User intent	Topic	Start here
"What is CLAUDE.md / memory / rules?"	Product	`references/product/memory.md`
"Shift+Tab modes / plan / auto / bypass"	Product	`references/product/permission-modes.md`
"allow deny Bash(git …) rules"	Product	`references/product/permissions.md`
"settings.json scopes / managed / local"	Product	`references/product/settings.md`
"claude flags / -p / --agent"	Product	`references/product/cli-reference.md`
"/commands in the REPL"	Product	`references/product/commands.md`
"CLAUDE_CODE_* env vars"	Product	`references/product/env-vars.md`
"Tool names for permissions or hooks"	Product	`references/product/tools-reference.md`
"Plugins vs `.claude/` standalone"	Product	`references/product/plugins.md`
"Push events into a session"	Product	`references/product/channels.md`
"Skill vs subagent vs hook vs plugin?"	Agents	`references/agents/execution-decision.md`
"Subagents / Explore / fork"	Agents	`references/agents/subagent-overview.md`
"How to build a custom agent"	Agents	`references/agents/custom-subagent-structure.md`
"Agent frontmatter fields"	Agents	`references/agents/frontmatter-fields.md`
"Multiple sessions as a team"	Agents	`references/agents/agent-teams.md`
"Agent patterns and composition"	Agents	`references/agents/agent-patterns.md`
"Automate with hooks"	Hooks	`references/hooks/hook-lifecycle.md`
"Hook events catalog"	Hooks	`references/hooks/hook-events.md`
"Match syntax / if filters"	Hooks	`references/hooks/matchers-and-if.md`
"Hook input/output/exit codes"	Hooks	`references/hooks/input-output.md`
"Hook security / block commands"	Hooks	`references/hooks/security.md`
"Debug hooks / /hooks menu"	Hooks	`references/hooks/debugging.md`

Workflow

Identify user intent — What is the user asking about? (see table above)
Route to topic — Product, Agents, or Hooks
Load the reference — Read the corresponding file
Follow the guidance — Apply the instructions, checklists, or examples
Escalate if needed — If the reference says "go deeper," load the deeper reference

When to Combine Topics

Agent + Hook: When building an agent that needs deterministic guardrails (see references/agents/hooks-in-agents-bridge.md)
Product + Agent: When setting up project structure for agents (see references/agents/repo-structure.md)
Product + Hook: When configuring permissions alongside hooks (see references/product/permissions.md)
All three: When designing a complete .claude/ project setup

Rules

Never answer from memory alone — always load the relevant reference.
If the user question spans multiple topics, load references in parallel.
If the reference is missing, fall back to the official docs URL in the frontmatter.
If the user asks for a checklist, load the matching checklist from checklists/.
If the user asks for an example, load from examples/.
Never duplicate deep implementation detail in the router — push it to references.

Error Handling

Missing reference: Ask the user to clarify which topic they need.
Outdated reference: Check the last_reviewed date in the frontmatter and warn if stale.
Conflicting guidance: Prefer the official docs URL over the local reference.

claude-code-guide

Claude Code Guide

Quick Router