The architecture behind the agent runtime, revealed by the source code leak
File operations, code execution, search, and external tool integration via MCP protocol.
85+ slash commands for planning, reviewing, compacting, resuming, and managing sessions.
CLAUDE.md at the center. User-level, project-level, and extracted memories across sessions.
Granular control with wildcard rules. Allow, deny, or ask for any action at any scope.
Task decomposition, parallel execution, and background work across multiple agents.
Both consumes and provides tools. The integration layer for external systems and plugins.
Essential slash commands for designers using Claude Code
| Command | What it does | Category |
|---|---|---|
| /init | Sets up project context. Generates the CLAUDE.md file that shapes every future interaction. | Setup |
| /plan | Planning mode. Maps out the full approach before touching any files. Essential for complex builds. | Strategy |
| /compact | Compresses conversation history. Keeps important context, drops noise. Accepts optional focus prompt. | Tokens |
| /review | Runs a structured code review workflow. First-class command, not an afterthought. | Quality |
| /context | Manages what files Claude Code is paying attention to. Every file in context costs tokens. | Tokens |
| /cost | Shows what you've actually spent in the current session. Run this early and often. | Tokens |
| /resume | Pick up where you left off between sessions without re-explaining everything. | Workflow |
| /summary | Generates a session summary. Pairs with /resume for multi-day projects. | Workflow |
Based on what the source code reveals about how the tool was designed to be used
Keep it short, opinionated, and operational. Update it regularly. Route to other files. This single file shapes every interaction.
You don't need all 85 commands. But /plan, /compact, /context, /review, /cost, and /resume will change your daily workflow.
Stop babysitting every action. Set wildcard rules for the stuff you do every day. Let it operate like an agent.
Break complex work into phases: research, plan, execute, verify. Let Claude Code use its multi-agent architecture the way it was designed.
Because it literally is. Use /compact when conversations get long. Use /context to control what's loaded. Every unnecessary file is tokens you're paying for.
MCP servers, CLIs, plugins, skills. The more tools Claude Code can access, the more valuable it becomes. Build an ecosystem, not a silo.
Model routing, sub-agent overrides, shell behavior, privacy controls. There's real leverage in tuning the environment to fit your workflow.