ora

Use this agent to explore and understand codebases or search local project files. Powered by semantic search (codebase-search) — finds answers across code, config, docs, and any text files in the project. Preferred over the built-in Explore agent for any codebase exploration task. Examples: <example> Context: User wants to understand how a feature works user: "How does authentication work in this project?" assistant: "I'll use the ariadne agent to trace the auth flow across the codebase." <commentary> User needs to understand a cross-cutting concern — ariadne explores multiple files and connections. </commentary> </example> <example> Context: User needs to find where something is implemented user: "Find all the API endpoints and how they connect to the database" assistant: "I'll use the ariadne agent to map out the API layer." <commentary> Broad codebase exploration that requires searching patterns, reading files, and following references. </commentary> </example> <example> Context: User is onboarding to unfamiliar code user: "Give me an overview of this project's architecture" assistant: "I'll use the ariadne agent to explore the project structure and key components." <commentary> Architecture overview requires systematic exploration of directories, entry points, and dependencies. </commentary> </example>

Use this agent to orchestrate multi-step plan execution with learning accumulation between tasks. Examples: <example> Context: A validated plan with multiple implementation tasks user: [Plan approved with 6 tasks, some parallelizable] assistant: "I'll use the atlas agent to organize these tasks into execution waves with learning carry-forward." <commentary> Multi-step plan benefits from wave-based dispatch — atlas groups tasks by dependencies, specifies which agent handles each, and defines what learnings to capture between waves. </commentary> </example> <example> Context: First wave completed, need to plan next wave with learnings user: [Wave 1 results: auth middleware done, discovered project uses custom error classes] assistant: "I'll re-invoke atlas with the wave 1 learnings to plan the next wave." <commentary> Between-wave re-invocation — atlas receives accumulated learnings, compresses them, and adapts the next wave's task prompts to include relevant discoveries. </commentary> </example>

Use this agent to research external sources — documentation, websites, GitHub repositories, and any information available on the internet. Examples: <example> Context: User needs docs for a library or API user: "Find the docs for React Query's useInfiniteQuery" assistant: "I'll use the clio agent to look up the documentation." <commentary> User needs external documentation for a specific API — clio fetches and summarizes it. </commentary> </example> <example> Context: User needs to make a technical decision user: "Should I use Zustand or Jotai for this project? What are the trade-offs?" assistant: "I'll use the clio agent to research both libraries and compare." <commentary> Technical decisions require up-to-date comparison of docs, community patterns, and real-world usage. </commentary> </example> <example> Context: User wants to look up information from a website or research a topic online user: "What does the Tailwind v4 migration page say about breaking changes?" assistant: "I'll use the clio agent to look up that information." <commentary> Web research — fetching and summarizing content from websites or general internet sources. </commentary> </example> <example> Context: User asks about code or patterns in a specific GitHub repository user: "How does the Zustand repo implement its middleware system?" assistant: "I'll use the clio agent to explore the Zustand repository." <commentary> GitHub repo exploration — searching and understanding code in public repositories without cloning. </commentary> </example>

Use this agent for autonomous deep work — isolated implementation tasks that run independently in worktrees. Examples: <example> Context: User needs a module refactored while continuing other work user: "Refactor the payment module to use the new Stripe API" assistant: "I'll use the hephaestus agent in a worktree to handle this refactor autonomously." <commentary> Self-contained refactoring task — hephaestus works in an isolated worktree, freeing the main conversation for other work. </commentary> </example> <example> Context: Atlas dispatches a task from a multi-step plan user: [Atlas wave dispatch specifies: implement auth middleware] assistant: "I'll spawn hephaestus in a worktree with the task context and learnings from prior waves." <commentary> Plan execution task — hephaestus receives goal + accumulated learnings, works autonomously, returns changes + new learnings. </commentary> </example> <example> Context: User wants parallel implementation of independent features user: "Add the search endpoint and the notification service — they're independent" assistant: "I'll spawn two hephaestus agents in separate worktrees to work on both in parallel." <commentary> Independent tasks benefit from parallel worktree execution — each hephaestus instance works without interference. </commentary> </example>

Use this agent to analyze requests BEFORE planning to prevent AI failures. Examples: <example> Context: User wants to upgrade SDK/framework version user: "Upgrade from Expo SDK 53 to 54" assistant: "I'll use the metis agent to analyze scope and risks before planning." <commentary> SDK upgrade touches multiple packages, has breaking changes, and needs migration research — metis classifies intent, surfaces hidden requirements, and produces directives. </commentary> </example> <example> Context: User wants to build a feature touching multiple modules user: "Add push notification consent flow with analytics tracking" assistant: "I'll use the metis agent to analyze this request before planning." <commentary> Multi-system feature (notifications + consent + analytics) with ambiguous scope — metis explores existing patterns and flags risks before planning begins. </commentary> </example> <example> Context: User wants to refactor a system user: "Refactor the payment module to use the new API" assistant: "I'll use the metis agent to assess scope and risks first." <commentary> Refactoring request — metis identifies what behavior must be preserved, flags regression risks, and prepares directives for the planner. </commentary> </example> <example> Context: User gives an ambiguous request user: "Make the app faster" assistant: "I'll use the metis agent to clarify intent and scope." <commentary> Vague request that could go many directions — metis forces intent classification and generates clarifying questions before any work begins. </commentary> </example>

3 hooks across 2 events