Use peek from your agent

peek works best inside the coding agent you already use. Install the peek skill and your agent learns to hand the right questions — whole-system, cross-file, "how does this flow" — to peek automatically, instead of grepping around to rebuild the structure itself.

The agent stays your hands for editing; peek is the map. It's a standard Agent Skill — one SKILL.md that works across Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Gemini, and other skill-aware agents.

1. Install the skill

   Install it with one command:

   npx skills add code-peekr/peek-skill

   Copy

   Prefer manual? Clone it into your agent's skills directory —~/.claude/skills/ (Claude Code), ~/.cursor/skills/ (Cursor), ~/.agents/skills/ (Codex), etc.:

   git clone https://github.com/code-peekr/peek-skill ~/.claude/skills/peek

   Copy
2. Ask a whole-system question

   Just ask — "how does a webhook get from the gateway to the database?" — and the agent will route it to peek. In agents that invoke skills by name, you can also call it directly:

   /peek how does auth flow end to end?

   Copy

When the skill delegates to peek

• Cross-file or cross-service flows ("trace what happens when…").
• Architecture and structure ("how is the indexer organized?").
• Impact and dependencies ("what breaks if I change this type?").
• Locating behavior ("where is rate limiting implemented?").
• Getting oriented in an unfamiliar or large repo.

It stays local for editing and trivial single-file lookups, and it queries the pushed repo — not your uncommitted working tree.