practice makes claude perfect

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)

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 · Prompt Engineering · Extend Claude Code

Feature	Location	Description
Agent Teams	.claude/agents/<name>.md	Multiple agents working in parallel on the same codebase with shared task coordination
Voice Mode	built-in command	speak to prompt - /voice to activate
Remote Control	built-in command	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

See orchestration-workflow for implementation details of Command → Agent → Skill pattern.

claude /weather-orchestrator

• RPI
• Boris Feb26 workflow
• Ralph plugin with sandbox
• Human Layer RPI - Research Plan Implement
• AgentOs - 2026 its overkill (Brian Casel)
• Github Speckit
• GSD - Get Shit Done
• OpenSpec OPSX
• Superpower
• Andrej Karpathy Workflow
• Creator of Clawd Bot Workflow

■ Workflows

• CLAUDE.md should not exceed 150+ lines. 60 lines in humanlayer (still not 100% guaranteed)
• use multiple CLAUDE.md for monorepos — ancestor + descendant loading
• 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
• agent teams and git worktrees for parallel development
• use Ralph Wiggum plugin for long-running autonomous tasks

■ Daily

• update Claude Code daily and start your day by reading the changelog

■ Hourly

• commit often, as soon as task is completed, commit.

■ 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
• explore settings.json features like Output Styles (e.g. Explanatory when learning a new codebase), Plan Directory, Spinner Verbs for a personalized experience

■ 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
• use a different model for QA — e.g. Codex for plan and implementation review

• Boris setup - 5 tips | 03/Jan/26 ● Tweet
• 10 tips for using claude code by team itself | 01/Feb/26 ● Tweet
• 12 ways how people are customizing their claudes | 12/Feb/26 ● Tweet
• Git Worktrees - 5 ways how boris is using | 21 Feb 2026 ● Tweet
• Seeing like an Agent - lessons from building Claude Code (Thariq) | 28 Feb 2026 ● Article