Skip to main content
The Hiveku CMS is designed to be edited by both humans and the AI. The AI has a small set of tools for reading and writing the manifest, listing entries, and reading/writing/deleting them — which means you can describe content changes in plain English and let the AI carry them out. This page covers what the AI can do, the tools it uses (high level), and how to prompt it well for content tasks.

​
What the AI Can Do

Scaffold a manifest

From an empty project, propose a complete manifest based on what your site needs

Add or change collections

Add new collections, add fields to existing ones, change field types

Write and edit entries

Create entries from a description, an existing file, or a paste of content

Refactor components

Replace hardcoded content in React components with reads from the CMS

Bulk operations

Migrate from CSV/JSON, normalize data, rename slugs, deduplicate

Schema migrations

Rename fields and migrate existing entry data to the new shape

​
The Tools the AI Uses

The AI has six tools dedicated to the CMS, exposed through the project’s MCP server. You won’t call them directly, but knowing they exist helps you prompt the AI more precisely.
ToolWhat it does
manage_cms_scaffoldCreate a fresh hiveku.cms.json (and optional starter content) for a new project
manage_cms_read_manifestRead the current manifest
manage_cms_update_manifestUpdate the manifest — add, remove, or modify collections and fields
manage_cms_list_entriesList entries in a collection (with search, filter, sort)
manage_cms_read_entryRead one entry’s contents
manage_cms_write_entryCreate or update one entry
manage_cms_delete_entryRemove an entry
Each tool runs the same validation as the CMS panel, so the AI can’t write an invalid entry or break the manifest schema. Failed writes return error messages the AI can use to retry or ask you for clarification.

​
Prompting Patterns

​
Set up a CMS for a new project

Best when starting from scratch or with a minimal site: 
Set up a CMS for this site. We'll need a blog with posts, an author
collection that the blog references, and a testimonials collection.
The AI will:
  1. Read the project to understand the framework and existing routes
  2. Propose a manifest with the three collections and sensible fields
  3. Confirm with you
  4. Save the manifest
  5. Optionally seed each collection with one example entry

​
Migrate hardcoded content to the CMS

Best when you have content embedded in code that you want to extract: 
Look at app/blog/page.tsx — it has 12 hardcoded blog posts in an
array. Move them into a blog collection in the CMS, and refactor
the page to read from the CMS instead.
The AI will:
  1. Read the file and extract the hardcoded data
  2. Propose a blog collection schema based on the shape of the data
  3. Confirm with you
  4. Add the collection to the manifest
  5. Write one entry per post
  6. Refactor the React component to import from the CMS
  7. Show you a diff and let you verify
See Migrate an Existing Site to the CMS for the full walkthrough.

​
Write a new entry

Best for one-off entries with rich content: 
Add a blog post titled "Why we switched to Hiveku" — make it 500
words, professional but conversational tone. Tag it as "case-study"
and "engineering". Set publishedAt to today.
The AI will draft the post, write it as an MDX file, and (optionally) show you a preview before saving.

​
Bulk edits

Best when many entries need the same change: 
For all blog posts published before 2025, set status to "archived"
and add the tag "legacy".
The AI will list matching entries, confirm the count, and apply the change in a batch.

​
Schema changes

Best when you need to add a field across an existing collection: 
Add a "readingTime" field (integer, minutes) to the blog collection.
Calculate it for every existing post (~200 words per minute) and
backfill the value.
The AI will:
  1. Update the manifest
  2. Read each existing entry, compute the reading time, write it back
  3. Report how many entries were updated

​
Refactor a component to use the CMS

Best when you want existing UI to start reading from a collection: 
Refactor app/team/page.tsx to read from the team collection in the
CMS. Right now the team members are hardcoded in a JSX array.
The AI will rewrite the component to import from the CMS, type the data correctly, and remove the hardcoded array.

​
Prompting Tips

A few patterns that consistently get better results:
  • Anchor the AI to a file or feature. “Look at app/blog/page.tsx” beats “Migrate my blog” because the AI doesn’t have to guess where the content lives.
  • Specify the schema if you know it. If you already know your blog needs title, excerpt, publishedAt, body, tags, author, say so. The AI will fill in sensible defaults but it can’t read your mind.
  • Do one collection at a time. A single migration is easier to verify than three at once. The AI can chain them, but you get cleaner diffs from sequential prompts.
  • Ask for a confirmation first. “Plan the migration but don’t write anything yet.” Then after you’ve reviewed the plan, “Now do it.”
  • Use natural language for filters. “Posts from last quarter,” “products under $50,” “FAQs in the billing category” — the AI translates these into list filters automatically.

​
Trust Boundaries

A few things the AI is conservative about:
  • Deletes. The AI confirms before deleting any entry. You can override this with explicit prompts (“delete every post tagged ‘old’ without asking”).
  • Schema removals. Removing a field from the manifest is non-destructive (data stays in the file), but the AI calls it out in the diff so you don’t lose track.
  • Renames that orphan data. Renaming a field in the manifest doesn’t move data on disk. The AI flags this and offers to write a migration.
  • Reference cascades. Deleting an entry that’s referenced by others will leave dangling references. The AI warns and offers to clean them up.

​
Inspecting What the AI Did

Every CMS write the AI makes goes through the same versioning system as your manual edits. After any AI-driven change:
  • Open the CMS panel — the new or modified entries are in the entry list
  • Click any entry → three-dot menu → Version history to see the AI-authored snapshot
  • Open the file in the code editor to inspect the raw MDX/JSON
  • Click Restore this version if you want to roll back

​
Example: Migrating a Site With AI

A condensed version of the full migration guide:
1

Describe the source

“My homepage has a hardcoded <TestimonialsSection /> with 6 testimonials embedded in JSX. I want non-developers to be able to edit them.”
2

Let the AI plan

“Plan the migration — propose a schema and show me the field shape.”The AI returns a draft manifest entry with name, company, quote, avatar, featured fields.
3

Approve and execute

“Looks good. Add the collection, migrate the 6 testimonials, and refactor the component to read from the CMS.”The AI:
  • Updates hiveku.cms.json with the testimonials collection
  • Writes 6 JSON files to content/testimonials/
  • Rewrites <TestimonialsSection /> to read the collection
  • Deploys a preview
4

Verify in the CMS panel

Open the panel, switch to the testimonials collection, edit one. The change appears in the live preview.

​
What’s Next?

Migrate a Site

Full walkthrough for moving hardcoded content into the CMS

Initialize Your CMS

First-time setup

Editing Content

The CMS panel for human editors

The Manifest

What the AI is reading and writing