AI skills
Teach your agents once. Share it with all of them.
A skill is knowledge plus a trigger. Import one or write your own. Assign it to the agents that need it, and it works on every surface.
Only the index sits in the prompt. The agent loads a skill when its task matches the trigger.
The idea
Knowledge plus a trigger
Most organisations solve this by pasting the same instructions into every prompt, or by writing a long system prompt that nobody maintains. Both approaches drift, and both cost tokens on every request whether the knowledge is relevant or not.
A skill is a piece of knowledge with a plain-language description of when it applies. Write the trigger the way you would explain it to a colleague — “when the user asks for a code review” — and the agent works out the rest.
What the agent needs to know. A checklist, a house style, a runbook, a set of controls.
When it applies, in plain language. The agent matches its current task against it.
How you organise them. Filter by tag or by collection namespace as the library grows.
Import
Bring in work you did not write
Import a skill from the skills.sh community registry or from a public GitHub repository. Pin a GitHub skill to a tag or branch so an upstream change cannot alter how your agents behave without your say-so.
- Import from skills.sh or a GitHub repository
Pin GitHub skills to a tag or branch, and re-sync when you choose
- Import a whole collection and sync it as a unit
- Mix imported and inline skills in one library
Only the index sits in the prompt. The agent loads a skill when its task matches the trigger.
Loading
You pay for the skills you use
Skills load on demand. Only a compact index of names, descriptions, and triggers sits in the agent’s system prompt. When the current task matches a trigger, the agent loads that skill’s full content and continues.
That keeps the baseline prompt near 3,000 tokens rather than the roughly 49,000 it would take to inline every skill. A large library costs you almost nothing until it is needed.
Built in
Skills you did not have to write
On QuantGov, every organisation starts with 19 built-in skills. They cover ISM topics, four Essential Eight strategies, WCAG 2.2 AA accessibility, the Australian Privacy Principles, and Quant platform reference. Your agents know the compliance context before you configure anything.
Access control, logging, cryptography, data handling, networks, and secure development.
Application control, hardening, MFA, and patching.
Guidance across all four WCAG principles.
Australian Privacy Principles as they apply to handling data.
Disable any built-in skill for a platform agent through an overlay. Disabled skills are removed from the index and blocked at execution, so an agent cannot reach for one you have turned off.
From nothing to a shared library
Three steps, and no format to learn.
Import or write
Pull a skill from skills.sh or GitHub, or write one with content and a trigger condition. Both live in the same library.
skills.sh/s/code-reviewAssign to agents
Skills become available to the agents you choose. One import, and everyone on the team gets the benefit.
available to 6 agentsSync on your terms
Pin a GitHub skill to a tag or branch. It does not move until you change the pin and re-sync.
ref: v1.4.2What teams put in their library
The pattern is always the same: knowledge one person had, now everyone's agents have.
House style and standards
Your coding conventions, your writing style, your review checklist. Every agent applies them without being told each time.
- Consistent output
- No prompt copying
- One place to update
Runbooks and procedures
Deployment steps, incident response, escalation paths. The agent loads the runbook when the situation matches.
- Triggered by context
- Version controlled
- Shared across the team
Compliance context
Controls and obligations your work has to satisfy. Built-in skills cover the common frameworks; add your own for the rest.
- Built in
- Add your own
- Disable per platform agent
Build your skill library
Import what exists, write what does not, and give it to the agents you run.