The highlights of the Anthropic Claude Coding Best practices:
1. Tailor Your Environment
Use CLAUDE.md files: Placed in your repo (root, parent, child, or home directory), this auto-included file allows you to document project-specific instructions—bash commands, style guidelines, setup tips, common caveats, etc. Keep it concise and iterate it over time.
Tune permissions: Claude Code prompts for permission when performing actions like file editing or running bash commands. You can customize allowed tools via prompts, using .claude/settings.json, or CLI flags to streamline workflows.
Integrate gh CLI: Installing GitHub CLI enhances Claude’s ability to manage issues, PRs, and more directly.
2. Expand Claude’s Toolbox
Leverage bash tools: Claude inherits your shell tools—just document custom tools in CLAUDE.md so it recognizes them.
Use MCP (Model Context Protocol): Claude can access external MCP-capable tools (e.g., Puppeteer, Sentry). Define them in project/global configuration or .mcp.json so they’re automatically recognized.
Create custom slash commands: Store reusable workflows (e.g., “fix GitHub issue”) in .claude/commands using $ARGUMENTS placeholders. These show up as slash commands for easy reuse across your team.
Anthropic literally dropped the smartest one-pager on using AI at work pic.twitter.com/jS7K9wWsoe
— Aadit Sheth (@aaditsh) August 7, 2025
3. Follow Established Workflows
Explore → Plan → Code → Commit: Have Claude inspect files, plan using “think” commands (like “think harder”), write code, then commit with a PR. Planning first improves effectiveness.
Test-driven development (TDD) cycle: Ask Claude to write tests, commit them, then write code to pass tests—and iterate, possibly using subagents to verify correctness.
Visual iteration workflow: Supply prototypes (e.g., screenshots), ask Claude to implement, review results, and iterate. Repeat until visually accurate.
Safe YOLO mode: Use –dangerously-skip-permissions in quarantined containers for bulk edits—risky but efficient in isolated environments.
Codebase Q&A: Claude can answer questions about unfamiliar codebases—substitute for pair programming to explain logic, file behavior, or code design.
Git and GitHub integration: Claude can write commit messages, handle revert or rebase conflicts, generate and triage pull requests, review code, and interact with GitHub using gh.
Jupyter notebook support: Claude can interpret outputs and visuals in notebooks, reshape code aesthetics, and improve readability.
4. Optimize Your Workflow
Be precise in prompts: Clear, detailed instructions yield better results and reduce iterations. For example: specify edge cases, avoid vague commands like “add a calendar widget.”
Use visuals: Paste or embed images to help Claude align to visual requirements—essential for UI tasks.
Reference files and URLs: Use tab-complete to specify files, and include URLs for docs. Use allowlists to avoid repeated permission prompts.
Course-correct early: Use /clear, interrupt with Escape, or tweak prompts to guide Claude. Regularly clean context to avoid drift.
Use checklists/scratchpads: For multi-step tasks (like refactoring or bulk lint fixes), have Claude produce checklists in Markdown and tick them off as it progresses.
Use data injection: Feed logs or CSVs via pipes, bash, or tools to help Claude reason over real data.
5. Automate with Headless Mode
Use claude -p for non-interactive usage (e.g., CI scripts), combined with –output-format stream-json for structured responses—great for automating reviews, migrations, or batch processing.
Use Claude as a linter: Automate code reviews via styles or semantic checks that traditional linters miss.
6. Advance with Multi-Claude & Parallel Workflows
Claude reviewing itself: Run one Claude to write code and another (separate session) to review. Repeat context resets to refine output.
Parallel worktrees: Use multiple Git worktrees or clones with separate Claude sessions running independent tasks—great for parallel feature work.
Headless harnesses & pipelining: Programmatically call Claude in loops or pipelines for bulk tasks like migrations or scraping.
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.