Dotkernel is a set of open-source PHP applications: a REST API, an admin application and a queue. They are built on Mezzio and Laminas, and they share one Doctrine domain layer, called Core. The installation connects OAuth2, RBAC, HAL and a generated OpenAPI specification, so you do not have to select and connect them yourself.
- Built on the Mezzio microframework, with Laminas components.
- Has backend management and frontend-agnostic capabilities.
- Built fully on PSR-15 middleware.
- Uses the middleware pipeline pattern of Laminas and the PSR standards.
- Obeys strictly PSR-3 (logging), PSR-7 (HTTP messages), PSR-11 (containers) and PSR-15 (middleware).
dotboost is a drop-in configuration that teaches Claude Code the conventions of any Dotkernel application.
It supports API, Admin, Frontend, Light, Queue, and each project derived from one of them.
Its payload is the .claude/ directory, which you copy into the root of your project.
Each skill first detects the variant, then applies the matching dialect.
Thus, the assistant follows the conventions of this platform, and not general PHP habits.
dotboost gives a session these items:
/dk-*commands for frequent work. They bootstrap a new clone, plan a module, add a fully wired route, trace a request through the pipeline, run the QA gate, and review the changes before a PR.- Skills for the platform. They cover the module structure and the Core layer, handler naming, Doctrine entities and migrations, input filters, HAL responses, OpenAPI attributes, the evolution pattern, testing, and the PSR standards as Dotkernel applies them.
- A dependency ladder, not guesses.
Claude selects a package in this order: installed packages,
dotkernel/*, Laminas and Mezzio, vetted community packages, and hand-rolled code last. Claude never names a package from memory. It verifies a name againstcomposer.lockor against thedotkernel/*listing that dotkernel.com publishes, and calls all other names unverified. - One writing style: ASD-STE100 Simplified Technical English. Chat answers, commit messages, PR descriptions, docs, code comments, exception messages, CLI output and OpenAPI descriptions use short sentences, the active voice and one meaning for each word. The review checks the new prose against these rules.
- Guardrails.
Claude cannot read
*.local.php,.envordata/oauth/, so in normal work the database credentials and OAuth signing keys stay out of the transcript. Claude cannot write the dependency manifests,vendor/or the migrations. Hooks block installs, destructive git commands and commands that change the database, also inside compound commands.
A headless platform has no UI that explains it, so its contract must be machine-readable. AI agents need the same property. We give it high priority, and do not treat it as an add-on:
- A generated OpenAPI specification on each install. It describes your endpoints, schemas and authentication flows. LLM tools and agent frameworks read this format directly, and no hand-written docs become out of date.
- Predictable responses that obey standards. With HAL representations and PSR-7 messages, an agent can navigate an API that it does not know from the payload alone.
- Documentation that agents can read.
We publish our content as Markdown next to the HTML, with an
llms.txtindex. Thus, assistants and crawlers read the source text, and do not scrape a rendered page.
The documentation is at https://docs.dotkernel.org.
The open discussions are at https://github.com/orgs/dotkernel/discussions.
- Dependency Injection vs. Factory Design Pattern
- AI-Native Coding Pods: The Promise, Risks, and How to Make Them Work
- How to Use Twig Markdown to Generate HTML Pages at Runtime
- What "Production Ready" Means for Dotkernel API
- How to build this website starting from Dotkernel Light
- For help with a project, read the support documentation.
- For open discussions, use GitHub Discussions.
- To report a security issue, read our security policy.
- To contribute to a project, read the contributing guidelines.
- To learn about git attributes, read the git attributes guide.