Rules & Skills
With rules you permanently specify how the AI should work – such as language, style or prohibitions. With skills you create larger capabilities that only take effect when they are needed. You'll find both in Apis AI under Settings, then Extensions, in the Rules and Skills cards.
Rules – Persistent Instructions
Rules are short instructions that simply always apply in your chats – for example "Answer in English", "Briefly explain changes before you make them" or "Never touch files without mentioning it". You write them once and don't have to explain them again in every chat.
When creating a rule you choose one of two variants:
- Always active – the rule applies in every chat.
- Only for matching files – the rule applies only when you are working on a matching file. In the File globs field you enter patterns, e.g.
**/*.javafor everything ending in.java.
A proven combination: general style and language rules permanently active, project-specific instructions only for matching files.
Project Instructions (AGENTS.md)
Many projects carry their own ground rules in a file called AGENTS.md – for example how the project is built, which style is expected or which folders must not be touched. Apis AI looks for that file on its own as soon as the AI works in one of your folders, and follows the instructions it finds – you don't have to attach the file or remind the AI of it.
The search starts in the folder you are working in and goes upward through all parent folders – but never beyond the folders you have shared with the AI. If several such files are found, the more precise ones count: instructions from a subfolder are read after those from the parent folder.
- A file in the project folder therefore applies to everything inside it.
- Only reasonably small files are used (up to 64 KB); unreadable or very large files are skipped.
- Each file is read once per chat – changes to it take effect in the next chat.
So if a rule belongs to the project rather than to you personally, put it into an AGENTS.md in the project folder. For things that should apply to you in every chat, rules remain the better choice.
Skills – Reusing Capabilities
A skill is a package of instructions that the AI draws on when needed – for example "Frontend design", "Writing SEO copy" or your team's conventions. Once created, it is available to you in every chat.
When creating one you fill in:
| Field | Meaning |
|---|---|
| Name | This is what the skill is called in your list. |
| Description (when to use?) | Briefly, what the skill is good for – this is how the AI recognizes when it fits. |
| Instructions | The actual content: what should the AI do? |
Only if the skill should trigger automatically do two more fields appear:
- File globs – patterns for file types, e.g.
**/*.cssor**/*.java. - Keywords – terms such as "landing page, hero, call to action".
How a Skill Becomes Active
At the top of the skill is the Active switch: with it you turn it completely off or on. In addition, under Activation you choose how it starts – exactly one option:
| Option | What happens |
|---|---|
| Always active | The skill applies in every chat. Good for short, permanently valid instructions – but it takes up space in every conversation. |
| Auto-trigger | The skill takes effect on its own as soon as a matching file or one of your keywords appears. Ideal when it is only sometimes relevant. |
| As a tool | The AI fetches the skill itself when it needs it. You don't have to do anything else. |
| Manually per chat | The skill is available for selection in chat and only becomes active when you enable it there. |
Adopting Ready-Made Skills
You don't have to write skills yourself. Via the library you fetch ready-made packages and adopt them with one click:
- Curated – selected skills for direct addition.
- Online (mcp.directory) – browse and search a public skill directory.
- GitHub repo – enter
owner/repo(e.g.anthropics/skills), load the repository and adopt individual matches from the list. - Import – paste a link (raw file, GitHub or mcp.directory link, e.g.
https://mcp.directory/skills/frontend-design). The skill is loaded and appears as Add or Added.
Adopted skills land in your list and can be adjusted or deleted there at any time.
Built-in Agents
Apis AI already brings along ready-made agents – for example the Kotlin Expert: a senior Kotlin engineer for code reviews, debugging and idiomatic code (JVM, Android, Kotlin Multiplatform, coroutines and Flow). Such an agent is simply a skill with a complete expert instruction.
You don't have to install anything for this: on the first start the agent is placed in your skills and appears there with the activation Manually per chat. From then on it is completely yours – you can rename it, adjust its instructions, switch it off or delete it. A deleted agent does not come back on the next start either.
Enabling Skills in Chat
Skills with the Manually per chat activation can be found in chat under the "+" menu, then Skills. There you enable them specifically for the current conversation without changing your other settings.
If you open this view and haven't created such a skill yet, the app points out to you that you can create a skill with "Manually per chat" under Settings, then Extensions.