Modular Agent Skills that teach AI coding agents — Claude Code, Cursor, Codex, OpenCode, Gemini CLI — to write secure-by-default WordPress code and to audit and harden existing plugins and themes.
You do not need to understand skills, frontmatter, or per-agent install paths. Open your AI coding agent in the WordPress project you want to work on, then copy and paste this prompt:
Set up the WordPress security skills from
https://github.com/wpultimatesecurity/WordPress-Security-Skills for this project.
Read the repository's README and inspect my agent's existing configuration.
Use the install location supported by this agent and its current version;
prefer project-scoped installation unless I ask for a global installation.
Copy the skill directories together with their references. Preserve existing
skills and settings; ask before replacing anything with the same name.
Install instructions only. Do not activate the PHP examples, change my
WordPress site, access production credentials, commit, or push anything.
Report the source revision, installed skill names, destination, and any
reload needed. Verify discovery using the agent's available skill listing;
if you cannot verify it, say so rather than assuming installation worked.
ℹ️ What you are installing — Markdown instructions with PHP and configuration reference examples. This is not a WordPress plugin, a scanner, or automatic protection for your website. Review examples before adapting them to your project.
💡 Every future project — Add: “Install these globally for all my projects.” Use one scope where possible to avoid duplicate or conflicting skill copies.
SKILL.md and
its linked references before working. If it cannot open links, provide the local
files. Pasting a URL alone does not prove the instructions were loaded.
❓ Verify it worked — Ask: “Which WordPress security skills can you discover, and where were they loaded from?” Restart or reload the tool if required.
Prefer doing it yourself? Jump to manual installation.
Browse the skills · Get better results · Report a security issue · Contribute
Scope is deliberately security only. These skills guide development and review; they do not automatically enforce controls or prove that a website is secure.
AI-generated WordPress code can omit nonce checks, authorization, escaping, or prepared queries. These skills describe those failure modes and show defensive patterns for reviewers and developers to adapt.
Guidance links to the official WordPress reference. Check API behavior against your supported WordPress/PHP versions. Automated validation checks structure, PHP syntax, and coding standards where available; it is not an independent security audit or proof that every example is correct.
WordPress plugin and theme developers who use AI coding agents and want the generated code to be secure without hand-holding — plus reviewers auditing AI-written or third-party code.
Each skill targets a documented failure mode in AI-generated WordPress code:
Every skill follows the same structure: When to use · Core principles · Step-by-step ·
Common AI mistakes (wrong→right) · Correct code examples · Checklist · Official references,
with integration examples under each skill's references/ directory. A Supporting
references table in Step-by-step explicitly routes every artifact by load condition. These are
not a plugin bundle: adapt prefixes, permissions, storage, and any documented
asset/template dependencies before running them in a local WordPress environment.
Examples generally use a PHP 7.4 syntax baseline, with newer WordPress/PHP requirements noted where relevant. PHP 7.4 is end-of-life: use a supported PHP release and maintained WordPress version for production. Syntax compatibility does not imply that an older runtime is secure or supported.
These skills conform to the open Agent Skills specification,
so any compatible agent can load them. Each is a directory with a SKILL.md
(name + description frontmatter) plus a references/ folder for progressive disclosure.
Skills are plain directories — install by copying the ones you want (or the whole skills/
folder) into your agent's skills directory.
With Node.js installed, the skills CLI can install directly from this repository:
# Interactive (pick skills and scope):
npx skills add wpultimatesecurity/WordPress-Security-Skills
# Non-interactive, all skills:
npx --yes skills add wpultimatesecurity/WordPress-Security-Skills# Personal (all your projects):
mkdir -p ~/.claude/skills && cp -r skills/* ~/.claude/skills/
# Project-scoped (commit with the repo):
mkdir -p .claude/skills && cp -r skills/* .claude/skills/Claude Code reads skills from ~/.claude/skills/<name>/SKILL.md (personal) and
.claude/skills/<name>/SKILL.md (project). It loads each skill's description at startup and
the body on demand. See the Claude Code skills docs.
# Project-scoped:
mkdir -p .cursor/skills && cp -r skills/* .cursor/skills/Cursor discovers skills in .cursor/skills/<name>/SKILL.md (also .agents/skills/) anywhere
in the repo. See the Cursor skills docs.
# Project-scoped:
mkdir -p .opencode/skills && cp -r skills/* .opencode/skills/
# Global:
mkdir -p ~/.config/opencode/skills && cp -r skills/* ~/.config/opencode/skills/OpenCode loads from .opencode/skills/ (project) and ~/.config/opencode/skills/ (global),
and also reads ~/.claude/skills/ and .claude/skills/. See the
OpenCode skills docs.
Any agent implementing the Agent Skills standard can point at this skills/ directory. Where
an agent reads .claude/skills/ or .agents/skills/ (several do), the Claude Code paths above
work as-is.
- Give the agent context. State your WordPress and PHP versions, whether this is a plugin or theme, multisite/WooCommerce usage, relevant user roles, and whether it is working locally or on staging. Never paste production secrets.
- Load the relevant guidance. Start new code with secure-plugin-development. For existing code, start with security-auditing-code-review and load focused skills as the review identifies trust boundaries.
- Ask for evidence, not a security score. Require file locations, affected roles, exploit prerequisites, and a concrete verification for each finding. Separate confirmed vulnerabilities from hardening recommendations.
- Keep changes reversible. Review the diff, verify the changed behavior on local/staging data, and maintain a tested backup before deployment. Skill installation is not authorization to modify a live site.
For an existing project, try:
Use the installed WordPress security skills to review this project.
Start read-only: identify entry points, authorization boundaries, and risky
inputs and outputs. Report confirmed findings with file/line references,
exploit prerequisites, impact, and recommended fixes. Separate assumptions
and hardening suggestions from confirmed bugs. Propose a prioritized plan
before editing; do not access production or publish changes.
Skills improve the instructions available to an agent; they do not guarantee secure output or replace updates, backups, monitoring, or a qualified review. For reproducibility, record the installed revision. Review upstream changes before updating your local copies, then repeat discovery and behavior checks.
New skills and fixes are welcome — see CONTRIBUTING.md for the SKILL.md contract, description-writing rules, and the WordPress-correctness requirement (verify every API against developer.wordpress.org; never invent functions).
Content in this repository is developed with AI assistance under human direction; see docs/ai-authorship.md for what our validation does and does not establish.
- Security — Common APIs Handbook
- Plugin Security — Plugin Handbook
- Data Validation · Sanitizing · Escaping · Nonces
- Hardening WordPress
- WordPress Coding Standards
- OWASP Top Ten · OWASP Cheat Sheet Series
MIT.