Claude Skills for Directus: 9 SKILL.md Files for Items and Collections
Quick answer: Save each Directus chore as a Claude skill that names the tool to call and the table to hand back. Ryze AI is the best Claude connector for running these skills on a Directus instance in 2026, with one caveat: it connects one Directus instance per workspace, using a static token whose user role decides which collections the skills can see.
Below: 9 SKILL.md files, which of the 3 Directus tools each one calls, install steps, the limits that apply and a short troubleshooting list.
Ryze AI · SEO Starter
Directus skills on a live instance
- 3-day free trial, $129 a month
Used by 2,000+ marketers
The 9 Claude skills for Directus at a glance
Five skills only read, three write and one removes. Each names the Ryze AI tool it calls.
- 01Data model reviewList collections and their fields
Tool used: directus__read
Cadence: Monthly
- 02Status backlogFind items still in draft
Tool used: directus__read
Cadence: Weekly
- 03Field gapsFind items with empty fields you name
Tool used: directus__read
Cadence: Per release
- 04File metadata checkList files with no title or description
Tool used: directus__read
Cadence: Quarterly
- 05Flow inventoryList flows and what they are set to do
Tool used: directus__read
Cadence: Quarterly
- 06Draft an itemCreate one item from your text
Tool used: directus__read, directus__write
Cadence: Per article
- 07Refresh item fieldsChange chosen fields with a before and after
Tool used: directus__read, directus__write
Cadence: Per refresh
- 08Bulk status changeUpdate the status of the items you list
Tool used: directus__read, directus__write
Cadence: Per batch
- 09Delete itemsDelete the items that match a rule you confirm
Tool used: directus__read, directus__delete
Cadence: Quarterly
| # | Skill | Job | Tool used | Cadence |
|---|---|---|---|---|
| 01 | Data model review | List collections and their fields | directus__read | Monthly |
| 02 | Status backlog | Find items still in draft | directus__read | Weekly |
| 03 | Field gaps | Find items with empty fields you name | directus__read | Per release |
| 04 | File metadata check | List files with no title or description | directus__read | Quarterly |
| 05 | Flow inventory | List flows and what they are set to do | directus__read | Quarterly |
| 06 | Draft an item | Create one item from your text | directus__read, directus__write | Per article |
| 07 | Refresh item fields | Change chosen fields with a before and after | directus__read, directus__write | Per refresh |
| 08 | Bulk status change | Update the status of the items you list | directus__read, directus__write | Per batch |
| 09 | Delete items | Delete the items that match a rule you confirm | directus__read, directus__delete | Quarterly |
The tool names and descriptions come from the Ryze AI integration source and the Ryze AI help center. Directus was not connected in any Ryze AI workspace available to us, so no skill here was run against a live instance and no numbers from a real instance appear. Run each skill on a test collection first.
What are Claude skills for Directus?
A Claude skill for Directus is a SKILL.md file that tells Claude how to run one content task and what to return. Anthropic's Agent Skills docs say the file opens with a name and a description. Claude loads those two fields at startup and reads the rest of the file only when a request matches the description.
The name takes up to 64 characters, lowercase letters, numbers and hyphens, and cannot contain the reserved words anthropic or claude. The description is capped at 1,024 characters and must say what the skill does and when to use it. Every skill here starts with directus for that reason.
What it is
- Chat prompt
- One question typed into chat
- Skill
- A saved file with fixed steps and a fixed output
- Directus connection
- Tools Claude calls to reach the instance
Setup
- Chat prompt
- None
- Skill
- Save once for each Claude app
- Directus connection
- An Instance URL and a static token entered in Ryze AI
Repeatable
- Chat prompt
- Only if you retype it
- Skill
- Yes, the same steps every run
- Directus connection
- Depends on the prompt or skill
| Chat prompt | Skill | Directus connection | |
|---|---|---|---|
| What it is | One question typed into chat | A saved file with fixed steps and a fixed output | Tools Claude calls to reach the instance |
| Setup | None | Save once for each Claude app | An Instance URL and a static token entered in Ryze AI |
| Repeatable | Only if you retype it | Yes, the same steps every run | Depends on the prompt or skill |
A skill holds instructions and no credentials, so one file works on any instance you connect. For the tool list, setup per client and the comparison with the Directus MCP server, see the Directus MCP page.
How to install the Directus skills in Claude
Claude Code
Put each block in ~/.claude/skills/<skill-name>/SKILL.md for yourself, or in .claude/skills/<skill-name>/SKILL.md inside a project to share it. Type /directus-draft-item to call a skill by name, or let Claude pick it from the description (Claude Code skills docs).
claude.ai
Zip each skill folder and upload it as a custom skill. Anthropic lists custom skills on Pro, Max, Ultra, Team and Enterprise plans with code execution enabled.
Connect the instance
Connect Directus inside Ryze AI, then add the Ryze AI connector to Claude. The steps are in the next sections, on the Directus MCP page and in the Ryze AI help center.
Anthropic states that custom skills do not sync between surfaces: a skill uploaded to claude.ai is absent from the API, and Claude Code skills live on disk apart from both. Anthropic also advises using skills only from sources you trust, so read each file before you save it, the ones on this page included.
The four skills that write or delete carry disable-model-invocation: true in their frontmatter. Per the Claude Code docs, that stops Claude from starting a skill by itself and leaves it to you to type the slash command. It applies to Claude Code, and it does not replace per-tool approvals in the connector.
Read skills: review the instance before you touch it
These five call only directus__read, which runs a read-only Directus REST request. Items, collections, fields, files, users, flows and server info are within its reach, and none of these skills changes anything. What comes back is limited to what the token's user role can see.
01. Data model review
Lists the collections on the instance with their fields, so you know the model before you ask for edits.
Use it when: you inherit an instance or plan a bulk edit.
02. Status backlog
Finds items in one collection that were never published. Many Directus collections carry a status field with values such as draft and published, though the field and its values are your own choice, so the skill asks which one to use.
Use it when: a post exists in the Data Studio but not on the site.
03. Field gaps
Checks the fields you name for empty values across one collection. Directus returns 100 items when a request sets no limit, so the skill asks for paging.
Use it when: a content team wants every post to carry a summary and a meta description.
04. File metadata check
Lists files in the Directus file library with an empty title or description. Alt text for images often starts from this metadata, so gaps matter for accessibility and image search.
Use it when: a page audit flags images with no alt text.
05. Flow inventory
Lists the flows on the instance, which are Directus automations, so you can see what runs when items change.
Use it when: you are about to run a bulk update and want to know what else reacts to it.
Write skills: create, refresh and bulk update items
These three call directus__write, which creates or updates items through the Directus REST API (POST or PATCH), including bulk updates of many items. In Claude, open Customize > Connectors > Ryze AI > Tool permissions and set write and delete tools to Needs approval, the setting the Ryze AI help center recommends. Each change then waits for your click.
06. Draft an item
Creates one item from text you paste. The skill sets the status field to its draft value when the collection has one, so nothing goes live by accident.
Use it when: you have a finished article and want it in the collection without pasting it into the Data Studio. For scheduled publishing, Ryze AI's Content Plan publishes articles into the collection you choose at setup.
07. Refresh item fields
Changes chosen fields on an existing item. The skill reads the whole item first and shows a before and after, so fields you did not mention stay as they are.
Use it when: a post needs a corrected fact or a new intro.
08. Bulk status change
Updates the status of the items you list in one go. The write tool supports bulk updates, which is why the skill insists on a numbered list and a count before it runs.
Use it when: an editor signed off on a batch. The token user needs update access to the collection and its status field; if the role lacks it, Directus refuses the change.
Cleanup skill: delete items with a confirmation step
directus__delete deletes one or many items through the Directus REST API. Deleting is final, so the tool sits apart from write, and you can allow edits while keeping removals on approval.
09. Delete items
Use it when: you are clearing test items before a launch. Keep Needs approval on the delete tool in Claude, in addition to the skill's own confirmation step.
How to run the skills on a live Directus instance
A skill describes the work and Claude needs a connection to do it. The Ryze AI connector gives Claude 3 Directus tools:
directus__read
- What it does
- Runs a read-only Directus REST request, such as items in a collection, collections, fields, files, users, flows and server info
- Used by
- All 9 skills
directus__write
- What it does
- Creates or updates items through the Directus REST API (POST or PATCH), including bulk updates of many items
- Used by
- Skills 06, 07, 08
directus__delete
- What it does
- Deletes one or many items through the Directus REST API
- Used by
- Skill 09
| Tool | What it does | Used by |
|---|---|---|
| directus__read | Runs a read-only Directus REST request, such as items in a collection, collections, fields, files, users, flows and server info | All 9 skills |
| directus__write | Creates or updates items through the Directus REST API (POST or PATCH), including bulk updates of many items | Skills 06, 07, 08 |
| directus__delete | Deletes one or many items through the Directus REST API | Skill 09 |
To connect, open a Directus user with the right role for your blog collection, click Generate next to Token, copy it and save the user. In Ryze AI, open Integrations, click Connect on the Directus card and enter the Instance URL (the Data Studio origin, https://, no path), the token and your public Blog URL. Then pick the collection articles publish to and, for translated collections, a language code (Ryze AI Directus guide).
The connector needs an active Ryze AI plan or trial, and connecting it costs nothing. SEO Starter is $129 a month with a 3-day free trial, and annual billing is 30% off the first year (pricing). After you connect Directus in Ryze AI, remove the Ryze AI connector in Claude, add it again and start a new chat, because Claude keeps the tool list from the moment the connector was added (Ryze AI help center).
Two scope limits matter for skills. A workspace holds one Directus connection, so a skill cannot compare two instances. The token also follows its user's role: Directus says static tokens inherit the permissions of the user they belong to, so a collection the role cannot see is invisible to every skill.
Directus ships its own MCP server for v11.12 and later. You turn it on under Settings > AI and generate a token for a dedicated MCP user, and its tools cover schema, items, collections, files and assets (Directus MCP guide). Its tool names differ from the Ryze AI names, so a skill written for directus__read would need editing before it works there. The Directus MCP page carries the full comparison.
Limits that shape these skills
Directus connections per Ryze AI workspace
- Value
- 1
- Effect on a skill
- Skills run against one instance at a time
Token role
- Value
- Same permissions as the token's user
- Effect on a skill
- Skills 06 to 09 fail where the role cannot create, update or delete
Default page size (QUERY_LIMIT_DEFAULT)
- Value
- 100 items
- Effect on a skill
- Skills 03 and 04 ask Claude to read in pages
Maximum page size (QUERY_LIMIT_MAX)
- Value
- -1, no cap by default
- Effect on a skill
- An admin may set a cap on your instance
Rate limiter (RATE_LIMITER_ENABLED)
- Value
- Off by default; 50 points per 1 second when on
- Effect on a skill
- Large audits may slow down on instances that turned it on
Custom skill upload in claude.ai
- Value
- Pro, Max, Ultra, Team and Enterprise with code execution
- Effect on a skill
- Free users can still use the connector but not uploaded skills
Free plan custom connectors
- Value
- 1 in total
- Effect on a skill
- Ryze AI would use that single slot
Skill name
- Value
- 64 characters, lowercase letters, numbers, hyphens
- Effect on a skill
- Names cannot include claude or anthropic
| Limit | Value | Effect on a skill |
|---|---|---|
| Directus connections per Ryze AI workspace | 1 | Skills run against one instance at a time |
| Token role | Same permissions as the token's user | Skills 06 to 09 fail where the role cannot create, update or delete |
| Default page size (QUERY_LIMIT_DEFAULT) | 100 items | Skills 03 and 04 ask Claude to read in pages |
| Maximum page size (QUERY_LIMIT_MAX) | -1, no cap by default | An admin may set a cap on your instance |
| Rate limiter (RATE_LIMITER_ENABLED) | Off by default; 50 points per 1 second when on | Large audits may slow down on instances that turned it on |
| Custom skill upload in claude.ai | Pro, Max, Ultra, Team and Enterprise with code execution | Free users can still use the connector but not uploaded skills |
| Free plan custom connectors | 1 in total | Ryze AI would use that single slot |
| Skill name | 64 characters, lowercase letters, numbers, hyphens | Names cannot include claude or anthropic |
The Directus limits come from its security limits reference, opened November 19, 2026. Each is an environment setting, so your instance, or a managed one, may use other values.
Troubleshooting: when a Directus skill fails
Claude has no Directus tools
Claude keeps the tool list from when the connector was added. Connect Directus in Ryze AI first, then remove and re-add the connector and start a new chat.
The token is rejected
A token generated without saving the user afterwards is rejected. Open the user, generate the token again, save the user and reconnect.
A collection is missing from results
Collections the token's role cannot see are not available. Give the role read access to the collection in Directus.
Instance URL is refused
Use the Data Studio origin with https:// and nothing after the domain.
Lists stop at 100 items
Directus applies a default limit of 100 when a request names none. Ask Claude to read in pages.
Connect link refused
A connection link is refused while a Directus connection already exists in the workspace. Click View, disconnect with the trash icon and connect again.
A skill does not start in Claude Code
Check the path is ~/.claude/skills/<name>/SKILL.md. Skills marked disable-model-invocation: true start only when you type the slash command.
Sources
The verdict
Turn each recurring Directus chore into a skill that names its tool and ends with a confirmation step. Start with the data model review and the status backlog, then add the write skills once approvals are set.
Ryze AI runs these skills in Claude through 3 Directus tools on one instance. The token user's role decides what each skill can see and change.




