amelnagdy/wp-guard
Review generated or changed WordPress code — plugins, themes, and blocks — before it ships. Best used reactively after an agent writes, edits, or reviews code touching WordPress APIs: add_action/add_filter, shortcodes, meta boxes, AJAX handlers, REST routes, WP_Query or $wpdb, widgets, or WP-CLI commands. Use on 'review this plugin', 'is this safe to ship', 'make this translatable', 'speed up this query', or after tasks like 'write a plugin' or 'add an endpoint/shortcode/meta box'. Enforces escaping and sanitization, nonces plus capability checks, prepared database queries, core-API-first development, translation-ready strings, and query/caching discipline. DO NOT USE for WooCommerce-specific order, product, or checkout logic (use woo-guard), non-WordPress PHP, generic code quality review (use clean-code-guard), test code review (use test-guard), server or hosting configuration, or conceptual WordPress questions.
npx skills add https://github.com/amElnagdy/guard-skills --skill wp-guard
You are reviewing generated or changed WordPress code before it ships. Apply the rules below as a guard pass after the first implementation pass. Be a sharp reviewer, not a pedantic one: flag what creates vulnerabilities, breaks translations, or melts servers — ignore cosmetic preferences WPCS tooling already handles.
These rules exist because AI agents produce WordPress code with systematic failures: raw echo of request data, AJAX handlers with neither nonce nor capability check, SQL built by string interpolation, English hardcoded into user-facing strings, posts_per_page => -1 on sites with a million posts, and hand-rolled replacements for APIs core already ships. Each one looks fine in a demo and fails in production.
Guard-pass mode (recommended): after WordPress code has been generated or edited, apply the rules to the diff or target files, then run the self-check before delivery. Fix violations before showing the user.
Live mode (explicit): when the user invokes this skill before writing WordPress code, apply the same rules while writing, then run the self-check before delivery.
Review mode (the user asks you to review, audit, or rate WordPress code): walk references/review-checklist.md against the target files and produce a structured findings report. Do not edit code in review mode unless asked.
Pair this skill with clean-code-guard when both are installed: clean-code-guard owns generic code quality; wp-guard owns the WordPress layer.
phpcs.xml/WPCS config, and composer.json. Project conventions win on conflict.esc_html(), esc_attr(), esc_url(), or wp_kses()/wp_kses_post() for rich content. Data passed to inline JS goes through wp_json_encode() + wp_add_inline_script() — esc_js() is legacy, for single-quoted strings in inline attributes only. Escaping happens at output, not at storage. echo $anything; without an esc_* wrapper fails review.$_POST, $_GET, $_REQUEST, $_SERVER) never touches logic raw: wp_unslash() first, then the type-correct sanitizer (sanitize_text_field(), sanitize_key(), absint(), sanitize_email(), …). Sanitization is not escaping; doing one never excuses the other.current_user_can()) AND a nonce (check_admin_referer(), check_ajax_referer(), or REST nonce handling). A nonce is not authorization. A REST permission_callback of __return_true on a writing route fails review.$wpdb->prepare() for every query containing a variable. Placeholders (%s, %d, %f, and %i for identifiers on WP ≥ 6.2), never interpolation or concatenation. Prefer WP_Query, the meta and options APIs over raw SQL when they can express the query.wp_remote_get()/wp_remote_post(), never curl. Assets via wp_enqueue_script()/wp_enqueue_style(), never echoed <script>/<style> tags. Scheduling via WP-Cron or Action Scheduler. Redirects via wp_safe_redirect() followed by exit. File writes via WP_Filesystem. Simple persistent data via options/transients, not a custom table.add_action(), add_filter(), or calling a core/plugin function, confirm it exists in the supported versions — read the source or the project's installed code. Hallucinated hooks fail silently in WordPress: no error, no behavior. Also match the hook to the moment — front-end code does not load on admin_init, queries do not run before init expects them.get_settings, data, api_key) are collisions waiting for the next active plugin.ABSPATH check (or equivalent project convention).__(), _e(), _x(), _n(), or the escaping combos esc_html__(), esc_attr__()), a literal text domain matching the plugin slug — never a variable or constant — translator comments on every placeholder, _n() for plurals (never sprintf with a hardcoded singular/plural choice), and no sentence assembly by concatenation. Dates and numbers through date_i18n()/wp_date() and number_format_i18n(). Details and JS i18n: references/i18n.md.10. Query discipline. No posts_per_page => -1 and no query_posts(), ever. Use 'fields' => 'ids' when only IDs are needed, 'no_found_rows' => true when not paginating, and never query inside a loop what could be primed once (meta/term caches). Details: references/performance.md.
11. Cache expensive work, load assets where used. Remote calls and heavy computations go behind transients or the object cache with a deliberate TTL. Options that are large or rarely read register with autoload => false. Scripts and styles enqueue only on the screens that use them.
echo, print, <?=: is every variable output escaped with the context-correct function?$_POST, $_GET, $_REQUEST: unslashed? sanitized? nonce-verified? capability-checked?$wpdb->: every variable behind a placeholder?WordPress-Extra + WordPress-Security) without warnings you cannot justify?If any answer is wrong, fix it before showing the user.
**Rule N violation** in `path/file.php:<line or function>`
- What: <one sentence>
- Risk: <XSS / SQLi / CSRF / broken i18n / scaling — one phrase>
- Fix: <one sentence>
Group by file, lead with security findings. If a file is clean, don't mention it.
$wpdb->prepare details, file uploadsTake amelnagdy/wp-guard from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.