LLM Wiki

Appearance

Built 26/04/17 09:39commit 8de3d61

Affaan Mustafa Claude Code Shorthand Guide ​

中文 | English

Summary ​

This source is a practical inventory of a mature Claude Code setup: skills versus commands, hook lifecycles, subagent scoping, user and project rules, MCP and plugin budgeting, and the terminal-editor pairing patterns that keep day-to-day coding sessions fast and legible.

Source ​

Key Contributions ​

  • Separates skills, slash commands, hooks, rules, subagents, MCPs, and plugins into distinct control surfaces instead of treating "configuration" as one blur.
  • Makes context-budget management concrete through explicit MCP and plugin disabling, plus a rule of thumb for how many tools should stay active.
  • Shows how lightweight terminal practices such as tmux, git worktrees, and named forks reduce risk before heavier multi-agent orchestration is needed.
  • Documents the editor layer as part of the control plane: file watchers, git review, auto-save, and quick navigation all affect agent reliability.

Strongest Claims ​

  • Claude Code setup should be treated like fine-tuning an operating environment, not like building a complicated architecture from day one.
  • Context window health is a first-class constraint, so unused MCPs and plugins should be aggressively disabled.
  • Parallel work is only worth it when scopes are orthogonal and code overlap stays low.
  • Hooks are the right place to automate repeated guardrails such as formatting, type checks, and push-time review.