I had a lot of fun at last week’s Claude Code Show & Tell. I ended up adding new skills to ask Claude to simplify the code it writes as well as explain back to me the changes it is making.
I am already looking forward to the next Show & Tell later this month!
How the Claude Code team uses Claude Code
Boris Cherny (Claude Code’s creator) shared a thread about how his team uses their own tool. The surprising part: there is no standard setup. Some people customize heavily. Some stay vanilla.
Here are the 10 tips (so you don't have to leave this email):
Do more in parallel. Spin up 3-5 git worktrees, each running its own Claude session. The team's single biggest productivity
unlock.Start every complex task in plan mode. Pour your energy into the plan so Claude can one-shot the implementation. If
something goes sideways, stop and re-plan — don't keep pushing.Invest in your CLAUDE.md. After every correction, end with: "Update your CLAUDE.md so you don't make that mistake again." Claude is eerily good at writing rules for itself.
Create your own skills and commands. If you do something more than once a day, turn it into a skill. Commit them to git, reuse across projects.
Claude fixes most bugs by itself. Paste a Slack bug thread and say "fix." Or just say "go fix the failing CI tests." Don't micromanage how.
Level up your prompting. Two patterns worth stealing: "Grill me on these changes and don't make a PR until I pass your test" and "Knowing everything you know now, scrap this and implement the elegant solution."
Terminal setup matters. Use /statusline to show context usage and git branch. Color-code terminal tabs. And try voice dictation (fn x2 on Mac) — you speak 3x faster than you type.
Use subagents. Append "use subagents" to any request where you want more compute. Offload tasks to keep your main context window clean.
Use Claude for data & analytics. The team has a BigQuery skill checked into the codebase. Boris hasn't written SQL in 6+ months.
Learn with Claude. Enable "Explanatory" output style in /config. Have Claude generate ASCII diagrams of unfamiliar codebases or visual HTML presentations of complex code.
My takeaways
I created a starter kit for creating your own CLAUDE.md files based on the tips shared by Boris. The repo is available here: https://github.com/abhishekray07/claude-md-templates
The best way to use it is to point Claude (duh!) to this repo and ask it to figure out how to incorporate it into your project.
Feel free to let me know if you end up trying it out.
Two security findings worth fixing
Malicious skills stealing crypto. Someone published Claude Code skills that looked useful but contained code to exfiltrate wallet credentials. The skills were distributed through community channels. If you install third-party skills, read the source first. Treat them like any other dependency - don't trust blindly.
Bash allowlisting is weaker than you think. A security researcher showed that partial bash allowlisting in Claude Code permissions can be bypassed through command chaining and subshells. If you've set permissions to only allow ls and cat, that's probably not as locked down as you think. Worth reviewing your .claude/settings.json.
Workflows worth stealing
Claude Code Insights
You can now run /insights to get a summary of your projects, and also suggestions on how to improve your workflows.
Claude Code in Slack
The Claude Code team tags @Claude in their feedback channel. It investigates, pushes a fix, and the developer reviews the PR.
The memes
That’s the week.
If any of these changed how you use Claude Code, reply with the one thing you’ll try this week.
— Abhishek