michi-explore

The explore skill is a structured conversation for investigative work. It applies Michi discipline — surfacing assumptions, noticing drift, capturing decisions — to thinking together. Use it when you need to understand something before you can plan it, or when the work is research rather than implementation.

When to use

• Orienting in a new session, project, or unfamiliar area of a codebase
• Researching tools, evaluating options, surveying libraries
• Brainstorming ideas and exploring directions before committing to one
• Pre-planning work when you don’t yet know enough to write a milestone plan
• Investigation that runs alongside an active epic — supporting research, spikes, library evaluations

What it does

The skill has two modes: peek and open exploration.

Peek mode (/michi-explore peek) is orient-first. The agent reads project docs at the requested depth (S through XL), summarizes what it understands, and waits. It builds shared context before proposing action. You can point it at a specific file, folder, or topic to focus the orientation.

Open exploration follows the thread wherever it goes. The agent alternates between dialogue (brainstorming, asking questions, thinking together) and investigation (reading files, analyzing code, building comparisons). Throughout, it maintains three disciplines: surfacing assumptions explicitly (“I’m assuming X because Y”), noticing when the conversation drifts from its starting point and making that visible, and capturing decisions when choice points emerge.

The agent tracks two kinds of questions as they arise — prerequisite questions (need an answer to continue) and discovery questions (emerged during exploration, often the most valuable output).

What it produces

• An artifact written to an agreed location — typically docs/epics/<topic>.md (flat-file epic, grows to a subdirectory if it earns one). The format follows what emerged from the conversation: an evaluation with options and a recommendation, research findings by theme, an exploration summary, or brainstorm results with initial assessments.
• A mini-reflection covering the key insight, decisions to capture, anything worth adding to memory, and what comes next.
• STATUS.md update if the exploration changes what’s active or understood about the project.

Key things to know

• This skill is flexible by design — a posture, not a checklist. It adapts to the conversation rather than imposing structure.
• Peek mode is useful for getting oriented at the start of a session without the agent immediately proposing work.
• The agent will name drift when it happens but won’t force the conversation back. The drift might be the discovery — you decide whether to follow it or return.
• After exploring, the natural next step is /michi-planning if a direction is clear, another /michi-explore if there’s more to investigate, or /michi-workshop if the work is small and self-contained.

For the full agent instructions, see the SKILL.md source.