurations to set on each zone before a penetration test. For denial-of-service tests it points to a separate page on the required notification.
- Evidence to keep
- The policy URL and read date, the owner's signed authorisation, and a record of each configuration change made for the test.
Where Cloudflare sits in front of your application
Cloudflare security starts with a routing fact.
Cloudflare describes its DNS and CDN services as what let it reverse proxy the web traffic to and from a domain, and with the default full DNS setup it becomes the primary authoritative DNS provider for that domain. Each DNS record then carries a proxy status. Once the domain is active, a proxied record is answered with Cloudflare anycast addresses, so HTTP and HTTPS requests for that name pass through Cloudflare on their way to your server.
A DNS-only record is answered with your server's actual address, and Cloudflare does not route its web traffic.
Only A, AAAA and CNAME records can be proxied; MX, TXT and other types are always DNS-only. Cloudflare recommends proxying every A, AAAA and CNAME record that serves web traffic, and says proxying is on by default when a domain is onboarded through the dashboard.
Its list of what proxying makes possible includes applying WAF rules and caching to incoming traffic, and for a DNS-only record that serves web traffic it says Cloudflare cannot optimise, cache and protect those requests. The worksheet therefore starts with the DNS table.
Some controls depend on the plan. The availability tables on the Cloudflare pages cited here have columns for Free, Pro and Business, then a further plan, which this page calls the fourth plan. The rate limiting page's table splits that plan into two columns, and this page gives neither column's detail.
Each plan condition on this page is the one the named Cloudflare page gave when it was read on 14 November 2026. Check that page again before relying on a plan.
This guide stays with the settings of a Cloudflare zone and account. Choosing a WAF, and the limits any WAF has, belong to the web application firewall guide, and separate guides in this library cover rate limit design and cache attacks; each gets a short pointer here.
What Cloudflare handles and what stays with you
Some protection runs without the customer setting it up.
Cloudflare's security features interoperability page says DDoS protection is always on for all plans and needs nothing turned on or configured. The same page describes Managed Rules as pre-configured rulesets maintained by Cloudflare, and Cloudflare's reference for its Managed Ruleset says it is created by the Cloudflare security team and updated frequently to cover new vulnerabilities and reduce false positives.
Other Cloudflare pages describe work at the origin, in the DNS table or in the application itself:
- Origin certificates. Cloudflare's encryption modes page says Automatic SSL/TLS will not move a zone to a less secure mode, giving an expired origin certificate as its example, and tells the customer: "You must ensure the validity of your origin SSL/TLS configuration at all times."
- Origin CA expiry. Cloudflare's origin CA page, as read on 14 November 2026, lists as a known limitation that Cloudflare does not send expiration notifications for origin CA certificates, and says that if you rely on long-lived origin CA certificates, you should track their expiry in your own certificate inventory or monitoring system.
- The address list at your origin. Cloudflare's instructions for restoring visitor addresses on nginx say the list of Cloudflare prefixes needs to be updated regularly.
- DNS hygiene. Cloudflare's guide to protecting an origin server advises reviewing existing DNS-only records, SPF and TXT among them, to make sure no origin address appears in them, and changing the origin's addresses once the domain is onboarded.
- Application flaws. Cloudflare's OWASP Core Ruleset page says some OWASP Top 10 risks can be addressed by a web application firewall while others require different solutions or must be mitigated during application development, naming Cryptographic Failures, Insecure Design, Identification and Authentication Failures, and Security Logging and Monitoring Failures.
The choices the worksheet records
On this page's reading, each of the settings that follow is a choice the account makes: which hostnames are proxied, how Cloudflare connects to the origin, which rules run and in what order, who can change them, and whether the origin answers requests that did not come through Cloudflare. Each of those choices becomes a row in the worksheet on this page.
Cloudflare TLS settings: the encryption mode and the origin connection
A zone's SSL/TLS encryption mode controls two connections: visitor to Cloudflare, and Cloudflare to your origin.
If possible, Cloudflare strongly recommends Full or Full (strict) to prevent malicious connections to the origin. The modes it lists under Custom SSL/TLS differ in that second connection:
- Off: no encryption between visitors and Cloudflare or between Cloudflare and the origin.
- Flexible: HTTPS from visitors to Cloudflare, plain HTTP from Cloudflare to the origin. Cloudflare's page for this mode calls the result partially secure, and says to use Full or Full (strict) instead when the application holds sensitive information, giving personalised data and user login as examples.
- Full: Cloudflare connects to the origin with the scheme the visitor used and does not validate the origin's certificate, which can be expired, self-signed or issued for another hostname. Cloudflare's page for this mode says that without Full (strict) a malicious party could technically hijack the connection and present their own certificate.
- Full (strict): Full plus validation of the origin certificate, which must be unexpired, issued by a publicly trusted certificate authority or Cloudflare's Origin CA, and match the hostname. Without such a certificate, visitors may see a 526 error.
- Strict (SSL-Only Origin Pull): always HTTPS to the origin, whatever the visitor used, with the same validation as Full (strict). Cloudflare's page for it says it is only available on zones on the fourth plan.
Reading the mode actually in effect
Two details change what the zone-wide value tells you. Cloudflare lists Automatic SSL/TLS as the default and, when its page was read on 14 November 2026, said the feature was being rolled out gradually and that zones not yet migrated show only the Custom SSL/TLS options, so record whether the mode was chosen or set automatically.
And Cloudflare says a configuration rule can set the mode for specific hostnames instead of the whole zone, so the worksheet also asks for every rule that does.
The visitor side: HTTPS redirects, HSTS and the minimum TLS version
Always Use HTTPS, HSTS and Minimum TLS Version act on the first connection, and Cloudflare's page for each lists it on every plan in its availability table. Always Use HTTPS redirects every http request to https for all subdomains and hosts, and is not shown in the dashboard while the encryption mode is Off; if only part of an application can serve HTTPS, Cloudflare says to use a single redirect instead.
HSTS adds a header that directs compliant browsers to turn HTTP links into HTTPS links and to prevent users from bypassing SSL browser warnings, for the period set by its Max Age Header.
It is sent on HTTPS responses only, and with includeSubDomains on, subdomains that do not support HTTPS become inaccessible.
Cloudflare's HSTS page warns that removing HTTPS before turning HSTS off, or before the max-age has run out, leaves the site inaccessible to visitors for the duration of the Max Age Header or until HTTPS is enabled again.
Minimum TLS Version refuses HTTPS connections from visitors that cannot use the chosen version or newer. A per-hostname value comes with Advanced Certificate Manager, Cloudflare Pages hostnames cannot have one, and the zone setting does not apply to R2 custom domains. Cloudflare's page for the setting gives a one-line curl command for testing which TLS versions a hostname supports, and the worksheet uses it.
The same page notes that a local VPN or a device security client may prevent connections that use legacy protocols.
Origin CA certificates
Cloudflare's Origin CA issues certificates for the second connection. Its page lists them on every plan in its availability table, says they are compatible with strict mode, and frames them for origins that only receive traffic from proxied records: visitors may see untrusted certificate errors if Cloudflare is paused or proxying is turned off for a subdomain that uses one, because the certificate only covers the connection from Cloudflare.
The certificate and its private key are saved as separate files, and the key cannot be seen again once the creation screen is closed.
The worksheet records each certificate's hostnames, expiry and server, and where the key is kept, never the key.
Cloudflare WAF rule order: which rule sees a request first
Cloudflare runs its Ruleset Engine security features in a fixed sequence of phases, and a terminating action such as Block or Managed Challenge ends the request there, so later phases do not evaluate it.
Its security features interoperability page gives the request phases in this order:
- HTTP DDoS Attack Protection, in the ddos_l7 phase.
- Custom rules, in http_request_firewall_custom.
- Rate limiting rules, in http_ratelimit.
- Managed Rules, in http_request_firewall_managed.
- Super Bot Fight Mode, in http_request_sbfm.
What runs outside that order
Within each phase, account-level rulesets run before zone-level ones. The interoperability page says IP Access Rules, Zone Lockdown, User Agent Blocking, Browser Integrity Check, Hotlink Protection and Security Level are not powered by the Ruleset Engine, are evaluated independently and do not follow the phase order. Bot Fight Mode sits outside the phase system too.
A further gap is easy to miss: when a Worker or snippet fetches a proxied hostname in the same zone, Cloudflare says the WAF custom rules and managed rules on that hostname do not apply to the subrequest.
Skip rules and the phases they can bypass
The interoperability page says a custom rule with the Skip action can bypass rate limiting rules, Super Bot Fight Mode and Managed Rules, and that a terminating action in custom rules prevents Super Bot Fight Mode from evaluating the request. Cloudflare's custom rules page adds that custom rules are evaluated in order, and that if an earlier rule blocks a request, later rules do not run for it.
Inside the managed phase, Cloudflare's managed rules page says WAF exceptions are evaluated in list order and a skip rule only bypasses the execute rules listed after it.
On this page's advice, every Skip rule and every exception goes in the worksheet with what it matches, what it skips, who added it and why.
Custom rule limits by plan
Cloudflare's custom rules page lists 5 custom rules on Free, 20 on Pro, 100 on Business and 1,000 on the fourth plan. The Log action, which Cloudflare's actions reference recommends for validating a rule before a more severe action, is listed only on the fourth plan. Regular expressions start at Business, and account-level custom rulesets, which apply across zones, are a fourth plan feature.
Cloudflare's interoperability page, in its scenario for an e-commerce site, says a Pro or Business plan adds Super Bot Fight Mode, custom rules and Managed Rules; this page gives each page's statement as it was read.
Finding which feature acted
To see which feature blocked a request, Cloudflare's interoperability page says to find it in the Events tab of the Analytics page and read its Service field. Its Security Events page adds that one request can produce one or more security events, that the dashboard shows those events rather than the requests themselves, and that for large volumes of traffic the sampled logs may not show every event.
Cloudflare WAF managed rules on each plan
Cloudflare's managed rules page lists what each plan may deploy.
The Cloudflare Free Managed Ruleset is on every plan and provides protection against high-impact and widely exploited vulnerabilities; the page says a zone that has deployed the Cloudflare Managed Ruleset does not need it. The Cloudflare Managed Ruleset and the Cloudflare OWASP Core Ruleset are listed on Pro, Business and the fourth plan, not on Free.
Sensitive Data Detection, which runs in a response phase, is listed on the fourth plan only. The Exposed Credentials Check is marked deprecated, with Cloudflare recommending its leaked credentials detection instead.
Deploying a ruleset is not the same as running all of it. Cloudflare's reference page for its Managed Ruleset says some rules are disabled by default to balance protection against false positives, recommends enabling the rules whose tags match your technology stack, such as wordpress, and advises against enabling every rule through overrides except in a proof of concept.
Enabling all rules through the ruleset status affects the rules that are off by default and every rule added in future.
Cloudflare's managed rules page adds that an override set at ruleset or tag level also reaches rules added later, not only those already there, and that exceptions have priority over overrides.
The OWASP Core Ruleset is Cloudflare's implementation of the OWASP ModSecurity Core Rule Set version 3.3.0. It adds up a threat score for each request and acts when the score passes the configured threshold. Cloudflare's own caution is that it is prone to false positives and offers only marginal benefits on top of the Cloudflare Managed Ruleset and WAF attack score, and that a zone that deploys it will need to monitor and adjust its settings based on its own traffic.
Managed rules inspect a request body only up to a size that varies by plan, and content past that limit may not be fully analysed. Where the plan includes request body fields, a custom rule can act on http.request.body.truncated. What a body limit means for any WAF is covered in the web application firewall guide.
Cloudflare bot protection: Bot Fight Mode, Super Bot Fight Mode and Bot Management
Bot Fight Mode, which Cloudflare's interoperability page lists as available on Free plans, is an on and off switch that challenges traffic matching patterns of known bots.
It does not run on the Ruleset Engine, so its page says no WAF custom rule or Page Rule can skip or bypass it; it does not trigger when an IP Access rule matches the request first. JavaScript Detections comes with it and cannot be turned off, which matters for a site that sends a Content Security Policy: among the steps Cloudflare lists, anything under /cdn-cgi/challenge-platform/ needs to be allowed.
Cloudflare's Super Bot Fight Mode page says it is included on Pro, Business and the fourth plan, where an account without the Bot Management add-on gets the Business version. The interoperability page says it sets separate actions for Definitely automated, Likely automated and Verified bots traffic.
It runs after custom rules in the http_request_sbfm phase, and its page says exceptions are configured with a custom rule using the Skip action; the Bot Fight Mode page points to Super Bot Fight Mode for exceptions such as your own API clients or monitoring tools.
Among its considerations, the Super Bot Fight Mode page also says Bot Fight Mode and Super Bot Fight Mode cannot be customised, adjusted or reconfigured through WAF custom rules. The same page says to turn Bot Fight Mode off when upgrading to Super Bot Fight Mode, and warns that if the organisation also uses Cloudflare Tunnel, Definitely Automated should stay set to Allow, or Super Bot Fight Mode may block tunnel connections.
The interoperability page describes Bot Management as an add-on for the fourth plan that gives every request a bot score from 1 to 99, which custom rules can then act on. Cloudflare's bot pages say Bot Fight Mode and Super Bot Fight Mode may challenge API or mobile app traffic, so the worksheet asks which of your hostnames serve APIs, mobile apps or tunnels, next to the bot settings and the Skip rules that exempt traffic.
Rate limits and cache rules at the edge
Rate limiting rules run after custom rules and before managed rules.
Cloudflare's rate limiting page says they are not designed to let a precise number of requests reach the origin: updating the counters can lag detection by up to a few seconds, so excess requests could still get through before the action applies.
What a rule can count by depends on the plan: the page lists IP on Free and Pro, with IP with NAT support added on Business. Cloudflare's page on request rate calculation says counters are not shared across data centres, with an exception for data centres associated with one geographical location, and its parameters page gives 429 as the default response when a rate limiting rule blocks. How to choose limits for sign-in and costly endpoints is covered in the rate limiting guide.
Cloudflare's default cache behaviour page says it follows the origin's cache headers, which an Edge Cache TTL cache rule can override, caches by file extension rather than MIME type, and does not cache HTML or JSON by default.
Among the cases in which the same page says it does not cache a resource are a response that carries Set-Cookie and a Cache-Control of private or no-store. On this page's advice, a cache rule that makes HTML eligible for cache or overrides the origin's headers is the first place to look when checking whether a signed-in page could be stored and served to someone else.
For web cache deception, Cloudflare documents Cache Deception Armor, turned on inside a cache rule. The rule compares a URL's extension with the Content-Type that comes back, and Cloudflare does not cache the response when the two disagree in a way that could allow a web cache deception attack, with exceptions its page lists.
The same page warns that the protection may be overridden by the origin's own Cache-Control header or by an Edge Cache TTL cache rule.
Cloudflare's cache responses page says the cf-cache-status header shows whether a resource is cached, with values that include HIT, MISS, BYPASS and DYNAMIC. The web cache deception guide explains how those attacks are tested.
Cloudflare origin security: the route around the proxy
Everything above acts on requests that arrive through Cloudflare.
Cloudflare's IP addresses page states the gap directly: if someone discovers the origin's address, for example through historical DNS records or mail server configuration, they could send traffic straight to the server, bypassing Cloudflare's security protections entirely. Proxying keeps the address out of DNS answers, but Cloudflare's exposed IP addresses page says it is still possible, though more difficult, to find an origin behind the proxy, and that a DNS-only record pointing at the same server reveals the address to anyone who looks it up.
Cloudflare's origin protection guide lists several ways to make the origin turn away requests that did not come through Cloudflare. Among them are these, and they rest on different checks:
- Allow only Cloudflare's addresses. Cloudflare's IP addresses page recommends blocking all traffic that does not come from its IP addresses or those of trusted partners, vendors or applications. Its origin protection guide rates this moderately secure and notes it is vulnerable to IP spoofing. The IP addresses page also says Cloudflare's address ranges are shared by all proxied hostnames.
- Authenticated Origin Pulls. Cloudflare's page describes AOP as mTLS, with Cloudflare presenting a client certificate to the origin. It works with Full or Full (strict), not with Off or Flexible, and Cloudflare's page lists it on every plan in its availability table. Global AOP uses a Cloudflare-provided certificate shared across all accounts, which Cloudflare says proves a request came from its network but not from your account; to tie requests to your account, it says to use zone-level or per-hostname AOP with your own certificate. Each level is enabled separately and each needs configuration on the origin server as well, so the worksheet records both halves.
- Cloudflare Tunnel. A cloudflared daemon on the server makes outbound-only connections to Cloudflare, Cloudflare's Tunnel page lists it on all plans, and its Cloudflare One documentation says the server's firewall can then allow only those outbound connections and block all inbound traffic. Authenticated Origin Pulls has no effect on tunnelled hostnames, where the tunnel's connector credentials authenticate origin traffic instead. Cloudflare says anyone with a remotely-managed tunnel's token can run the tunnel, and recommends rotating the token at a regular cadence.
Closing the DNS trail
The same origin protection guide goes further than the record audit above. If possible, it says, keep a mail service off the server that hosts the web resource, since mail sent to an address that does not exist is bounced back and reveals the mail server's address. It also says DNS records are in the public domain and historical records are kept, which is why it advises rotating origin addresses once onboarded.
The worksheet then checks the result from outside with two owner-run rows that send single requests from your own machine to your own origin.
Trusted headers: the visitor address your origin believes
Cloudflare's instructions for restoring visitor addresses say that, because Cloudflare acts as a reverse proxy, the origin logs a Cloudflare address by default, and that the original address arrives in a header called CF-Connecting-IP.
They show web servers configured to take it only from trusted proxy addresses: RemoteIPTrustedProxy for Apache's mod_remoteip and set_real_ip_from for nginx, each filled with Cloudflare's published ranges. The nginx instructions add that the list needs to be updated regularly.
Cloudflare's HTTP headers reference explains the alternatives. When a request reaching Cloudflare already carries an X-Forwarded-For header, Cloudflare appends the address of the proxy that connected to it. Cloudflare recommends that logs and applications read CF-Connecting-IP or True-Client-IP instead, and says each has a consistent format holding only one address.
True-Client-IP is listed for the fourth plan only. Cloudflare's Managed Transforms reference says the "Remove visitor IP headers" Managed Transform removes cf-connecting-ip and true-client-ip from requests to the origin, removes the visitor address from x-forwarded-for only when another CDN proxied the request first, and cannot be used together with the transform that adds True-Client-IP.
In its instructions for Caddy, the same restoring visitor addresses page advises also accepting traffic only from Cloudflare's IP addresses, because otherwise the header could be spoofed. On this page's advice, treat every web server and application that reads CF-Connecting-IP the same way: if it takes the header from any address, assume a request sent straight to the origin can set it, and that the value then reaches the logs, rate limits and allow lists that read it.
The worksheet therefore records which header your web server and application trust, and from which addresses, next to the result of the direct-origin check.
Account roles, two-factor sign-in, API tokens and the audit log
Every setting above can be changed by anyone whose role or token allows it, so the account belongs in a review of the zone.
Cloudflare's roles page lists Super Administrator - All Privileges as able to edit any setting, manage members, and create and manage all account-owned API tokens, and the Administrator role as reaching the full account without managing members or the billing profile. Other roles cover narrower jobs, among them Firewall (WAF, IP Access rules, Zone Lockdown, Cache Rules and Cache Response Rules), DNS, Cloudflare Access (which can also edit Cloudflare Tunnel) and Audit Logs Viewer, while Administrator Read Only can see the full account in read-only mode.
Cloudflare recommends two-factor authentication for every user, and a Super Administrator can turn on 2FA Enforcement to require it of all members. Its page describes security keys as phishing-resistant, and recommends at least two different factors plus safely stored backup codes.
API access comes through user tokens, account API tokens and the legacy Global API key. User tokens act for a user with a subset of that user's permissions. Account API tokens can act as service principals with their own permissions, which Cloudflare's page describes for integrations that need to keep working after the user who configured them has left, and only a Super Administrator can view all of them; other members see the ones they created.
Cloudflare's token creation page gives each token a policy naming the accounts or zones it covers and its permissions, most of which offer Edit or Read, and lets token use be limited by expiry and client IP address.
The Global API key is the legacy scheme: it has the same permissions as its user across all of that user's resources, one per user, without those limits, and Cloudflare recommends API tokens instead.
A token secret is shown once, and Cloudflare says anyone holding it can perform the actions it allows.
Audit Logs, which Cloudflare's version 1 audit log page lists on all plan types, record account actions such as login as well as zone configuration changes. They can be searched by user email or domain, filtered by date and downloaded as CSV, and are kept for 18 months.
The same page notes that most beta features do not appear in them until they leave beta, and that Audit Logs version 2 is available.
Cloudflare's scans and penetration testing policy
Cloudflare's scans and penetration testing policy, which showed a last-updated date of 20 April 2026 when read on 14 November 2026, says customers may scan and test application and network-layer aspects of their own assets, such as their zones within their Cloudflare accounts, with certain restrictions and provided they follow the policy.
All scans or testing must be limited to customer-owned IPs, Cloudflare's designated public IPs and the customer's registered DNS entries. Cloudflare-owned destinations such as *.cloudflare.com are allowed only as part of its Public Bug Bounty program.
Scans should be throttled to a reasonable rate and should identify the presence of vulnerabilities without attempting to actively exploit them, and the policy recommends leaving /cdn-cgi/ endpoints out.
Before a penetration test, the policy asks for security configurations on each zone tested: the Cloudflare Managed Ruleset deployed with all rules enabled, the OWASP Core Ruleset at paranoia level PL4 with a score threshold of High, 25 and higher, custom rules on the WAF attack score and on malicious uploads, Super Bot Fight Mode on Pro and Business plans without Bot Management, or Bot Management enabled where the zone has it, and rate limiting rules on key endpoints.
Each of those changes the zone's configuration, so on this page's advice the account owner decides on them outside the worksheet, which only records the decision. The policy also says other features active on the account or zone can influence test results.
The policy's important remarks bear on reading a scan report for a zone behind Cloudflare: Cloudflare's anycast network will report ports other than 80 and 443 as open, which the policy calls expected behaviour that does not indicate a vulnerability, and findings about the ROBOT vulnerability are false positives for assets behind Cloudflare.
For denial-of-service tests the policy points to a separate Cloudflare page on simulating test DDoS attacks for the required notification. Under its additional resources, it says customers can download a Penetration Test Report of Cloudflare from the dashboard.
The owner-run rows in the worksheet send individual requests rather than run a scan: one per hostname in rows 4 and 5, one or two in row 15 and one in row 16, each to your own hostname or your own origin's address.
Keep them inside the policy's limits all the same, to your own zone and your own origin's address. If your origin runs with a hosting provider, check that provider's terms before sending anything to it.
How to fill in the zone worksheet
Fill in a separate worksheet for each zone, with that zone open in the dashboard and signed in as a member whose role can read every setting the rows name.
When your plan lacks a control, say so in the cell, naming the plan and the Cloudflare page that gives the condition: an empty cell cannot be told apart from a row nobody checked.
- Rows 4, 5, 15 and 16 send requests, and row 12 is a page view in your own browser while signed in as yourself. Every other row reads something: a setting, a list, a log, a file on your own server or a policy page.
- Each request goes once, from your own machine, to a hostname in your own zone or to your own origin's address. No address ranges, no other ports than the row names, no repeated runs and no load.
- Record that an API token, origin certificate, private key or tunnel credential exists, its scope and who holds it, never its value. Leave the Global API key's View button, the token secret screen and a tunnel's installation command alone, because each displays a credential, and leave the Global API key's Change button and a tunnel's Refresh token option alone, because each replaces one.
- Where a row reads a web server configuration file or a log, record whether a secret is present and where, never its value, and do not print the file to a shared screen.
- Put the date on each screenshot and keep the link to the Cloudflare page behind each plan condition. Strip out the names of anyone outside the business before the worksheet leaves the team.
- Fill in a fresh copy when the plan changes, a hostname, member or token is added, the origin moves, or the encryption mode or rule order changes.
What a review of your own zone cannot show
The worksheet records configuration and a handful of responses.
It shows which hostnames pass through Cloudflare, how Cloudflare reaches the origin, which rules run in what order, and who can change them. What your application then does with a request the rules let through is outside it: whether one customer can read another's records, whether a session ends when it should, or whether an API returns more than the screen needs.
Cloudflare's OWASP Core Ruleset page says risks such as Insecure Design and Identification and Authentication Failures depend more on how the application is built or how the monitoring pipeline is set up.
A rule log is not proof either way. Cloudflare's Security Events page says it covers requests that Cloudflare security products acted on or flagged, shows events rather than requests, and uses sampling for large volumes of traffic. On this page's reading, a quiet log is therefore not a record that nothing was tried, and a rule that blocked one test payload says nothing about a variation it never saw.
Nor does the worksheet reach Cloudflare's platform. Evidence about that comes from Cloudflare, such as the penetration test report its policy says customers can download. And it is a snapshot: the audit log is the record of what changed after the day you filled it in.
Where a Clever Ops assessment fits
This is an informational guide, and nothing on it describes a Clever Ops test of Cloudflare's platform.
The Clever Ops methodology leaves services run by other businesses outside the method, and Cloudflare's policy allows Cloudflare-owned destinations only as part of its Public Bug Bounty program.
The application behind your zone is where a Clever Ops assessment can start. For a custom web application and its API, that is the web application and API penetration test, scoped with two test accounts for every role. Active testing of an application behind Cloudflare sits within that scoped assessment.
Before the first request, each target's owner signs an authorisation and the host's testing policy is checked. During testing the request rate stays capped, and a firewall that starts blocking, or a request from your contact, pauses testing on that target.
For a brochure or marketing site, the website security check looks from outside at what is visible to anyone on the internet without signing in, such as certificates, transport security and security headers.
Testing also needs a list of the exact targets: each domain, subdomain and API address in scope, and anything that has to stay untouched. A completed worksheet already holds your proxied hostnames and whether the origin answers directly. Load and denial-of-service tests are referred to a specialist in load testing, and they need your host's written approval in advance.
If Clever Ops built or hosts the application behind the zone, the tester would be checking work Clever Ops did, so the engagement is a security review rather than an independent penetration test, and its report records who built the application and who hosts it.
Clever Ops does not run incident response or watch systems around the clock. Where a token or the account itself may have been misused, Cloudflare's support and an incident response specialist come before any review.
Frequently asked questions
Cloudflare's security features interoperability page says DDoS protection is always on for all plans and lists Bot Fight Mode as available on Free. Its managed rules page lists the Free Managed Ruleset on every plan, while the Cloudflare Managed Ruleset and OWASP Core Ruleset start at Pro. Its custom rules page lists 5 custom rules on Free, without the Log action, while the interoperability page's scenario for an e-commerce site says a Pro or Business plan adds custom rules. Origin CA certificates, Authenticated Origin Pulls, HSTS and Always Use HTTPS each appear on every plan in their pages' availability tables. Cloudflare's proxy status page adds that it cannot cache and protect requests to a DNS-only record that serves web traffic.
Cloudflare's encryption modes page strongly recommends Full or Full (strict) where possible. Flexible sends traffic from Cloudflare to your origin over plain HTTP, and Cloudflare's page for it says to use Full or Full (strict) when the application holds sensitive information such as personalised data and user login. Full matches the visitor's scheme and does not validate the origin's certificate. Full (strict) validates it, so the origin needs an unexpired certificate from a publicly trusted authority or Cloudflare's Origin CA that matches the hostname. The page for Strict (SSL-Only Origin Pull) lists it for zones on the fourth plan only.
Cloudflare's proxy status page says a DNS query for a proxied record is answered with Cloudflare anycast addresses rather than your origin's. A DNS-only record that points at the same server reveals it, and Cloudflare's exposed IP addresses page says it is still possible, though more difficult, to find an origin behind the proxy. Its IP addresses page gives historical DNS records and mail server configuration as examples of how an origin's address can be discovered. Among the ways Cloudflare's origin protection guide lists to turn away direct requests are allowing only its addresses at the origin, Authenticated Origin Pulls and Cloudflare Tunnel.
Cloudflare's security features interoperability page gives the request phases in order: HTTP DDoS Attack Protection, custom rules, rate limiting rules, Managed Rules, then Super Bot Fight Mode, with account-level rulesets before zone-level ones in each phase. A terminating action such as Block stops the request before later phases. A custom rule with the Skip action can bypass rate limiting, Managed Rules and Super Bot Fight Mode. IP Access Rules, Zone Lockdown, User Agent Blocking, Browser Integrity Check, Hotlink Protection and Security Level are evaluated independently of that order, and Bot Fight Mode also sits outside it.
Cloudflare's scans and penetration testing policy says customers may scan and test application and network-layer aspects of their own assets, such as their zones, with certain restrictions and provided they follow the policy. Testing must stay within customer-owned IPs, Cloudflare's designated public IPs and the customer's registered DNS entries, and Cloudflare-owned destinations are allowed only through its Public Bug Bounty program. Before a penetration test, the policy asks for specific WAF, bot and rate limiting configurations on each zone tested, and for denial-of-service tests it points to a separate page on the required notification. Your origin host's terms may add conditions of their own.
Bot Fight Mode does not run on Cloudflare's Ruleset Engine, so its page says WAF custom rules and Page Rules cannot skip or bypass it. It does not trigger when an IP Access rule matches the request first. For exceptions such as your own API clients or monitoring tools, the same page points to Super Bot Fight Mode, which it says supports Skip rules. Cloudflare's Super Bot Fight Mode page says it is included on Pro, Business and the fourth plan.
Scope the right test for your application
Tell us what you have built, who uses it and when it needs to launch. You will get a plain-English recommendation: a website audit, a web application test or a referral to a specialist.