Agent mode
Autonomous multi-step builds. The agent plans, writes code, tests in a sandbox, runs a security audit, and asks for your approval before deploying — all without you writing a line.
All features
Cursor for WordPress. Meet the Snippet Engineer.
Describe the change in plain English. The agent writes the PHP, CSS, and JavaScript, tests it in an isolated sandbox, runs a security audit, and waits for your approval before anything ships.
4
Agent modes
4+
AI providers
0
Lines you write
100%
Approved deploys
How it works
From plain English to deployed WordPress code.
The Snippet Engineer is a multi-step autonomous agent. It plans, writes, validates, self-corrects, and deploys — with you in control at every decision point.
01
Describe what you need
Type your request in plain English. "Add a WooCommerce discount that gives 15% off orders over $100 for logged-in users." The agent handles the rest.
02
Agent writes the code
For multi-file projects, parallel subagents write PHP, CSS, and JS simultaneously. You see each file appear in a live activity card as it's generated.
03
Sandbox & security audit
Before anything goes live, the code runs in an isolated sandbox probe to catch fatal errors. A security risk score is calculated and checked against your configured ceiling.
04
You approve, it deploys
Every deploy is human-in-the-loop by default. Review the plan, approve the publish, and the agent deploys. If something breaks post-deploy, it automatically rolls back.
Four modes, one panel
A Cursor-style interface built into WordPress.
The Snippet Engineer lives in a unified side panel. Four modes cover different workflows, from fully autonomous builds to quick read-only questions.
Chat mode
Conversational help with your snippets. Ask questions, get explanations, request refactors. The agent can read and modify snippets directly in the conversation.
Ask mode
Read-only questions about your snippet library. Ask about what hooks a snippet uses, how it interacts with other code, or what a function does — no changes, just answers.
Plan mode
Review a proposed approach before execution. The agent lays out exactly what it plans to do, and waits for your go-ahead before writing a single line of code.
Context mentions
@-mention the exact context the agent needs.
Type @ in the composer and autocomplete surfaces snippets, projects, hooks, and plugins from your library. Point the agent at exactly the right code — no over-broad context.
- @snippet — reference a specific snippet by name
- @project — scope the agent to a workspace
- @hook — target a WordPress action or filter
- Conversations persist across sessions
Capabilities
Everything you'd expect from a professional agent.
Parallel subagents
Multi-file requests fan out across 2–4 focused code-generation jobs running simultaneously. Each saves its output the moment it finishes — files are written in parallel, not sequentially.
Security audit gates
Every deploy enforces sandbox testing and a security risk score check. You configure the risk ceiling. The agent cannot publish code that fails the audit.
@-mention context
Type @ in the composer to reference specific snippets, projects, hooks, or plugins. The agent gets exactly the right context for the task.
Live streaming
Token-by-token streaming means you watch code appear in real time. No waiting for a spinner — see the agent's thinking and output as it's generated.
Project-aware scaffolding
The agent knows your active project's folder structure. Ask it to scaffold an entire WooCommerce customization workspace with separate snippets for pricing, cart, and emails.
Multi-provider AI
Switch between Anthropic, OpenAI, xAI (Grok), and Google Gemini from the composer. Change models mid-conversation without losing context.
Audit log & signed tokens
Every write, deploy, rollback, and filesystem action is signed and logged. Full traceability for teams and agencies who need to know exactly what ran.
Autopilot mode
For lower-risk work, enable autopilot to let the agent deploy without a per-request approval prompt. All other security gates still apply.
Also new in v1.3.0
Project Workspaces for libraries at scale.
Projects group related snippets and folders into self-contained workspaces. The agent is fully project-aware — it knows the active folder tree and can scaffold a workspace for you.
- Global project switcher — move between workspaces instantly
- Dedicated folder tree and snippet list per project
- Agent scaffolds multi-file projects with correct organisation
- Existing snippets auto-migrated to Default project on upgrade
Built for production
You stay in control. Always.
Every deploy goes through server-enforced gates — not optional, not bypassable.
01
Sandbox probe
Code runs in an isolated probe request before deployment. Fatal PHP errors are caught here — before they can reach your live site.
02
Security risk score
Every snippet gets a risk score from the security audit engine. Your configured ceiling is enforced at deploy time — the agent cannot publish code that exceeds it.
03
Human-in-the-loop
Deploy approval is always required by default. The agent asks, you confirm. Full audit log, signed deploy tokens, and filesystem path restrictions protect every run.
FAQ
Questions
That's the right mental model. Cursor is an AI-native IDE for general development. The Snippet Engineer brings the same autonomous agent experience — multiple modes, parallel subagents, context mentions, live streaming — directly into the WordPress admin, purpose-built for PHP/CSS/JS snippet workflows.
No, by default every deploy is human-in-the-loop. The agent writes and audits the code, then waits for your explicit approval before publishing. Autopilot mode is available as an opt-in for lower-risk work, and all security gates still apply in autopilot.
The agent automatically rolls back to the previous revision if a deployment causes a failure. The sandbox probe also catches fatal PHP errors before deployment, so most issues are caught before they can reach production.
Yes. For larger requests, the agent fans out across parallel subagents that each write one file simultaneously — PHP handler, CSS stylesheet, JavaScript module — and saves them as separate snippets the moment each one finishes.
You can connect Anthropic (Claude), OpenAI (GPT-4o, o1), xAI (Grok), and Google Gemini. The model picker is in the composer — switch providers or models mid-conversation.
The security posture is designed for production. Sandbox testing, security audit risk scoring, human-in-the-loop approval, signed deploy tokens, filesystem path restrictions, and a full audit log are all enforced before any code goes live.
More features
Explore the rest of the workflow.
Get started
Build snippets with an agent at your side.
The Snippet Engineer is included in every SnipVault plan. One-time purchase, lifetime updates.