Using Skills

Enabling a skill

Open your agent in the Console
Go to the Skills tab
Browse or search the skill library
Click Enable on the skill(s) you want
Save

An agent can have multiple skills enabled. Skills provide step-by-step guidance the agent follows when you ask it to perform that type of work.

You can also manage an agent’s skills without leaving a chat: open the Skills section in the workspace side panel to see the agent’s skills, turn them on or off instantly, add one from the library, or install one from a link. Changes apply to your next message.

Turning a skill off means the agent can’t fall back on it either — a disabled skill is unavailable, not merely unadvertised.

When does the agent use a skill?

Skills are attached to your agent at configuration time. The agent uses a skill when:

You run it directly from the message box (below)
You ask for it in words — “Using the code review skill, review the changes in PR #45”
The skill’s instructions are relevant to the task and the agent chooses to apply that workflow as part of its reasoning

Each skill has an Invocation setting that decides which of those are allowed:

Invocation	Who can start it
Automatic	You or the agent — the default
On request only	Only you, by running it
Agent only	Only the agent, when it judges the skill relevant

Running a skill on demand

Type / at the start of the message box to open the skill menu. It lists the skills attached to this agent, one line each with its description. Pick one and it becomes a removable Skill chip in front of your message — whatever you type after it is passed to the skill as the instruction for this run.

One skill per message. The chip applies to that message only.
If the agent has no skills attached, the menu tells you so and points you at the Skills panel.
Guests and customer-facing conversations don’t get the menu.

Seeing which skill the agent is following

While a skill is in effect, a Following skill pill sits in the run header, naming it and saying whether it was started by you or by the agent. The activity rows underneath name it too — Using skill: weekly-report, Reading skill file: extract.py — so a skill-driven run is never a black box.

If a long conversation gets condensed, the skill is re-attached afterwards and a one-line note says so, so the agent doesn’t silently drift off the procedure halfway through.

What a skill is allowed to do

A skill can carry a list of the actions it expects to use. That list only ever narrows what the agent can do — it can never widen it.

Anything the skill doesn’t list, and that writes, sends, deletes or spends, asks you first, even if your autonomy level would normally let it through.
Anything the skill explicitly rules out is refused.
Reading is never restricted this way — a skill can’t blind the agent.
A skill can never grant a permission the agent doesn’t already have, and never lowers the safety floor for payments and credentials.

This means installing a skill from a link is a safe thing to do: at worst it asks you more often than you’re used to.

Skills that bring files

Some skills ship with scripts or reference documents. When the agent first needs one, the files are placed in the agent’s sandbox under /workspace/.skills/<skill-name>/ and the agent is told where they are.

Nothing in a skill runs by itself. A script that arrives with a skill is just a file until the agent decides to run it, and when it does, that’s an ordinary command subject to your autonomy level and approval rules. No part of a skill executes before you or the agent has seen its instructions.

Skills that are only instructions never touch the sandbox at all.

Which skills are actually earning their place

The agent’s Skills tab shows, per skill, when it was last used on this agent and how many times. Skills that have never been used carry a Never used on this agent badge.

That matters for more than tidiness: every attached skill takes up a little of the agent’s attention on every run. If a skill is never used, detaching it makes the agent slightly sharper.

Creating a custom skill

Go to Console → Skills
Click New Skill
Choose a category and give it a name
Write the instructions — the step-by-step procedures the agent should follow
Save and publish
Enable it on your agent from the Skills tab

Custom skills use the same instruction format as the built-in ones. Write them like a standard operating procedure:

When the user asks to [trigger]:

1. [Step 1 — what to do first]
2. [Step 2 — what to check or gather]
3. [Step 3 — how to proceed]
4. [Output format — what the final response should look like]

A skill can also be set to run as a helper agent instead of as instructions your agent follows itself — useful for self-contained jobs. See Helpers.

Adding a skill from a link

If someone shares a skill as a link — a GitHub repository or folder, a skills.sh page, or a direct SKILL.md link — you can install it without copying anything by hand. Auteryn uses the open Agent Skills (SKILL.md) format, so skills shared anywhere in that format work here.

Go to Console → Skills and click Add from link
Paste the link and click Preview
Review what would be installed — the skill’s name, description, full instructions, any bundled scripts, where it comes from, and a safety review
Click Install

You can also add a skill from a link inside an agent’s Skills tab (it’s attached to that agent on install), or by pasting the link to your agent in chat and asking it to install the skill — the agent shows the same details on a confirmation card and installs only after you approve.

A few things to know:

Nothing is added until you approve it. Existing skills are never replaced — if the name is taken you’re asked to pick a new one.
Skills that bundle runnable scripts are clearly flagged; review them before approving.
The preview lists anything in the file Auteryn doesn’t support, by name, so you know exactly what was ignored rather than assuming it took effect.
Imported skills show an Imported badge and a source panel with a Check for updates button. Updates are shown as a summary of what changed and only apply when an owner or admin confirms.

Disabling a skill

Open the agent → Skills tab
Find the skill → Disable