Who did what, to which honeypot, and whether it actually worked — superadmin-only, and it never forgets on purpose.

Every route or background job that mutates something — or that refuses to because a safeguard tripped (a typed confirmation that didn't match, a missing pinned host key, a bad self-registration token, a failed or locked-out login) — writes one AuditLogEntry: actor, action, outcome, source IP, and a plain-language summary. A background job (the scheduler, a retention purge) logs under a fixed actor label instead of a user, since it has no request to read one from.

In the reader's language. The summary is always recorded in English — exports, syslog, the Wazuh rules and the REST API stay English, so a machine-consumed record never varies by who triggered it. On the Audit log page an English reader sees that summary; any other language shows the translated label of the action code instead (audit.action_label.<code> in each locale file, falling back to the parent code — honeypot.power.reboot → honeypot.power — and then to the English summary), with the full English summary as the cell's tooltip. The details column is shown as key: value · key: value. A test fails when a new audit action has no label in every shipped language (tests/test_audit_labels_and_nav.py).

The Audit log page itself live-refreshes over WebSocket the moment a new entry is written — no need to reload to see something that just happened. See Honeypot Management's "Live updates over WebSocket" for how that mechanism works across the app.

Every entry is hash-chained: entry_hash covers this entry's own fields plus the previous entry's entry_hash, so altering or deleting an entry breaks the chain from that point on. Concurrent writers (the web app and every Celery worker child can all log at once) serialize through a single locked row for the duration of one entry's write, so two concurrent entries can never both link to the same previous hash.

Settings → Security → "Verify chain" walks the whole table and reports the first broken link, if any — a proactive integrity check, not something that only fires when someone goes looking.

Logging commits independently of whatever the caller is doing, after its own commit — a failed audit write can never roll back the action it describes, and a rejected/blocked action still gets its own record even though nothing else was worth committing.

Kept audit_log_retention_days (Settings → Checks & retention, default 90) before the daily purge job deletes old entries — None means keep forever. Exportable as CSV or JSON (/audit/export) for archival or compliance, with the same spreadsheet-formula-injection guard the REST API's own event export uses (a leading =/+/-/@ gets a defusing prefix, since a raw one opens as a live formula in Excel/Sheets).

A separate, optional target (Settings → Integrations, app.audit_syslog) mirrors every entry to a syslog server over UDP/TCP/ TCP-over-TLS as it's written — best-effort, a delivery failure is logged and swallowed, never blocking the write it's mirroring. This is the global, audit-only target — never a honeypot alert; those have their own three targets entirely, see Honeypot Management.

Wazuh users: wazuh/honeypotshelf_decoders.xml + wazuh/honeypotshelf_rules.xml (repo root) ship a ready-made decoder + rule set (id range 107000-107099) for both this feed and the honeypot-alert one — covers every audit action this app emits and every OpenCanary module, plus brute-force correlation rules for repeated failed logins. Both files are plain XML with no comments (some Wazuh manager versions' own upload/file-editor path chokes on comments and non-ASCII characters), so the install steps live here instead:

1. Copy honeypotshelf_decoders.xml to /var/ossec/etc/decoders/ and honeypotshelf_rules.xml to /var/ossec/etc/rules/ — or point a <decoder_dir>/<rule_dir> pair in ossec.conf's <ruleset> block at wherever you keep this wazuh folder instead:

   <ruleset>
     <decoder_dir>etc/decoders</decoder_dir>
     <rule_dir>etc/rules</rule_dir>
     ...
     <decoder_dir>/path/to/wazuh</decoder_dir>
     <rule_dir>/path/to/wazuh</rule_dir>
   </ruleset>

2. Point Honeypot Shelf's syslog target(s) at this manager (or an intermediate syslog-ng/rsyslog relay forwarding to it) — Settings → Integrations for the global audit target, each Company's own page and/or "All honeypots" → Integrations for alerts.
3. /var/ossec/bin/wazuh-control restart (or just restart wazuh-manager) to load both files.
4. Verify with /var/ossec/bin/wazuh-logtest against a captured line before relying on it in production, as always.

The rules file matches on the <decoded_as> names the decoders produce (honeypotshelf-audit / honeypotshelf-honeypot-alert — each decoder needs <use_own_name>true</use_own_name>, or Wazuh reports the parent decoder's name instead and <decoded_as> never matches) and on the fields both decoders expose from each feed's JSON body via JSON_Decoder — every top-level key in the payload, referenced by its own name with no data. prefix (JSON_Decoder doesn't add one): event, id, timestamp, action, actor, ip, outcome, target_type, target_id, target_label, summary, details for the audit feed; event, id, timestamp, companies, honeypot, honeypot_ip, type — OpenCanary's own numeric logtype id, as a string — label, src_ip, src_port, dst_port, source, raw for the honeypot-alert feed).

Watch out for Wazuh's 13 reserved "static" field names (srcip, dstip, srcport, dstport, protocol, user, status, id, url, data, extra_data, action, hostname) — the audit feed's own action key collides with one. A rule can't match it via <field name="action">...</field> (Wazuh rejects the whole rule file at upload with a bare "XML syntax error"/code 1113, no line number); it needs the dedicated <action type="pcre2">...</action> element instead — same attributes, different tag. honeypotshelf_rules.xml already does this.

Severity levels: every honeypot-alert rule (107001, 107045- 107069) sits at level 8 or higher — a honeypot has no legitimate traffic, so any interaction with one is inherently an attacker doing something, never routine noise. Credential submissions/login attempts step up to 9, and the brute-force correlation rules (107007, 107050) stay at 10. The audit-feed rules (107000-107043) stay mostly informational (level 3-6) and only escalate to 8+ for genuinely security-relevant events (failed/locked-out logins, CSRF rejection, rate limiting, destructive actions) — ordinary CRUD by a logged-in admin is expected, routine activity, not something worth an analyst's attention.