An AI agent can help with a task, but it needs to understand how your team works: what context to consult, which rules to follow and what outcome to prepare. Skills package that knowledge as reusable instructions and procedures an agent can follow when requested.
Prefapp applies this approach in Firestartr. A skill can interpret a plain-language request, gather the organisation’s context, check relevant rules and prepare a change for review. For example, Firestartr’s operational skill can propose platform changes and bring them to a pull request for the team to review before they are applied.
A skill’s instructions are not, by themselves, a security boundary. We combine them with least-privilege technical permissions, automated validation and human approval points. This lets AI accelerate work without granting unlimited authority or turning a suggestion into an uncontrolled production change.
What skills provide
Explicit context
They specify which information about repositories, teams and systems the agent needs before acting.
Repeatable procedures
They turn team practices into reusable steps instead of relying on improvised prompts.
Rules and permissions
Instructions guide the agent; technical permissions and policies limit what it can execute.
Human review
Changes are presented as reviewable proposals, for example through pull requests, before they are applied.
A workflow you can review
Firestartr’s website shows this pattern for operating a platform: the agent interprets a request, places it in the organisation’s context, validates the rules and prepares a change for review before deployment. Prefapp’s public skills include both development workflows and the Firestartr operational skill.