Skip to main content

Use of Claude in this project

Claude Code is Anthropic's AI-powered CLI. It is used in this project as a development assistant, on the tasks described below.

The division of responsibility matters more than the list of tasks. Claude proposes options, argues for one and writes code. Every decision that shapes the product—what to build, which approach to take, what to ship—is made by the maintainer. Where a design choice has consequences the maintainer should weigh, Claude is expected to surface the trade-off rather than pick quietly.

Designing and building features​

Claude works on whole features, not only on isolated changes. The scaffolding CLI and template sync tooling, the chatbot with its pluggable provider support, the managed URL reuse system and the shared dashboard data layer were all developed this way: Claude researches the existing code, proposes a design and implements it once the approach is agreed.

That includes design work. Data shapes, build-time and run-time boundaries, and the structure of new subsystems are drafted by Claude and reviewed before implementation begins. Designs are frequently sent back, and some are rejected outright.

Automating manual tasks​

Claude handles repetitive or scripted work that would otherwise require manual effort—for example, bulk-updating Frontmatter fields across many files, reformatting content to match a new structure or generating boilerplate that follows a known pattern. These are tasks where the shape of the work is clear but execution is tedious.

Developing React components and dashboards​

Claude is used to write and iterate on React components and data dashboards. This includes implementing new UI components, adjusting layout and styling and wiring up data from the TinaCMS GraphQL API. Claude works within the existing component patterns in the project rather than introducing new abstractions.

Defining CMS schemas​

Claude writes the TinaCMS collection and template definitions that drive the editing forms, including field validation and the rules that show or hide fields based on other values.

Writing tests​

Claude writes and maintains the project's test suite. Tests are expected to check behaviour against an authoritative source—an external standard, a specification or the real generated output—rather than restating what the code already says.

Tracking down bugs​

Claude helps narrow down the source of bugs by reading relevant code, checking recent changes and reasoning about likely causes. It is particularly useful for bugs that span multiple files or require cross-referencing configuration with behaviour.

Writing style rules​

Claude is used to write Vale rules for the docStatic style set. This covers authoring new rules, testing them against sample content and refining patterns to reduce false positives. Rules are committed to .vale/styles/docStatic/ alongside the documentation.

Maintaining information architecture​

Claude helps maintain the information architecture of the doc set—reviewing taxonomy, tags and page structure to ensure consistent categorisation and logical navigation. This is an ongoing task, typically run alongside content reviews.

Sanity checking content​

Claude reviews documentation for factual consistency, structural gaps and logical flow. This is a lightweight check—it does not replace a formal editorial review but is useful for catching obvious problems before content reaches reviewers.

Drafting documentation​

Occasionally Claude produces a first draft of a documentation page, which is then reviewed and edited by a human author. Drafts are treated as starting points, not finished content.

Validating changes

Before work is considered done, Claude runs yarn lint (Biome) against code changes and vale against documentation changes, and fixes anything either tool flags.

Scope of AI involvement

docStatic's concept, product direction and priorities are the maintainer's. Claude contributes to the design and implementation of individual features within that direction, and its work is reviewed before it lands. At the time of writing, Claude is a co-author on roughly six per cent of commits.

Related topics

  • MCP Server Integration – Using docStatic with AI assistants through the Model Context Protocol (MCP) server
  • Chatbot – Add an in-browser AI chatbot that answers questions about your documentation, with no external service required.
  • Deployment – Deploying your docStatic site.
  • Prose linting with Vale – Use Vale to enforce a consistent writing style across your docStatic documentation.
  • Static assets – Static assets are the non-code files that are directly copied to the build output. Learn about how they are handled and the best practices for using static assets.