ATK Rule Doctor Lite: why do my custom Wazuh rules never fire?

A free, read-only check for Wazuh 4.x. It lists every custom rule that did not fire and tells you the most likely reason, including rules that Wazuh silently threw away while loading them.

One Python 3 file, standard library only. It reads files; it changes nothing on your manager and sends nothing anywhere.

curl -sLO https://github.com/xuxu298/rule-doctor-lite/releases/latest/download/rule-doctor-lite.py
sudo python3 rule-doctor-lite.py                 # on the manager, reads /var/ossec
python3 rule-doctor-lite.py --ossec-dir ./copy   # or on a copy of /var/ossec

Or from PyPI (same file, with a rule-doctor-lite command):

pipx install rule-doctor-lite                    # or pip install rule-doctor-lite inside a venv
sudo "$(command -v rule-doctor-lite)"             # on the manager, reads /var/ossec

Manager in docker? Copy the four things it reads, then point --ossec-dir at the copy:

C=single-node-wazuh.manager-1; mkdir -p copy/ruleset copy/etc copy/logs/alerts
docker cp $C:/var/ossec/ruleset/rules copy/ruleset/ && docker cp $C:/var/ossec/etc/rules copy/etc/
docker cp $C:/var/ossec/logs/ossec.log copy/logs/ && docker cp $C:/var/ossec/logs/alerts/alerts.json copy/logs/alerts/

What the labels mean

label	meaning
DROPPED-AT-LOAD	Wazuh ignored the rule at load time: its if_sid parent was not loaded yet (the file name sorts before the parent's file, or the parent is defined later in the same file), points to the rule itself, or does not exist. The manager still starts and wazuh-analysisd -t still exits 0; the only trace is warnings 7617/7619, printed by -t and written to ossec.log, that nothing fails on. Read from ossec.log when present, otherwise predicted from the load order. Chains are followed: a rule whose parent was itself dropped is dropped too, however deep.
SHADOW-CANDIDATE	a sibling that Wazuh evaluates first (higher level, or same level and loaded earlier) did fire. A candidate, not a finding: it proves the sibling matched something, not that your rule would have matched the same event.
NEVER-REACHES-MANAGER	no related rule fired and the manager logged event loss (rules 203/204). Only loss that raised 203/204 is seen: an agent with its client buffer off, or a manager dropping events at its EPS limit, raises neither, and Lite then says NO-MATCH.
NO-MATCH	a related rule fired but this one did not, or nothing suggests event loss.

The shadowed / never-reaches-the-manager / no-match split is Kislley Rodrigues's.

LOCATION-ANCHORED (0.2.3) is a separate list, not a silence label: custom rules whose <location> is anchored at the start of a path, like ^/var/log/auth.log. An agent that reads host logs through a mount, for example /hostfs in a containerised agent, reports location as /hostfs/var/log/auth.log. Measured on Wazuh 4.14.7, first with wazuh-logtest and then with a live wazuh/wazuh-agent:4.14.7 container reading /hostfs/var/log/auth.log: <location>/var/log/auth.log</location> still fires (it is a substring match), <location>^/var/log/auth.log</location> does not. The idea comes from Mattias Hemmingsson's wazuh-help notes on running agents in containers.

Every result is printed with the time window it was measured over. A rule written for a quarterly event is not dead after a day.

What it reads by default (0.2.1): the current logs/alerts/alerts.json plus the rotated alert files of the last 7 days (logs/alerts/<year>/<Mon>/ossec-alerts-<dd>.json, plain or .gz, as Wazuh writes them); change the window with --days N. From ossec.log it takes the 7617/7619 warnings of the latest rule load only (the block before the last Total rules enabled), so a rule you already fixed is not reported from an older load, and lines from wazuh-analysisd -t are ignored.

Tested on a real manager

wazuh/wazuh-manager:4.14.7, 27/09/2026. Two identical custom rules on if_sid 5715, one in 0094-test.xml (sorts before the stock 0095-sshd_rules.xml), one in 0500-ok.xml. A real sshd "Accepted password" event fired 100081 from 0500-ok.xml; 100080 in 0094-test.xml never fired. Rule Doctor Lite reported:

100080   level 10  0094-test.xml   DROPPED-AT-LOAD   ossec.log 7617: parent 5715 not found when the rule loaded

and, with ossec.log withheld, the same verdict from the load order alone:

100080   level 10  0094-test.xml   DROPPED-AT-LOAD   predicted: parent 5715 is in 0095-sshd_rules.xml, which loads after 0094-test.xml

Chains (0.2.0, 28/09/2026)

Same image. Anchor 910010 on if_sid 5715 in 0094-early.xml, child 910012 on if_sid 910010 and grandchild 910013 on if_sid 910012 in 0500-chain.xml. wazuh-analysisd -t exited 0; ossec.log had 7617 + 7619 for all three, each naming the rule above it as the missing parent. Lite, with ossec.log withheld:

910010   level 10  0094-early.xml   DROPPED-AT-LOAD   predicted: parent 5715 is in 0095-sshd_rules.xml, which loads after 0094-early.xml
910012   level 10  0500-chain.xml   DROPPED-AT-LOAD   predicted: parent 910010 is itself dropped at load
910013   level 12  0500-chain.xml   DROPPED-AT-LOAD   predicted: parent 910012 is itself dropped at load

The prediction matters because ossec.log can miss warnings: analysisd buffers the warnings of each rules file in a list capped at 50 (ERRORLIST_MAXSIZE) and flushes it after the file, so a file that produces more than 50 warnings loses the oldest ones (src/analysisd/analysisd.c L708-750 on v4.14.7).

tests/run.sh runs the fixture checks without docker.

Limits

• Wazuh 4.x rule syntax. Only if_sid parents are read; if_group, if_matched_sid and if_matched_group are not followed yet.
• NEVER-REACHES-MANAGER only sees loss that raised rule 203 or 204 (see the table above).
• SHADOW-CANDIDATE needs a replay to confirm or clear. Lite does not replay.
• "Never fired" means never fired in the alerts it read (7 days by default). Older alerts need --days or --alerts.

Run checks from your AI agent

The same team runs a free MCP server at https://atkvn.com/agent/mcp (no sign-up; listed in the MCP Registry as com.atkvn/security-checks). Add it to Claude, Cursor or any MCP client and your agent can:

• test_wazuh_rule: run your rules and decoders against your log lines on a real wazuh-manager 4.14.9 and see which rule fired, or why yours did not.
• audit_wazuh_rules: find rules that pass wazuh-analysisd -t but are dropped at load (file order, missing parents, colliding ids).
• check_php2shell: patched or not for CVE-2026-109618/109619, from php -v, php -i and package output it collects on your server. Detection rules for the crash and out-of-memory signals: PHP2Shell note.

Details and limits: https://atkvn.com/agent/

Guides

Step-by-step notes for what Lite points at, each with a one-minute check you can run yourself (measured on Wazuh 4.14.7):

• Wazuh rule dropped at load: 7617 and 7619, while wazuh-analysisd -t exits 0
• Wazuh rule matches in wazuh-logtest but never fires in production
• Wazuh rule shadowed by a sibling rule
• mapper_parsing_exception: the alert is in alerts.json but never reaches the dashboard
• Wazuh "Cannot read 'srcip' from data"
• Wazuh rule with <hostname> matches in logtest but never fires for agent events
• PHP2Shell: stock Wazuh does not alert on PHP-FPM crashes or out-of-memory floods (4.14.9)

Full version and done-for-you fixes

• Free 24-hour answer: email the --json report, your custom rule files and your Wazuh version to [email protected], and within 24 hours you get a note on every silent rule and why it does not fire.
• ATK Rule Doctor (USD 490 per cluster per year) replays the real events behind every SHADOW-CANDIDATE through a throwaway manager of your version to confirm or clear it, and fixes alerts the indexer rejects with mapper_parsing_exception: https://vct.atkvn.com/rule-doctor.html
• Wazuh Fix: we fix a silent rule or a mapping conflict on your version, USD 149 per problem (USD 490 for indexer or deployment work), and you pay only after the fix fires on your cluster: https://vct.atkvn.com/rule-doctor.html?src=gh

Licence: Apache-2.0. Made by ATK New Technology, Hanoi.