Skills
Review, edit, approve, archive and import the skills your assistant can load from GuruSup Brain.
A skill is a set of instructions, written in a SKILL.md file, that your assistant can load when a task matches it. The Skills page is where you decide which skills your assistant gets.
Only admins can open Skills. It sits under Knowledge in the sidebar. Members who try the address are sent back to their home page. Before the memory build starts, the page shows a preview with "Available once your memory is built" and an admin sees a Get started button. Members see "An admin in your organization needs to build it first."
Where skills come from
Every new skill starts as Pending and does nothing until an admin approves it.
| Source | How it arrives |
|---|---|
| A member's assistant | A member asks their assistant to publish a skill they wrote. This uses the skills_publish tool and only works for knowledge categories that member is allowed to see. |
| An admin | Import skills on this page, or Edit on an existing skill. |
| Brain | Brain can also propose a new skill, or a new version of one, when skill proposals are switched on for the service. Not every organization gets them. These arrive in Pending with a Why this was proposed section. If you never see any, add skills with Import skills or through an assistant. |
If a skill with the same identifier already exists, publishing or importing does not replace it. It adds a new Pending version. The version that is Active stays in use until you approve the new one.
How a skill reaches your assistant
When an assistant connects to Brain, Brain gives it a list of the available skills: name, description, categories, version and an identifier. The assistant loads the full skill only when one looks relevant to the task, using the skills_get tool.
Only Active skills are on that list, and only those whose knowledge access the person asking is allowed to use. A skill that is Pending or Archived is not.
The assistant decides whether to load a skill by reading its description. A skill without a description has no way to be picked. Write the description as a statement of when to use the skill.
The Skills page
The header reads "What the agent can do, how much it's used, and what's pending review." Refresh reloads the list and Import skills opens the import dialog. There are three tabs, each with a count.
| Tab | What's in it |
|---|---|
| Pending | Skills waiting for your review. This tab opens first. |
| Active | Approved skills your assistant can use. |
| Archived | Skills you've taken out of use. |
The table shows:
| Column | What it shows |
|---|---|
| Skill | The name, then the description. If a skill has no description, the row shows No trigger in place of it. |
| Updated | When it last changed. Hidden on narrow windows. |
| Tags | The first tag, and how many more there are. Hidden on narrow windows. |
| Usage | How many times an assistant has called the skill through MCP, all time. It doesn't say when it was last used. An active skill with no calls shows Never used. |
The default order depends on the tab. Pending and Archived show the most recently updated first. Active shows the most used first. Click Skill, Updated or Usage to sort, and click again to flip the direction.
Search by name, tag or author filters the current tab. It also matches the description, version and knowledge access. Press / to jump to it.
Everyone who opens this page is an admin, so there is no member version of the table. Members meet skills only through their assistant.
Shortcuts
Select ? to see the list.
| Key | Action |
|---|---|
/ | Search |
j and k | Next and previous skill |
e | Edit the open skill |
a | Approve the open skill |
Esc | Close the panel |
Links you can share
The address keeps what you are looking at, so you can share or reload it. It stores the tab (status), search (q), sort (sort and dir) and the open skill (skill). A link to a skill opens the right tab for it. If no skill has that identifier, you see No skill called "…". If it exists but is in a status without a tab, for example superseded, you see "… is Superseded and is not listed here."
Empty tabs
| Tab | Message |
|---|---|
| Pending | Nothing waiting for review. New proposals land here as they come in. |
| Active | Approve a pending skill and it becomes part of what the agent can do. |
| Archived | Nothing archived yet. |
A search with no matches says No skills match "…". If the list can't load, you see "Could not load skills." with the reason. Click Refresh.
Review a pending skill
Open the skill
Click a row in Pending. A panel opens on the right with the skill's name, its status and its actions. The arrows at the top move to the previous or next skill in the list, and the panel shows your position, for example "3 of 34".
Read what it does
The panel shows when it was updated, who created it and its version, then a short summary. Open SKILL.md to read the full instructions. On Pending it starts open.
If the skill has no description, the panel warns you with No description. Brain matches a skill's description against the task, so it will never use that skill until it has one. Add one with Edit.
Check the reason
When Brain or a colleague gave a reason, Why this was proposed shows it, together with a summary of what changed. For a new version of an approved skill the heading is Revision drift. If the skill looks like it touches knowledge it hasn't been given access to, the panel says so: "This skill looks like it touches knowledge it has not been given access to." Check Knowledge access before you approve.
Check when it should fire
Some skills carry examples of when they should be used. When it should fire shows the guidance and the expected behavior. Should trigger lists example requests that should load the skill. Near misses lists similar requests that should not. Each list shows up to four examples, and says "No examples yet." when there are none.
Check for duplicates
If active skills share tags or knowledge access with this one, Already covered by lists up to four of them, most overlap first. Click one to open it, and use Back to to return. Check the skill isn't a duplicate before you approve it.
Approve, edit or archive
Select Approve to make it Active. Select Edit to change it first. Select Archive to set it aside. Brain confirms with "[name] is now active." or "[name] is now archived." and moves you to the next skill in the list.
What else the panel shows
| Section | What it shows |
|---|---|
| Knowledge access | The categories the skill may answer for. |
| Tags | Its tags. |
| Description | The stored description, when the summary above it comes from Brain's own note on the skill's purpose. |
| Usage | "Called N times through MCP." or "Never called through MCP." |
| Resources | Counts for Assets, Scripts and Metadata, with the first five of each listed by name. |
| Earlier pending revisions | Shown when a skill has more than one pending version. |
If the same skill has more than one pending version, the table shows the latest one, and Approve approves that one. Earlier pending revisions lists the older ones, which stay pending.
When Approve fails
| Message | What happened |
|---|---|
| This version is no longer waiting for review. Refresh the page to see its current state. | Someone else already approved or changed it. Click Refresh. |
| A newer version of this skill was approved after this one was proposed. Review the latest version instead. | You are looking at a version built on an older approved one. Review the newer version. |
You also see "Could not update skill." followed by the reason for any other failure.
What approving does to the old version
When you approve a new version of a skill that is already Active, the old version is replaced. Assistants get the new one from then on. Other pending versions that were built on the replaced version become Outdated.
Edit a skill
Select Edit in the panel. You can change:
| Field | Notes |
|---|---|
| Name | Required. Up to 240 characters. |
| Version | Optional. Up to 80 characters. If you leave it empty, Brain uses 1.0.0. |
| Description | When Brain should use this skill. Up to 2000 characters. |
| Tags | Comma separated, for example sales, analysis. Brain stores them in lowercase and keeps at most 50. |
| Knowledge access | Who this skill may answer for. Pick from general, marketing, sales, finance, product, engineering, customers, hr and leadership. Ticking a category unticks general. If you clear them all, it goes back to general. |
| SKILL.md | Required. Up to 200000 characters. |
Brain rewrites the header at the top of SKILL.md from the skill's identifier and description each time it saves. Any other header fields in the file are dropped from the stored copy.
How saving works depends on the skill:
- A Pending skill is edited in place.
- An Active skill gets a Pending version with your changes. If it already has a pending version, that one is updated. The Active version stays in use until you approve the new one.
Select Save to keep your changes or Cancel to leave. Brain confirms with "[name] updated." If you cancel with unsaved changes, Brain asks Discard your changes? and you can choose Keep editing or Discard changes. The skill then goes back to how it was when you opened it.
If Name or SKILL.md is empty, Brain says "Give the skill a name." or "A skill needs its SKILL.md content." and doesn't save. Other failures show "Could not save skill." with the reason.
Archive and restore
Select Archive on a pending or active skill to move it to Archived. The whole skill leaves the list your assistant sees, but stays on the page.
There is no Reject button. To set a proposal aside, archive it.
In Archived, the buttons are Restore, Edit and, in one case, Approve:
- Restore brings the skill back to Active if it was approved before, or to Pending if it never was.
- Approve appears on an archived skill only when it was never approved before.
Archived is where skills wait, not where they are deleted. The page has no delete button.
Import skills
Select Import skills at the top of the page to add skills from files.
Choose your files
Drop Markdown files on the dialog or pick them. Only .md and .markdown files are accepted, and the whole file becomes the skill content. You can add several at once, and drop more later. If a file can't be read, you see "Could not read [file]."
Check what Brain read
Brain reads the file's header if it has one, then fills in each row.
| Field | Where it comes from |
|---|---|
| Name | The first # heading in the file. If there is none, the name in the header, then the file or folder name. |
| When to use it | The description in the header. If there is none, the box is empty and you must fill it in. |
| Tags | The tags in the header, as a list or comma separated. |
| Version | The version in the header. |
| Knowledge access | Never read from the file. Every row starts as general. |
A file called SKILL.md takes its name from the folder it sits in, when the folder comes with the drop. Without one it would be named SKILL, so check the name field.
Complete each row
Each file gets a name, and a box that asks "When should Brain use this skill?". Fill it in, because a skill without it will never be picked. Rows with a problem are marked, and the dialog says "Fix the marked rows before importing." Use the X on a row to remove it.
Set knowledge access
Change it per file with the categories button, or use Set knowledge access for all when you have more than one file, then Apply to every file. A row left empty goes to general. Tags in a file never change access.
Import
Select Import. Brain sends the files one at a time, and one failure doesn't stop the rest. When all files go in, Brain says "1 skill imported. It is waiting for approval." or "N skills imported. They are waiting for approval." and takes you to Pending.
Limits and row messages
| Field | Limit | Message |
|---|---|---|
| Name | 240 characters | "Name is N characters; the limit is 240." |
| Name | Required | "No name could be derived. Enter one." |
| When to use it | 2000 characters | "Description is N characters; the limit is 2000." |
| When to use it | Required | "Say when Brain should use this skill, or it will never be picked." |
| File | 200,000 characters | "File is N characters; the limit is 200,000." |
| File | Not empty | "The file is empty." |
| Version | 80 characters | "Version is longer than 80 characters." |
| File name | Needs letters or numbers | "Nothing here names the skill. Give the file a name with letters or numbers." |
| Two files, same skill | One per batch | "Same skill as [other file]. Remove one of the two." |
The missing-description message appears only after you press Import. Until then the box just shows its question.
Outcomes
After you press Import, each row shows what happened.
| Outcome | Meaning | What to do |
|---|---|---|
| Imported | The skill is in Pending. | Review it there. Imported rows leave the dialog. |
| Already exists | Brain refused the write and says why. This happens, for example, while your organization is being reset and nothing can be written. | Wait, then press Import again. |
| Rejected | The content is invalid. The row shows the reason Brain gives. | Fix the row and import again. |
| Failed | Something else went wrong. | Import again. If it keeps failing, contact support. |
Files that didn't go in stay in the dialog so you can fix them and press Import again. The footer then reads "N skills already imported and waiting for approval. M left." and Cancel becomes Done. If every file was refused because the organization is being reset, a notice says nothing can be written right now, your files are still there, and you should try again in a few minutes.
Brain identifies a skill by the name in the file's header, written as lowercase words joined by hyphens, or by the file name if there is no header name. The # heading does not affect identity. A file with the same identifier as an existing skill doesn't fail. It becomes a new Pending version of that skill.
Imported skills are never active straight away. Every one lands in Pending and needs your approval.
Related
- MCP tools reference: the
skills_getandskills_publishtools. - Ask questions: how the assistant uses what Brain returns.
- Roles: what admins and members can do.