Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 12 additions & 4 deletions docs/hook-order-compatibility.md
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,10 @@
Focused check on 2026-11-13: WebDecoy 2.11.4 at `92a92da`, PHP 8.5.10,
WordPress 7.1.4 hook dispatcher. Fourteen request cases passed.

A follow-up [installed-plugin HTTP matrix](plugin-stack-compatibility-2026-11-13.md)
now adds 122 measured cases using real WordPress and MariaDB. This page describes
the earlier, narrower hook fixture.

The general front-end check is registered on `init` at priority 1. A callback
that exits before it runs prevents detection of that request. Later redirects
do not erase an already recorded detection.
Expand All @@ -28,7 +32,8 @@ nonce, authentication, or enforcement behavior.

## Current companion sources reviewed

These are source reviews, **not installed-plugin compatibility certifications**:
These were source reviews during the initial pass. The linked installed-plugin
follow-up records the runtime cases now exercised and the remaining limits:

- Redirection 5.10.1: `includes/module/class-plugin.php` registers canonical
redirects and initialization at default `init` priority 10, with page
Expand All @@ -37,11 +42,14 @@ These are source reviews, **not installed-plugin compatibility certifications**:
default `init` priority 10 and its main login handling on `wp_loaded`.
Its earlier `plugins_loaded` handler rewrites login request paths and
`$pagenow`; consequently WebDecoy may inspect the rewritten path rather than
the original login URL. Complete mixed-plugin behavior remains untested.
the original login URL. The follow-up exercises the main login and redirect
flows, but not every setting.
- Wordfence 9.1.1 has earlier bootstrap protection and later hooks. Its optimized
firewall runs before WordPress. Requests it ends before WebDecoy runs are
outside this plugin's observation window; no complete Wordfence coexistence
or Extended Protection installation was tested.
outside this plugin's observation window; the initial hook fixture did not
exercise that runtime. The follow-up now
tests a manual Wordfence block in basic and native prepend modes; licensed
vendor signatures remain outside its scope.

Official packages reviewed:
[Redirection](https://downloads.wordpress.org/plugin/redirection.5.10.1.zip),
Expand Down
218 changes: 218 additions & 0 deletions docs/plugin-stack-compatibility-2026-11-13.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,218 @@
{
"tested_at": "2026-11-13",
"runtime": {
"wordpress": "7.1.4",
"php": "8.3.33",
"mariadb": "11.8.9",
"webdecoy": "2.11.4",
"wordpress_image_id": "sha256:8801a1239d7ba9fb340a5fc5ba0bf7f8d3652adbd64893e3fba7992ba618108e",
"mariadb_image_id": "sha256:2439dcd7d14010ecd1ff7a4e1c5abe8e208c34fe35290744deeeaac3569043c3"
},
"packages": [
{
"slug": "webdecoy",
"version": "2.11.4",
"url": "https://downloads.wordpress.org/plugin/webdecoy.2.11.4.zip",
"sha256": "6fc5d8a1e6c520f98ca67e4ab9d0274cdc1d7c06967c4ad6fbd5f0ee3a912e11"
},
{
"slug": "redirection",
"version": "5.10.1",
"url": "https://downloads.wordpress.org/plugin/redirection.5.10.1.zip",
"sha256": "f68a4be447ecba00c64e8a39d7517807ed49cc18e77b7a364c20cb31848235b7"
},
{
"slug": "wps-hide-login",
"version": "1.9.20",
"url": "https://downloads.wordpress.org/plugin/wps-hide-login.1.9.20.zip",
"sha256": "86fc072944700163a221c11ea4d132bd907e04229d38c3e98030e47a46ba6b72"
},
{
"slug": "wordfence",
"version": "9.1.1",
"url": "https://downloads.wordpress.org/plugin/wordfence.9.1.1.zip",
"sha256": "51acb206ebaef1756e03112df81c749a70e4086d1a71b6e3426373e393880ad1"
}
],
"scenarios": [
{
"name": "baseline",
"mode": "Wordfence default/learning where active",
"plugins": [
"webdecoy/webdecoy.php"
],
"http_cases": 10,
"passed": true,
"wordfence": null
},
{
"name": "redirection",
"mode": "Wordfence default/learning where active",
"plugins": [
"redirection/redirection.php",
"webdecoy/webdecoy.php"
],
"http_cases": 11,
"passed": true,
"wordfence": null
},
{
"name": "hide",
"mode": "Wordfence default/learning where active",
"plugins": [
"webdecoy/webdecoy.php",
"wps-hide-login/wps-hide-login.php"
],
"http_cases": 11,
"passed": true,
"wordfence": null
},
{
"name": "wordfence",
"mode": "Wordfence default/learning where active",
"plugins": [
"webdecoy/webdecoy.php",
"wordfence/wordfence.php"
],
"http_cases": 10,
"passed": true,
"wordfence": {
"api_key_configured": false,
"is_paid": false,
"waf_status": "learning-mode",
"waf_rules_file_present": true,
"basic_protection": true
}
},
{
"name": "all",
"mode": "Wordfence default/learning where active",
"plugins": [
"redirection/redirection.php",
"webdecoy/webdecoy.php",
"wordfence/wordfence.php",
"wps-hide-login/wps-hide-login.php"
],
"http_cases": 12,
"passed": true,
"wordfence": {
"api_key_configured": false,
"is_paid": false,
"waf_status": "learning-mode",
"waf_rules_file_present": true,
"basic_protection": true
}
},
{
"name": "all-reverse",
"mode": "Wordfence default/learning where active",
"plugins": [
"redirection/redirection.php",
"webdecoy/webdecoy.php",
"wordfence/wordfence.php",
"wps-hide-login/wps-hide-login.php"
],
"http_cases": 12,
"passed": true,
"wordfence": {
"api_key_configured": false,
"is_paid": false,
"waf_status": "learning-mode",
"waf_rules_file_present": true,
"basic_protection": true
}
},
{
"name": "all-waf-enabled",
"mode": "Wordfence enabled, native plugin bootstrap",
"plugins": [
"redirection/redirection.php",
"webdecoy/webdecoy.php",
"wordfence/wordfence.php",
"wps-hide-login/wps-hide-login.php"
],
"http_cases": 14,
"passed": true,
"wordfence": {
"api_key_configured": false,
"is_paid": false,
"waf_status": "learning-mode",
"waf_rules_file_present": true,
"basic_protection": true
}
},
{
"name": "all-waf-enabled-reverse",
"mode": "Wordfence enabled, native plugin bootstrap",
"plugins": [
"redirection/redirection.php",
"webdecoy/webdecoy.php",
"wordfence/wordfence.php",
"wps-hide-login/wps-hide-login.php"
],
"http_cases": 14,
"passed": true,
"wordfence": {
"api_key_configured": false,
"is_paid": false,
"waf_status": "learning-mode",
"waf_rules_file_present": true,
"basic_protection": true
}
},
{
"name": "all-extended",
"mode": "Wordfence enabled, native auto_prepend_file",
"plugins": [
"redirection/redirection.php",
"webdecoy/webdecoy.php",
"wordfence/wordfence.php",
"wps-hide-login/wps-hide-login.php"
],
"http_cases": 14,
"passed": true,
"wordfence": {
"api_key_configured": false,
"is_paid": false,
"waf_status": "learning-mode",
"waf_rules_file_present": true,
"basic_protection": true
}
},
{
"name": "all-extended-reverse",
"mode": "Wordfence enabled, native auto_prepend_file",
"plugins": [
"redirection/redirection.php",
"webdecoy/webdecoy.php",
"wordfence/wordfence.php",
"wps-hide-login/wps-hide-login.php"
],
"http_cases": 14,
"passed": true,
"wordfence": {
"api_key_configured": false,
"is_paid": false,
"waf_status": "learning-mode",
"waf_rules_file_present": true,
"basic_protection": true
}
}
],
"http_cases_passed": 122,
"independent_server_receipts_matched": 122,
"local_detection_rows_in_these_cases": 54,
"limitations": [
"Synthetic HTTP clients; no browser JavaScript/client scanner exercise.",
"No Wordfence license or downloaded vendor signatures; local manual UA rule tested in both basic and auto_prepend modes.",
"No cloud reporting, real OpenAI identity/DNS verification, Apache/FPM, CDN or full-page-cache certification."
],
"findings": [
"No conflict observed in the tested login, navigation, redirect, AI counter and WebDecoy trap enforcement flows.",
"Wordfence manual block returns503 before WebDecoy detection; auto_prepend cases terminate before WordPress loads.",
"Repeated already-blocked-IP requests return403 without a fresh detection row.",
"WPS Hide Login rewrites the hidden login path before WebDecoy sees its request context."
],
"production_changes": false,
"wordfence_downloaded_rule_feed_bytes": 0
}
114 changes: 114 additions & 0 deletions docs/plugin-stack-compatibility-2026-11-13.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,114 @@
# Installed WordPress plugin coexistence check — 2026-11-13

**122 synthetic HTTP cases passed across 10 scenarios**, matched one-to-one
with independent server-side access receipts. No conflict was observed in the
tested navigation, login, redirect, AI-counter or WebDecoy blocking flows.
No production detection-hook change is supported by this check.

This follows the [14-case hook-order check](hook-order-compatibility.md).
These tests use actual installed plugins, full WordPress bootstrap and a real
MariaDB database. They do not use the earlier in-memory service adapters.
The [measurement summary](plugin-stack-compatibility-2026-11-13.json) records
package hashes, runtime identities and per-scenario results. All traffic is
controlled fixture traffic, never customer traffic or detection-accuracy data.

## Versions and scenarios

WordPress7.1.4, PHP8.3.33, MariaDB11.8.9, public WebDecoy2.11.4,
Redirection5.10.1, WPS Hide Login1.9.20 and Wordfence9.1.1. Official WordPress.org
packages were staged before execution; no plugin code was patched.

| Scenario | HTTP cases | Result |
| --- | ---: | --- |
| WebDecoy alone | 10 | Passed |
| WebDecoy + Redirection | 11 | Passed |
| WebDecoy + WPS Hide Login | 11 | Passed |
| WebDecoy + Wordfence, default learning mode | 10 | Passed |
| All four plugins | 12 | Passed |
| All four, alternate load order | 12 | Passed |
| All four, Wordfence enabled with local manual rule | 14 | Passed |
| Same, alternate load order | 14 | Passed |
| All four, native Wordfence `auto_prepend_file` | 14 | Passed |
| Same, alternate load order | 14 | Passed |

The first six scenarios total 66; enforcement-enabled scenarios add 28 and native
prepend scenarios add 28. A later re-run checks the saved reproduction runner;
it does not add new coverage categories to the primary 122-case measurement.

## What was exercised

- A browser-style HTTP client retrieves the target page successfully. This is
an HTML request check, not a browser JavaScript test.
- Ordinary `python-requests` traffic creates real local detection rows from
WebDecoy scoring, independently of the reserved test-trigger control.
- A separate reserved Test UA returns its labeled403 JSON receipt.
- One synthetic GPTBot-UA request per scenario increments the real hourly AI
counter by exactly one. Its identity stays **claimed**, never verified.
- Correct credentials submitted to the real login form produce a login redirect
and an authenticated200 admin page, including WPS Hide Login's custom URL.
- An unauthenticated client-detection AJAX call without a nonce is rejected403
through the dedicated handler and creates no general front-end detection.
- A local tripwire in monitor mode records its violation and proceeds to404;
block mode returns403 and records the fresh violation.
- Redirection's actual installer/model API creates a301 rule; the redirected
bot request still has a WebDecoy detection row.
- WPS Hide Login hides the original login URL with404 while WebDecoy still
records the scored bot request.
- Wordfence's local manual UA rule returns503 and prevents target content from
being served. In native prepend mode the receipt proves WordPress has not
loaded and there is no WebDecoy detection row.
- Every extended-mode request independently proves that native
`WFWAF_AUTO_PREPEND` ran; merely configuring an option is insufficient.

## Observation boundaries confirmed

A Wordfence block before WebDecoy runs appears in the server access log and has
no WebDecoy detection row. That is expected earlier termination, not evidence
that a protected action ran. The native prepend block happens before WordPress
itself loads.

A second important distinction is inside WebDecoy: **an already-blocked IP can
receive403 without being scored again or creating a new detection row**. The
initial enforcement-enabled pass exposed state carried from a preceding case.
The fixture now resets only its dummy IP blocks between scenarios and separately
asserts the repeated-block behavior. Detection-row totals are not a total of all
blocked HTTP requests. Admin/AJAX exclusions and earlier edge/cache responses
also prevent one-to-one equivalence with the host's access log.

WPS Hide Login rewrites the original hidden login URI during bootstrap. A
subsequent detector can therefore see the rewritten path. Successful login,
original-login hiding, bot scoring and AI counting still passed in this setup.

These results support keeping the detector at its present lifecycle hook.
Moving it earlier would introduce a separate dependency/bootstrap risk and
would not make a WordPress plugin observe requests stopped before WordPress.

## Isolation, evidence and limitations

The two disposable containers share only a private loopback namespace. There
are no published host ports, external network routes, production credentials,
cloud keys or mail delivery. Code and toolchains are read-only. Database,
Wordfence state, logs and uploads use bounded tmpfs. Each container has one CPU,
384MB RAM limits and PID/file-size limits; PHP requests and HTTP clients have
execution deadlines. Only a dummy local administrator is created.

The enforcement cases use a test-only trusted-loopback proxy header carrying
`192.0.2.42`, avoiding Wordfence's loopback allowlist. This is not a real
reverse-proxy deployment certification. The extended listener uses native PHP
index fallback, with no router that might skip prepend execution.

Wordfence is unlicensed and offline. Its downloaded rules file is empty. A
local manual block is exercised in enabled and native prepend modes; vendor
signature efficacy, feed updates, scanning and cloud services are not tested.
There were startup missing-rule warnings. An initial fixture-only `WP_CLI`
bootstrap mistake was corrected before measured runs. The verified HTTP cases
show no PHP critical-error page; the full setup log is not claimed to be empty.

Other untested surfaces include the client-side scanner (disabled to isolate
server-side behavior), browser JavaScript, Apache/FPM and `.htaccess`, multisite,
actual CDN/full-page-cache behavior, real crawler identity verification, cloud
reporting, and settings combinations outside this matrix.

[Reproduction fixture and isolated setup](../tests/wp-compat/plugin-stack/README.md).
The fixture contains first-party test code only; WordPress and companion
packages are supplied separately. No production plugin source was changed.
8 changes: 8 additions & 0 deletions tests/wp-compat/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -48,3 +48,11 @@ python3 tests/wp-compat/hook-order.py /path/to/wordpress/wp-includes php

This supplements the Docker smoke tests; it is not a full plugin-stack test.
See [scope and measured results](../../docs/hook-order-compatibility.md).

## Installed companion-plugin matrix

The [isolated plugin-stack fixture](plugin-stack/README.md) exercises full
WordPress HTTP requests, real MariaDB storage, Redirection, WPS Hide Login and
Wordfence, including native `auto_prepend_file`. See the
[122-case measured results](../../docs/plugin-stack-compatibility-2026-11-13.md)
for the tested versions, observation limits and remaining scope.
Loading
Loading