Where custom rules load from
Custom rules go in /var/ossec/etc/rules/. Those files are not loaded after the stock ruleset: they are merged with /var/ossec/ruleset/rules/ and loaded in order of the full file name. On 4.14.7 the image has 168 stock rule files, all starting with a digit, the last being 0999-malicious-ioc-rules.xml, so local_rules.xml loads after all of them. We measured the same child rule of stock rule 5715 in five file names: 0094-test.xml and 0095-aaa.xml were ignored; 0095-zzz.xml, 0096-test.xml and local_rules.xml fired. Full measurement.
How to check in one minute
/var/ossec/bin/wazuh-analysisd -t 2>&1 | grep -E 'WARNING|ERROR|CRITICAL'
The exit code is 0 even when rules are ignored, so read the lines. Each will be ignored line names a rule that is not running.
Six ways, measured
1. The parent rule is in a file that loads later
A child rule in a file that sorts before its if_sid parent's file is read before the parent exists. Wazuh logs 7617 and 7619, ignores the rule, and the config check still exits 0:
WARNING: (7617): Signature ID '5715' was not found and will be ignored in the 'if_sid' option of rule '100080'.
WARNING: (7619): Empty 'if_sid' value. Rule '100080' will be ignored.
Rule dropped at load: 7617 / 7619
2. A static field written as <field>
action is a static field with its own <action> tag. Writing <field name="action"> instead logged ERROR: Failure to read rule 100210. Field 'action' is static. and the manager did not start: no alerts at all, not only for FortiGate.
FortiGate note, with the measurement
3. More than 50 load warnings
wazuh-analysisd -t and ossec.log keep at most 50 load messages; anything beyond that is dropped, oldest first, with no line anywhere. In our test 60 rules were ignored: 50 lines were printed, naming the last 25, and the first 35 were never mentioned. A public community ruleset printed 52 warnings where a static check found 55.
The 50-warning limit, measured
4. An if_group the load check cannot match
At load time, 4.14.7 accepted an upper-case, partial group name and a | list, and rejected a name starting with ^ that does not begin any group:
<rule id="100715" level="5"><if_group>AUTHENTICATION_FAI</if_group>...</rule>
<rule id="100716" level="5"><if_group>zzz_nope|authentication_success</if_group>...</rule>
<rule id="100717" level="5"><if_group>^uthentication</if_group>...</rule>
WARNING: (7610): Group '^uthentication' was not found. Invalid 'if_group'. Rule '100717' will be ignored.
Rules 100715 and 100716 loaded with no warning. This is the load check only; we did not test which events these rules then match.
5. JSON logs behind a syslog header
A line that is only JSON decodes on its own. The same JSON behind a standard syslog header matched no decoder, and behind a plain date it was claimed by the stock windows-date-format decoder with no fields, so rules on the JSON keys never see them.
6. A CDB list rule under load
A rule with lookup="not_match_key" fed only values that were in the list gave no false alerts at 1,000 events per second, 10 at 5,000, and 24 and 27 in two 50,000-event bursts; with a single rule-matching thread, or a negated <field> in place of the list, the same bursts gave none.
CDB list not_match_key false positives
7. Events the stock ruleset never alerts on
On 4.14.9, usermod adding a user to sudo has no decoder at all, MySQL Access denied over syslog lands on level 0 and MariaDB's on level 2, while the same group change through gpasswd alerts at level 5. A local decoder and four rules close the gap; one field set still needs a stock decoder change.
usermod to sudo and MySQL access denied: no alert
The same holds for PHP: a PHP-FPM worker killed by SIGSEGV is not decoded, and repeated out-of-memory fatals from one client stay at level 0 or 5, the two signals to watch for PHP2Shell (CVE-2026-109618/109619).
PHP2Shell: PHP-FPM crashes and out-of-memory floods
8. A stock decoder that stores the wrong IP
On 4.14.x the snort decoder drops the last octet of srcip when the source has no port and keeps the port in dstip. The fix merged for 4.14.11 still cuts some dstip values; a local decoder with one more change got all seven syslog samples right.
snort srcip and dstip truncated
Limits of what we measured
Each item measured on 4.14.7; details on the linked page. Item 4 was measured for this page with wazuh-analysisd -t only. Other versions, clusters and the Wazuh 5.0 engine (which does not load XML rules) were not covered here.