core-explore

name	core-explore
description	Read-only exploration — locate code, files, and conventions across the codebase and return conclusions, not file dumps. Fast, single-agent. Use when: locating where something lives in the codebase, learning a convention before editing, or answering a "where/how is X done" question without changing files.

One-Liner

Read-only reconnaissance that returns conclusions, not file dumps — find where things live and how the codebase does them, then hand the answer back.

Goal: $ARGUMENTS

Do

• Search broadly (by name, content, structure) to locate the relevant code/files.
• Read only the excerpts needed to answer; do not dump whole files into context (../../resources/token-budget.md).
• Return conclusions — where things live, how they connect, the naming convention — with file:line references.

Rules

• Read-only. Make no edits; this skill locates and reports.
• For a true multi-location sweep, fan out subagents (spawn-subagents, ../../resources/agent-tool-mapping.md) and synthesize.
• Caveman level — bulk file/grep/search results returned to the orchestrator are high-volume, low-ambiguity data → compress at ultra (per the Level selection matrix in ../../resources/token-budget.md).
• Hand findings to core-explain, core-fix, or a workflow as needed.