Skip to content

Unofficial community registry. Not affiliated with, endorsed by, or sponsored by xAI.

中文

Guides

How to read a risk level

safety-ratings

Low, medium, and high are an editor's judgment of permissions. They are not a certificate or a score.

for Grok Bot · unofficial, not affiliated with xAI

CNPS0.1.02026-11-18New

Paste this note into Grok Bot and say: follow this note.

Note: How to read a risk level
Low, medium, and high are an editor's judgment of permissions. They are not a certificate or a score.

For: People who want to know what a skill can touch before it goes into Grok Bot.

What to check:
1. Defines low, medium, and high by capability: local or public data, private reads or drafts, or the ability to send, delete, pay, or change live records.
2. Says plainly that this version has no high-risk entries. That is a choice of what was published, not a claim that high risk does not exist.
3. Gives you a check of your own, with no download count involved.

Steps:
1. Read the risk word, then the sentences under Safety. If they disagree, follow the sentences and treat the entry as the higher level.
2. Compare the connectors you actually enabled. An entry that says read-only becomes send-capable if you turned sending on.
3. Run it once with Test once before you trust it, and watch for an action the entry does not name.
4. If you are unsure, keep it out of a routine that runs by itself.

Low: reads public pages or public DNS, or writes files in your workspace. The task must not send or delete.
Medium: reads mail, CRM, notes, sheets, or team chat, or writes drafts. The task must still not send, delete, pay, or change DNS. Gmail can send and Slack can post.
High: can send, delete, pay, post, or change live domain, access, or production data.
No entry in this first version is marked high.

Do not:
- Do not treat a risk level as an xAI review or a security certification.
- Do not use a download count or a star rating to judge an entry.
- Do not ignore a wider permission on your machine because an entry looks narrower.

Paste this note into Grok Bot and say: follow this note.

What it does

  • Defines low, medium, and high by capability: local or public data, private reads or drafts, or the ability to send, delete, pay, or change live records.
  • Says plainly that this version has no high-risk entries. That is a choice of what was published, not a claim that high risk does not exist.
  • Gives you a check of your own, with no download count involved.

Who it is for

People who want to know what a skill can touch before it goes into Grok Bot.

How to put it in Grok Bot

  1. Read the risk word, then the sentences under Safety. If they disagree, follow the sentences and treat the entry as the higher level.
  2. Compare the connectors you actually enabled. An entry that says read-only becomes send-capable if you turned sending on.
  3. Run it once with Test once before you trust it, and watch for an action the entry does not name.
  4. If you are unsure, keep it out of a routine that runs by itself.

Example

What the three words mean

Low: reads public pages or public DNS, or writes files in your workspace. The task must not send or delete.
Medium: reads mail, CRM, notes, sheets, or team chat, or writes drafts. The task must still not send, delete, pay, or change DNS. Gmail can send and Slack can post.
High: can send, delete, pay, post, or change live domain, access, or production data.
No entry in this first version is marked high.

Connectors it needs

No connector.

Safety and permissions

  • Do not treat a risk level as an xAI review or a security certification.
  • Do not use a download count or a star rating to judge an entry.
  • Do not ignore a wider permission on your machine because an entry looks narrower.

Risk: Low

Questions

If it says low, can I skip the rest?

No. Low means the editor saw a narrow scope. You still check which files it reads and whether it will treat text on a web page as an instruction.

Why is there no score?

This version has no usage record. A score would be invented. The risk word describes capability, not popularity.