Explore

Output language: answer and write artifacts in the user's own language (match their messages, not the repo's history).

Overview

Turn a task's code references into a fact checklist the main agent can act on without re-reading the repo.

Core principle: the repo owns facts; the user owns decisions. Read everything the task names; verify everything the user asserts; deliver a checklist, not an opinion.

Who reads

The main agent dispatches a background explore subagent and tells it to load this skill. The subagent is read-only: no design, no questions to the user, no code. If you are that subagent, this skill is your entire task — follow it exactly.

When to Use

• Task names code to change or reference: "refactor X", "model after Y", "reuse Z's pattern"
• The prompt contains technical assertions to verify (signatures, fields, SQL, types, "try not to update" style constraints)
• Refactor / legacy rework / reuse, where guessing the code shape would waste grill questions