If you’re compromised: a step‑by‑step cleanup plan

If evidence of compromise exists, follow these steps in order. For high‑value sites consider professional incident response.

1. Contain: Put the site into maintenance mode or take it offline if necessary. Block suspicious IPs at the host or network level.
2. Preserve evidence: Make a full snapshot (files + DB) and preserve logs. Do not overwrite logs until investigation is complete.
3. Rotate credentials: Reset passwords for all admin, editor and contributor accounts. Invalidate active sessions. Rotate API keys and integration tokens if needed.
4. Remove the injected payloads: Search and sanitize or remove stored payloads in postmeta, options, widgets and plugin settings. Edit entries carefully — do not delete system data blindly.
5. Reinstall clean plugin files: Replace plugin files with fresh copies from the official source (after you have patched). Avoid attempting to “clean” modified plugin files unless you know exactly what changed.
6. Re‑scan and validate: Use multiple scanners and file integrity checks. Compare core and plugin files to known good sources.
7. Investigate the entry point: Identify which account performed the injection and how credentials were obtained (weak password, credential reuse, phishing, compromised local workstation).
8. Notify affected parties: If user data was exposed or clients are affected, disclose appropriately and notify stakeholders per local regulations and good practice.
9. Monitor and learn: Continue monitoring logs, re‑scan periodically, and improve role hygiene and onboarding processes to prevent recurrence.

For plugin developers: how this should be fixed properly

Developers must treat untrusted input as hostile. Key practices:

1. Never trust user input: Validate on input and escape on output. Input validation filters clearly invalid content; output escaping ensures safe rendering in each context.
2. Sanitize HTML with wp_kses / wp_kses_post: If limited HTML is allowed, use wp_kses() with an allowlist that excludes on* attributes and javascript: URIs.
3. Escape in the right context: Use esc_attr(), esc_html(), esc_js(), wp_json_encode() as appropriate for attributes, body text and inline JS.
4. Capability checks and nonces: Verify current_user_can() and nonces on all state‑changing endpoints. Use the most specific capability relevant to the action.
5. Sanitize storage and escape output: Apply both — sanitizing at storage reduces risk, but output escaping remains mandatory.
6. Least privilege: Avoid exposing administrative widget fields to low‑privilege users. If HTML input is necessary, consider restricting to trusted roles.

Example conceptual save handler:

 array(
    'href' => true,
    'title' => true,
    'rel' => true,
  ),
  'strong' => array(),
  'em' => array(),
  'br' => array(),
);
$cta = wp_kses( wp_unslash( $_POST['cta_field'] ), $allowed );
update_post_meta( $post_id, 'my_plugin_cta', $cta );
?>

Example output: