← Back to Blog
Product

Skills in Aii Cloud — Bring Your Workflow, Keep the Boundaries

·3 min read
productcloudskillsgithubpluginsworkflowsv0.6

The best AI workflow is rarely “be helpful.” It is the way your team already works: the review checklist, research method, decision format, or repository briefing procedure that makes an answer dependable.

Aii Cloud now has Skills for that layer.

A Skill is a SKILL.md instruction package you can inspect, install, and deliberately apply to a Chat or Work turn. It changes the method—not the product's permission model.

Start with a built-in workflow or make your own

Open Skills from the dashboard. You can:

Personal Skills remain private to your account. You can inspect, edit, download, update, and remove them from the same library.

From the composer, choose an installed Skill or type / at the start of the draft and select it from the Skills group. A visible token stays linked to the selected workflow until you send or remove it.

One Skill, one turn

Selecting a Skill arms it for the next accepted message. It is consumed after that turn and does not silently stay active for the rest of the chat.

Why so explicit? Because procedure is context. A decision framework that helps one request may distort the next. One-turn activation keeps the applied method visible and lets the current request control length, audience, and output format.

Afterward, the owner can see which exact Skill revision was applied. Continue with Skill deliberately re-arms it for a follow-up; an ordinary reply remains ordinary.

Review the exact GitHub snapshot

Installing from GitHub is a review flow, not a paste-and-trust shortcut.

Aii resolves the source to a full commit and shows the canonical repository, path, complete SKILL.md, content digest, supported-directory result, and license-file status. Only after confirmation does it install the exact bytes you reviewed.

Imported packages are intentionally narrow: one bounded instruction file plus ordinary license files. Scripts, symlinks, submodules, resource folders, and executable bundles are rejected. A mutable branch moving later does not change what is installed; updates require another explicit review.

The trust copy says what the system actually knows: instructions only, not reviewed by Aii, no tools or account access added.

Skills and Plugins are different kinds of authority

A Skill tells the model how to approach the work. A Plugin gives a turn access to an external capability.

The built-in Repository brief Skill makes that distinction visible. Selecting it does not grant GitHub access. To read a repository, you must also select the ready GitHub Plugin as a separate one-turn capability.

The Gateway checks the current installation and grant before sending a short-lived credential to the worker. Repository contents are treated as untrusted evidence, not as new instructions. A GitHub-enabled conversation stays inside the product's private-context boundary and cannot be publicly shared or mixed with public web search.

Instructions cannot manufacture authority. A Skill cannot:

A workflow you can read is a workflow you can improve

The point of Skills is not to collect more magic prompts. It is to make repeatable methods portable and inspectable.

Create a Skill for the work you do every week:

markdown
---
name: launch-readiness-review
description: Review a launch plan for evidence, ownership, rollback, and open risk.
---

Use the supplied launch plan as evidence.
Separate confirmed facts from assumptions.
Check owner, deadline, rollback, observability, and unresolved risk.
End with Blockers, Follow-ups, and a clear ship/no-ship recommendation.
Never invent dates, test results, or approvals.

Install it, arm it for one Work turn, and keep refining the source as the team's method improves.


Skills are available in Aii Cloud Chat and Work. They add reviewed instructions; Plugins remain the separate boundary for tools and connected accounts.