Mio — Personal Context for AI Agents

Mio gives you access to the user's personal context: their tech stack, coding conventions, communication style, project details, and preferences. Instead of asking the user to repeat themselves, check Mio first.

Session Start

At the beginning of every conversation, load the user's context before doing anything else.

For code tasks (writing, reviewing, debugging, refactoring):

list_categories()
get_context({ query: "<describe what you're working on>", sources: ["tech_stack", "conventions"] })

For general or broad tasks (planning, brainstorming, communication):

get_context({ query: "<describe what you're working on>" })

Omitting sources returns all available categories — use this when you're unsure what you need.

If working on a specific project:

list_projects()
set_active_project({ project: "<project-uuid>" })
get_context({ query: "<describe what you're working on>" })

Setting the active project means all subsequent get_context calls automatically include that project's context. Do this once, early in the session.

During Work

Use the context, don't parrot it. The point of loading context is to inform your decisions silently — not to quote it back to the user. If Mio says the user prefers TypeScript with arrow functions, just write code that way. Don't say "According to your Mio profile..."

Don't re-ask things Mio already knows. Before asking the user about their preferred language, framework, style, or conventions, check if Mio already has this information. Only ask when Mio doesn't have an answer.

Refresh context when switching topics. If the conversation shifts to a different project or domain, call get_context again with updated sources relevant to the new topic.

End of Session

When a conversation has been substantive — you made decisions together, debugged something complex, or learned something new — log it:

log_conversation({
  summary: "Migrated auth from JWT to session cookies. Chose httpOnly + Secure flags for XSS protection.",
  takeaways: ["Prefers session-based auth over JWT", "httpOnly cookies are a hard requirement"],
  tags: ["authentication", "security"]
})

What makes a conversation worth logging:

• Architectural decisions were made
• New user preferences were expressed
• A complex problem was solved
• The user taught you something about their workflow

Don't log:

• Trivial one-off questions ("what time is it in UTC?")
• Simple lookups or file reads with no decision-making
• Conversations that produced no new information

Profile Updates

When the user explicitly states a new or changed preference, suggest a profile update:

update_profile({
  category: "tech_stack",
  suggestion: "Uses Bun as package manager and runtime (switched from npm)",
  reason: "User explicitly said they switched from npm to Bun"
})

Only suggest updates when:

• The user clearly states a preference ("I use Bun now", "We switched to Tailwind")
• You learn a stable fact about their setup ("Our team does 2-week sprints")
• A convention is confirmed ("We always use conventional commits")

Don't suggest updates for:

• One-off decisions ("Let's use a Map here instead of an object")
• Assumptions or inferences you're not confident about
• Temporary context that won't apply to future sessions

All suggestions go to the user's review queue — they approve or reject each one. Never assume a suggestion was accepted.

Graceful Degradation

Not all Mio tools may be available. Work with what you have:

• Only get_context available? Load context at session start, skip logging and profile updates.
• No list_projects? Skip project setup, use get_context without project scoping.
• No log_conversation? Focus on using context, skip end-of-session logging.
• No tools at all? This skill doesn't apply — proceed normally.

Tool Reference

For detailed parameters, response formats, and advanced patterns for each tool, see references/tool-patterns.md.