Superpowers¶

Superpowers are like shortcuts for common coding tasks. Instead of figuring out how to approach a bug fix or a new feature from scratch every time, invoke a skill and let it guide your agent through a proven workflow.

Think of them as "recipes" — each one knows the best way to handle a specific type of task.

The Skills¶

Starting New Work¶

Before you write any code. This skill helps you explore the problem, ask the right questions, and choose the best approach before committing to an implementation.

"Use brainstorming to figure out how to build this feature"

Turn a vague idea into a step-by-step plan. Break down requirements, identify dependencies, and create a clear roadmap before coding starts.

"Use writing-plans to break this requirement into steps"

Fixing Bugs & Ensuring Quality¶

Fix bugs with evidence, not guesswork. Gather symptoms, form a hypothesis, test it, and verify the fix actually works.

"I found a bug — use systematic-debugging"

Never say "done" without proof. Run the tests, check the output, and confirm everything actually works before claiming success.

"Use verification-before-completion before finishing"

Working At Scale¶

Do multiple things at once. Split independent tasks across parallel agents, aggregate results, and handle failures gracefully.

"Use dispatching-parallel-agents to handle these in parallel"

Monitor your agent team. View real-time status, track outcomes, and find improvement opportunities.

"Check the team-ops monitoring dashboard"

Security¶

Check for security issues. Scan your configuration for exposed secrets, unsafe settings, and potential vulnerabilities.

"Run security-scan on the config"

How To Use Skills¶

You don't need to memorize skill names. Just describe what you want to do, and your agent will pick the right skill:

You: "I need to debug this error — the login page crashes when I click submit"
Agent: *activates systematic-debugging skill*
       1. Gather error symptoms...
       2. Form hypothesis...
       3. Test fix...
       4. Verify it works

You can also explicitly invoke a skill by name:

"Use brainstorming for this feature"
"Use writing-plans to plan the migration"
"Use systematic-debugging for this error"
"Run verification-before-completion before we commit"

Which Skill Should I Use?¶

Situation	Skill
"I have an idea but don't know where to start"	`brainstorming`
"I know what to build but need a plan"	`writing-plans`
"Something is broken"	`systematic-debugging`
"About to finish — make sure it actually works"	`verification-before-completion`
"Multiple independent things to do"	`dispatching-parallel-agents`
"Need to check security"	`security-scan`

RL Training System¶

AIOS includes a multi-environment reinforcement learning system. It trains a shared student policy using a unified control plane across shell, browser, and orchestrator tasks.

For details, see the Architecture page.

Where To Go Next¶

Case Library — real-world examples of skills in action
ContextDB — how skills interact with project memory
Agent Team — multi-agent collaboration details