LM Studio Skills in Bionic: Use, Create, Install

Bionic reads agent skills in the standard SKILL.md format

Bionic reads skills written in Agent Skills, the industry-standard format built around a SKILL.md file, so a skill you wrote for another agent loads without conversion. A skill is a shortcut that holds one piece of know-how: specialist knowledge, a capability, or a task you do again and again. Under the specification, each skill is a folder with at least a SKILL.md inside, and that file starts with YAML frontmatter and continues in Markdown.

Bionic is a separate app from LM Studio, released on July 16, 2026 as an agent for open models, and it gained skills on August 17, 2026. It runs local models on the LM Studio runtime, or open models on LM Studio Secure Cloud when a job is too heavy for your own machine. LM Studio itself stays available alongside it for low-level configuration.

Everything you do with skills in Bionic falls into three actions. The table shows where each one starts.

Type @ to use a skill, and some models will pick one on their own

To use a skill, type @ in the composer, the box where you write your message, and choose the skill from the palette that opens. The same palette lists files too, so one keystroke covers both kinds of reference. Once you reference a skill, Bionic reads its SKILL.md and works out how to act on it.

Bionic can also reach for a skill without being asked, depending on the model you run. LM Studio does not say which models qualify, so treat automatic use as a bonus and keep @ as the dependable route.

LM Studio's Docs advise using @ whenever you already know which skill Bionic should apply. From there, a simple habit follows: if a reply ignores a skill you expected it to use, reference that skill with @ and send the request again.

Ask Bionic to create or install a skill instead of writing the file

You do not have to write a SKILL.md by hand, because Bionic will create a skill or install one when you ask. For creation, LM Studio's post simply says to ask, and for installation it calls asking the agent the best way. Placing the file yourself is still an option when you already have a finished SKILL.md.

Create one from a task you just finished with the agent

The easiest skill to create is one drawn from a task you and Bionic have just completed. When a job went well and you expect to repeat it, tell Bionic to turn what the two of you just did into a skill. It asks clarifying questions to pin down the details, writes the file, and from then on you call it with @ like any other.

The second pattern starts from a blank page. You workshop a new skill with Bionic, describing the job while it drafts. For either route, the Docs suggest including examples, the outputs you require, and any constraints, and they advise reviewing a newly created skill before you rely on it for important work.

Install from a URL, or drop a SKILL.md file into Settings

To install a skill, give Bionic its URL and ask it to install it, or run @Install Skill. LM Studio describes asking the agent as the best way to install, just as it does for creating. If a SKILL.md is already on your machine, the alternative is to drop it into Settings > Skills, which makes it available to the agent.

Read third-party skills before you install them in Bionic

A skill is a set of instructions the agent acts on, so read a third-party SKILL.md before you install it. Running the model locally does not remove that step, and LM Studio's own tips say to review installed skills before relying on them for important work. Open the file, see what it asks the agent to do, and only then hand over the URL.

Docker has described what goes wrong when one skills store is shared between sandboxes. In Docker Sandboxes, supported agents mount the same host-side store with read-write access unless you opt out, so a skill changed inside one sandbox becomes input for the next one that loads it.

Docker also marks the limit of that risk: the store is sandbox state, an edited skill does not run on your host by itself, and the exposure moves between sandboxes rather than into the host. A related attack, where a path you already trusted is quietly redefined, is covered in Cursor vulnerability ran code you never approved.

Some skills are published as web pages rather than as a file you can download. Converting the page to Markdown gives you the instructions as plain text, which is easier to read line by line before you install anything.

Turn on the Claude Code and Codex skills you already have

Skills you already keep for Codex or Claude Code work in Bionic once you allow them under Settings > Skills > Skills from other apps. That screen is where you choose which of those skills Bionic can see, and anything you want it to ignore can be switched off from the ••• menu. The Docs label the same setting "Use skills found in other apps", so look for either wording.

LM Studio's Docs call these compatible skills, and the choice is yours rather than all or nothing. Start with the ones you call most often, then add more once you have seen how Bionic handles them. When you later want Bionic to see a skill you left out, go back to the same screen and make it visible.

Keep one library of skills for Claude Code, Codex, and LM Studio

LM Studio skills follow the same standard as the skills you use in Claude Code and Codex, so one library can serve all three. The format is SKILL.md, calling a skill takes @, creating or installing one is a request to the agent, and reusing skills from other apps is a switch in Settings.

For a first session, work in this order. Turn on the skills you already have, use Bionic normally until you notice a task you keep repeating, and then ask Bionic to turn that task into a skill. Before installing anything from outside, read its SKILL.md.

LM Studio's Docs name the best fit as workflows with repeatable steps, a consistent output format, or specialist instructions. Their examples include reviewing documents against a checklist, applying a team style guide, and following a recurring research or analysis workflow.