Features

Harness-Controlled Agent Architecture

Models propose. Agents request. The harness decides. File access, code changes, commands, network access and subagent creation all pass through a trusted execution controller that checks permissions before anything happens.

The AI model is never treated as the security boundary. Control stays with the harness.

Evidence-Backed Completion Gates

A model saying a job is complete is not enough. WP HoloDev requires real validation evidence before a job can reach completed status.

PHP syntax checks, WordPress Coding Standards, static analysis, automated tests, security validation and package integrity can all form part of the completion gate. If the required evidence does not pass, the job is not finished.

Recoverable Job State Machine

Every development job moves through tracked stages including analysis, planning, editing, validation, testing, review and packaging.

Objectives, plans, completed work, findings, tests and budgets are persisted throughout the process. If WP HoloDev restarts during a job, it can recover the state and continue from the correct point instead of starting again.

Built-In Cost and Resource Control

Set limits for model calls, agents, concurrent agents, tokens, spending and runtime. Budgets are checked before model calls, subagent creation and tool use, helping keep even complex development work predictable.

Usage and estimated costs are tracked by project, job and agent, with unknown pricing treated as an error rather than silently assumed to be free.

Seven-Type Persistent Memory

Important project knowledge survives between sessions through dedicated stores for project facts, architectural decisions, task history, project rules, code intelligence, validation history and release records.

Memory entries include provenance and confidence information, can be invalidated when their source changes, and remain visible and correctable by you.

Repository Intelligence Indexing

WP HoloDev maps the structure of your codebase, including classes, functions, namespaces, hooks, shortcodes, REST and AJAX endpoints, admin-post handlers, cron events, database tables, queries, options, dependencies and code references.

The index updates as the project changes, giving the agent precise and current code intelligence before it starts making decisions.

60 Built-In WordPress Specialist Skills

A comprehensive library of versioned WordPress skills covers core architecture, plugin and theme development, WPCS, security, REST, AJAX, admin interfaces, Settings and Options APIs, Transients, custom database tables, taxonomies, WP-CLI, cron, multisite, internationalisation, WooCommerce, performance, testing, packaging and wordpress.org submission requirements.

Trust-Levelled Skill System

Skills are classified by trust level, including built-in trusted skills, administrator-approved skills, project-local skills and untrusted skills.

Checksum verification, dependency resolution and controlled upgrades help protect the skill system, while permissions remain under harness control. A skill can request access, but it cannot grant itself additional power.

Scoped Specialist Subagents

The main agent can bring in specialist agents for focused tasks such as security review, repository analysis, database review, WPCS checking, automated testing, REST analysis, performance review and release verification.

Each specialist receives a defined objective, limited scope, permitted files, approved tools and its own resource budget under strict least-privilege controls.

Independent Code Review

Before completion, an independent reviewer can inspect the actual code and diff rather than relying on the primary agent's description of its own work.

Findings are returned with severity, evidence, file and line references, confidence, recommended actions and required tests, giving you a much stronger final review process.

Smart Multi-Provider Model Routing

Use OpenAI, Anthropic and OpenRouter-compatible providers through a unified model layer. Models can be organised into efficient, balanced and expert tiers, allowing WP HoloDev to choose the right level of intelligence for each task.

High-risk work such as security and release review can automatically use expert-tier models, while provider failover helps jobs continue when a service encounters an error.

Deterministic Validation Pipeline

Real development tools can verify every change, including PHP syntax checking, PHPCS with WordPress Coding Standards, PHPStan static analysis, PHPUnit unit and integration tests, Plugin Check and Playwright browser tests.

Your project rules determine which validators are required. The agent cannot simply waive a failed requirement.

Deterministic Security Validation

Rule-based security checks look for problems such as missing nonces, unescaped output, unprepared SQL, missing capability checks, unsafe REST or AJAX permissions, open redirects and file traversal vulnerabilities.

AI security review adds another layer of analysis, but never replaces objective security checks.

Regression-Aware Baseline Diffing

Before editing begins, WP HoloDev can record the project's existing warnings and errors. After the changes, findings are classified as pre-existing, resolved or newly introduced.

This makes it much easier to prove that a change has improved the project without allowing new defects to disappear inside old technical debt.

Disposable WordPress Test Environments

Create clean, isolated WordPress installations with controlled WordPress and PHP versions, dedicated databases, WP-CLI and test fixtures.

Compatibility profiles can test both current and minimum-supported environments, while disposable sites keep development tests away from real customer data and permanent production systems.

Real Browser Testing

Playwright can test the same workflows a real WordPress user would perform, including installation, activation, login, admin pages, settings, content creation, forms, AJAX calls, REST endpoints and role-based access restrictions.

This gives WP HoloDev the ability to verify complete workflows rather than stopping at static code analysis.

WordPress Coding Benchmarks

Built-in benchmark tasks let you measure how configured models perform on realistic WordPress development work. Results can be stored and compared side by side, helping you choose the right models for your own projects.

Immutable Source and Observable Diffs

Uploaded source remains read-only in the original workspace. Changes happen in a separate working area, while validated results are written to output.

Git baselines make every modification inspectable and reversible, while canonical path handling keeps file access inside its permitted workspace.

Hardened Execution Sandboxes

Development commands can run inside locked-down containers using non-root users, dropped Linux capabilities, read-only filesystems, process and resource limits, security profiles and network egress controls.

Sandbox containers never receive access to the Docker socket, helping keep the execution boundary isolated even when untrusted code is being tested.

Prompt-Injection Resistance

Repository content is treated as untrusted data, whether it appears in PHP, JavaScript, documentation, packages or uploaded archives.

Text inside a project cannot become system instructions or override the harness simply because it tells the AI to do so.

Memory Poisoning Protection

Memory records include provenance, and security-sensitive policy can only originate from trusted configuration rather than model inference.

Inferred information can be revalidated against current source before it is used, and memory remains inspectable so incorrect information can be identified and removed.

Secrets Stay Outside the AI Context

Provider credentials are encrypted at rest using Windows DPAPI and are kept out of prompts, project memory, logs, sandboxes and release packages.

Secret detection and automatic log redaction provide additional protection against accidentally persisting API keys or private-key material.

Safe Archive Extraction

Uploaded ZIP files can be checked before extraction for path traversal, absolute paths, symbolic and hard links, device files, excessive compression ratios, excessive file counts and unsafe filename lengths.

This protects the workspace before potentially hostile archive contents are allowed to reach the project filesystem.

Structured Audit Trail

Job transitions, agent creation, tool calls, sandbox activity, validation results, cancellations and security events can all be recorded in a structured audit history.

With automatic secret redaction, you get a detailed and reviewable account of what WP HoloDev actually did during a development job.

Release Packaging with Integrity Proof

Completed projects can produce distributable ZIP packages together with manifests, SHA-256 checksums and validation reports.

The result is a traceable release artifact ready for wordpress.org submission, testing, client delivery or deployment.

WordPress Bridge Deployment and Rollback

Pair WP HoloDev with a WordPress site using a companion bridge protected by HMAC-signed, timestamped and replay-resistant communication.

Validated releases can be deployed to the paired site, while rollback support gives you a controlled route back to the previous verified release when required.

Final-Test Release Locking

Before deployment, WP HoloDev can lock the release source, rerun the required validator suite, fingerprint the final package and bind the release directly to its final-test evidence.

This ensures the code being deployed is the same code that passed the release checks.

Simple Windows Desktop App

Get started without installing Python, PostgreSQL, Redis or a permanent Docker environment. The Windows application includes its own local server, embedded SQLite database, encrypted credential storage and first-run setup for connecting your preferred AI provider.

Docker remains optional and is only needed when you choose to use disposable WordPress test environments.

Self-Hosted Workspace

Manage projects, conversations, memory, jobs, agents, subagents, file changes, validation evidence, release artifacts, model usage and skills from a clean local workspace.

Your WordPress development environment stays under your control rather than becoming dependent on a closed hosted development platform.

Full REST API and Live Event Streaming

The same capabilities available through the interface can be exposed through a versioned REST API covering projects, jobs, memory, skills, artifacts, providers, deployments and usage.

Server-sent events provide live updates for job transitions, agent activity, tool calls and validation results as work happens.

Open-Source and Extensible

Typed extension interfaces allow model providers, sandbox backends, artifact stores, validators, skill sources, repository analysers, fixtures and memory retrieval systems to be replaced or extended without rewriting the core platform.

WP HoloDev is designed as an extensible WordPress development harness, giving developers the freedom to adapt the platform as their requirements grow.