Click on this badge to show the latest best practice
Click on this badge to show implementation in this repo
Click on this badge to see the Command → Agent → Skill orchestration workflow

Boris Cherny on X (tweet 1 · tweet 2 · tweet 3)

CONCEPTS

Feature	Location	Description
Commands	`.claude/commands/<name>.md`	Entry-point prompts for workflows — invoke with `/command-name`
Sub-Agents	`.claude/agents/<name>.md`	Custom agents with their own name, color, tools, permissions, and model
Skills	`.claude/skills/<name>/SKILL.md`	Reusable knowledge, workflows, and slash commands — load on-demand or invoke with `/skill-name`
Workflows	`.claude/commands/weather-orchestrator.md`
Hooks	`.claude/hooks/`	Deterministic scripts that run outside the agentic loop on specific events
MCP Servers	`.claude/settings.json`, `.mcp.json`	Model Context Protocol connections to external tools, databases, and APIs
Plugins	distributable packages	Bundles of skills, subagents, hooks, and MCP servers · Marketplaces
Settings	`.claude/settings.json`	Hierarchical configuration system · Permissions · Model Config · Output Styles · Sandboxing · Keybindings · Fast Mode
Status Line	`.claude/settings.json`	Customizable status bar showing context usage, model, cost, and session info
Memory	`CLAUDE.md`, `~/.claude/projects/<project>/memory/`	Persistent context via CLAUDE.md files and `@path` imports · Auto Memory · Rules
Checkpointing	automatic (git-based)	Automatic tracking of file edits with rewind (`Esc Esc` or `/rewind`) and targeted summarization
CLI Startup Flags	`claude [flags]`	Command-line flags, subcommands, and environment variables for launching Claude Code
AI Terms		Agentic Engineering · Context Engineering · Vibe Coding
Best Practices		Official best practices · Extend Claude Code

🔥 Hot

Feature	Location	Description
Agent Teams	`.claude/agents/<name>.md`	Multiple agents working in parallel on the same codebase with shared task coordination
Remote Control	CLI / claude.ai	Continue local sessions from any device — phone, tablet, or browser · Headless Mode
Git Worktrees	built-in	Isolated git branches for parallel development — each agent gets its own working copy
Ralph Wiggum Loop	plugin	Autonomous development loop for long-running tasks — iterates until completion

Workflow orchestration using the Command → Agent → Skill pattern.

Component	Role	Example
Command	Entry point, user interaction	`/weather-orchestrator`
Agent	Fetches data with preloaded skill (agent skill)	`weather-agent` with `weather-fetcher`
Skill	Creates output independently (skill)	`weather-svg-creator`

See orchestration-workflow for implementation details.

DEVELOPMENT WORKFLOW

TIPS AND TRICKS

■ Workflows

CLAUDE.md should not exceed 150+ lines. (still not 100% guaranteed)
use commands for your workflows instead of agents
have feature specific subagents (extra context) with skills (progressive disclosure) instead of general qa, backend engineer.
/memory, /rules, constitution.md does not guarantee anything
avoid agent dumb zone, do manual /compact at max 50%
always start with plan mode
vanilla cc is better than any workflows with smaller tasks
commit often, as soon as task is completed, commit.
agent teams and git worktrees for parallel development
use Ralph Wiggum plugin for long-running autonomous tasks

■ Utilities

iTerm terminal instead of IDE (crash issue)
Wispr Flow for voice prompting (10x productivity)
claude-code-voice-hooks for claude feedback
status line for context awareness and fast compacting
/permissions with wildcard syntax (Bash(npm run *), Edit(/docs/**)) instead of dangerously-skip-permissions
/sandbox to reduce permission prompts with file and network isolation
output styles: use Explanatory when learning a new codebase, Learning for coaching

■ Debugging

/doctor
always ask claude to run the terminal (you want to see logs of) as a background task for better debugging
use mcp (Claude in Chrome, Playwright, Chrome DevTools) to let claude see chrome console logs on its own
provide screenshots of the issue

REPORTS

Report	Description
Agent SDK vs CLI System Prompts	Why Claude CLI and Agent SDK outputs may differ—system prompt architecture and determinism
Browser Automation MCP Comparison	Comparison of Playwright, Chrome DevTools, and Claude in Chrome for automated testing
Global vs Project Settings	Which features are global-only (`~/.claude/`) vs dual-scope, including Tasks and Agent Teams
Skills Discovery in Monorepos	How skills are discovered and loaded in large monorepo projects
Agent Memory Frontmatter	Persistent memory scopes (`user`, `project`, `local`) for subagents — enabling agents to learn across sessions
Advanced Tool Use Patterns	Programmatic Tool Calling (PTC), Tool Search, and Tool Use Examples
Usage, Rate Limits & Extra Usage	Usage commands (`/usage`, `/extra-usage`, `/cost`), rate limits, and pay-as-you-go overflow billing