WP HoloDev turns AI coding into a structured development workflow. Instead of sending a prompt to a model and hoping the generated code is correct, the harness gives AI coding agents controlled access to the project, development tools, specialist skills, persistent context, testing and validation.
You describe what you want to achieve. WP HoloDev can inspect the existing project, work out what needs changing, carry out the development, test the result and continue when something fails. Permissions, approvals, models, external tools and release actions remain under harness control throughout.
1. Start With a Real Development Project
WP HoloDev works around real projects rather than isolated AI conversations. Each project provides the codebase, settings and persistent development context the harness needs to understand what it is working on.
New Project or Existing Codebase
You can start something new or point WP HoloDev at an existing project. For WordPress development that might be a plugin, theme or custom website containing PHP, JavaScript, CSS, templates, REST endpoints, database logic and supporting files.
The project becomes the home for its conversations, jobs, memory, model policies, working mode, validation and deployment information, keeping unrelated development work separated.
2. Decide How You Want the AI to Work
Before starting a job, you can control three separate parts of the AI development process: what kind of work WP HoloDev should optimise for, which models it is allowed to use and how much agent delegation is permitted.
Choose a Main Mode
Select Coding, WordPress Coding, Design, WordPress Design or Cyber Security.
The active mode influences the prompts, capabilities, skills, tools and validators used for the task without changing your model or agent controls.
Choose Model Assignment
Full Auto lets WP HoloDev choose the cheapest model that is actually sufficient for the work. Semi-Auto allows automatic selection only from the exact models and provider connections you approve. Manual locks the entire model-backed job to one exact provider, account and model.
Choose an Agent Level
Full allows specialist subagents when they are genuinely useful. Low allows a maximum of four subagents for the job. None keeps all model-backed work with the main agent while still allowing tools, skills, validation and testing.
3. Tell WP HoloDev What You Want
Development begins with an ordinary request. Describe the feature you want built, the bug you want fixed, the code you want reviewed or the result you need from the project.
Describe the Outcome Rather Than Every Step
You do not normally need to tell the agent which file to open, which search to run, which skill to load or which tool to use. WP HoloDev can investigate the project and select appropriate capabilities as the task progresses.
You might ask it to fix a broken settings screen, add a feature to an existing plugin, redesign an interface, investigate a REST endpoint or perform a defensive security review.
The conversation can stay centred on the result you want rather than becoming a stream of low-level development instructions.
4. The Harness Builds Context From the Project
Before changing code, the agent can inspect the actual project and identify the files, functions and relationships relevant to the task.
Project-Aware Development
WP HoloDev can search and analyse project files, inspect surrounding implementations and trace relationships instead of working only from a pasted code snippet.
For WordPress projects, specialist capabilities help the agent work with hooks, actions, filters, shortcodes, REST endpoints, AJAX handlers, database access, custom post types, taxonomies, WP-CLI and other WordPress structures.
This gives the agent a better chance of changing the correct part of the project without unnecessarily disrupting the architecture around it.
5. Existing Project Knowledge Can Be Reused
Useful context does not have to disappear at the end of every conversation.
Persistent Project Memory
WP HoloDev can retain useful project-specific information between development sessions, including important implementation details, decisions and other context that may help future work.
Memory remains associated with the project and can be inspected rather than becoming an invisible source of instructions.
This reduces unnecessary rediscovery when you return to a project later.
6. Skills, Tools and Models Are Selected for the Work
WP HoloDev can automatically assemble the capabilities required for the current task rather than forcing you to configure every part of the workflow manually.
Automatic Skill Selection
Relevant development skills can be selected from the task, project and active Main Mode. These can include debugging, testing, WordPress development, security, performance, PHPStan, REST APIs, WP-CLI, design and code review.
Automatic Tool Selection
The harness can also make relevant native tools and validators available automatically while still passing execution through the same ActionRegistry, permission and project-isolation controls.
Model-Specific Prompting
WP HoloDev compiles prompts for the actual model performing the work. If Full Auto or Semi-Auto reroutes a job to another suitable model, the correct model-specific prompt profile is compiled for the model that actually receives the task.
7. The Job Continues as Structured Work
A WP HoloDev job is more than a single model response. Requirements, TODOs, progress, delegated work and validation state can be tracked separately from the conversation.
The Model Does Not Decide When the Job Is Finished
If actionable structured work is still pending or in progress, a confident final message from the model does not automatically make the job complete.
The harness can continue the same job while outstanding work remains, preserving the job's project, Main Mode, Model Assignment, Agent Level, TODOs and completed work.
Jobs Can Survive Long Development Sessions
Continuation and context compaction allow long-running work to proceed without forcing every earlier tool result and conversation turn to remain in the active model context forever.
8. Larger Jobs Can Use Specialist Subagents
Some development tasks benefit from splitting independent or specialist work into separate model-backed assignments.
Purposeful Delegation
Subagents can be used for focused investigation, implementation, testing, debugging, design, WordPress development, review or defensive security work.
They inherit the running job's project and policy constraints rather than becoming uncontrolled independent agents.
You Control How Much Delegation Is Allowed
Full Agent Level permits subagents when useful, Low restricts the entire job to four logical subagents and None keeps all model-backed work with the main agent.
Model retries or provider fallback for the same logical worker do not artificially consume additional subagent slots.
9. Development Tools Are Used Through the Harness
The AI can do more than describe changes. WP HoloDev gives agents controlled access to tools that allow them to participate directly in development.
Inspect, Search, Edit and Execute
Depending on the task and permissions, an agent can inspect project files, search the repository, edit code, run development commands, examine command output and work across multiple related files.
Native tool calls go through WP HoloDev's ActionRegistry, argument validation, permissions and project boundaries rather than being executed simply because a model requested them.
Additional Capabilities Through MCP
Compatible MCP tools, resources and prompts can also become part of the workflow when MCP servers are connected.
MCP remains an extension of the harness rather than a route around its project boundaries and security controls.
10. Sensitive Operations Stay Under Harness Control
Giving an AI useful development capabilities does not mean giving it unrestricted access to the computer.
Project Boundaries
Filesystem operations are constrained to authorised project workspaces with protections against traversal, unsafe links, Windows reparse points, junctions and other path-escape techniques.
Controlled Commands and Tools
Tool registration, argument validation, command restrictions and execution controls are enforced separately from model instructions.
Protected Credentials
Provider credentials remain separate from ordinary project prompts and are protected against unnecessary exposure through logs, diagnostics and stored metadata.
The AI can ask to perform development work, but it is not itself the authority that decides which security boundaries apply.
11. Failures Stay Inside the Development Loop
Real development does not always work on the first attempt, and WP HoloDev is designed around that reality.
Inspect, Correct and Test Again
A failed test, syntax error, static-analysis finding, build failure or unexpected browser result can remain part of the active development task.
The agent can inspect the failure, return to the relevant code, make a correction and rerun the appropriate checks instead of treating the first failed attempt as the end of the task.
The workflow is designed around iteration: investigate, change, test, inspect, fix and test again.
12. Real Development Tools Check the Result
Generated code is not treated as proof that the task succeeded.
Deterministic Validation
Depending on the project, WP HoloDev can run checks such as PHP syntax validation, WordPress Coding Standards, PHPStan, PHPUnit, Plugin Check and other project-specific test or build commands.
Validation results become part of the development evidence rather than relying on the model to judge its own output.
Regression-Aware Checking
Existing findings can be identified before development begins so later validation can distinguish older technical debt from regressions introduced by the current change.
Holographic Plugin Check
For WordPress plugins, the bundled Holographic Plugin Check workflow provides focused code quality, security, optimisation and release-readiness checking.
It can also be accessed directly through commands such as /check, /holo and /audit.
13. The Agent Can Test What Happens in the Browser
Source code can be technically valid while the interface or website it produces is still broken.
Browser-Level Evidence
WP HoloDev includes Playwright-based browser automation so development work can inspect real pages and interfaces.
The agent can load pages, exercise supported interactions, investigate browser behaviour and capture screenshots during testing and debugging.
This is especially useful for administration screens, forms, JavaScript, CSS, responsive interfaces and other behaviour that cannot be confirmed by static code validation alone.
14. Review Does Not Have to Trust the Implementing Agent
The system that wrote a change does not have to be the only system evaluating whether the result is correct.
Independent Review
A separate review stage can inspect the actual implementation, project changes, test evidence and validation findings.
Important findings can return to the development loop for correction before the overall job is accepted as complete.
This is particularly useful for larger changes, security-sensitive work and final release reviews.
15. You Can Inspect What Actually Happened
The main conversation does not need to contain every internal operation, but those operations do not have to be hidden.
Execution Trace and Audit History
WP HoloDev can record model routes, tool usage, agents, validation results, lifecycle decisions and other development activity separately from the normal conversation.
This gives you a way to inspect how the result was produced without forcing every low-level action into the chat.
Common management operations are also available through slash commands such as /status, /skills, /tools, /memory, /models and /cancel.
16. Different Models Can Handle Different Work
The AI model is one component of the harness, and WP HoloDev does not require every task to use the same provider or model.
Full Auto Chooses the Cheapest Suitable Route
WP HoloDev can evaluate model capability, context requirements, required tools, availability and cost before choosing a route.
A cheaper model can handle straightforward work when it is sufficient, while a more capable model can be selected when the actual task requires it.
Semi-Auto Keeps Selection Inside Your Allowlist
You can restrict automatic routing to exact provider connections and models that you approve for a project.
Manual Means One Exact Model
If you choose Manual, every model-backed part of that job stays on the exact provider connection and model you selected without silent fallback to another account.
17. Move Validated WordPress Work Towards Release
For WordPress projects, the workflow can continue beyond implementation, testing and review.
Prepare the Release Package
WP HoloDev can prepare a distributable project ZIP after development and validation are complete.
Release artefacts can be fingerprinted and checked so later deployment can verify that the package being used is the one that was actually validated.
Deploy Through the WordPress Bridge
A project can be paired with an authorised WordPress installation through the HoloDev deployment bridge.
Authenticated requests, project authorisation and integrity protections allow deployment without giving the coding model ordinary FTP credentials or unrestricted server access.
Verify and Roll Back
Post-deployment verification can check the resulting installation, while rollback provides a controlled way to return to a previous verified deployment when necessary.
18. Local and Docker-Based Testing Can Work Together
WP HoloDev itself runs locally on Windows without requiring a large supporting server stack.
Normal Development Runs Locally
The harness includes its own Node.js runtime, application server and SQLite database. Python, PostgreSQL, Redis and Docker are not required simply to start using it.
Docker When Isolation Is Useful
When Docker Desktop is available, WP HoloDev can use isolated WordPress environments for runtime testing without performing development checks against real production or customer sites.
From Request to Verified Result
The WP HoloDev workflow is designed around a straightforward idea.
Choose how you want the AI to work. Tell the harness what you need. Let it understand the project. Give the agent controlled capabilities to do the work. Keep the job running while actionable work remains. Test the result. Fix what fails. Review the implementation. Complete the job when the required evidence exists.
The AI provides the reasoning and coding ability. WP HoloDev provides the project context, model control, tools, skills, agents, security and development workflow needed to turn that ability into practical software development.