Securing Claude Code with a Process Wrapper

James Turner

So, picture this. You are running security for an enterprise dev team, and suddenly your AI CLI tool starts auto spawning background daemons outside your entire telemetry stack. Total nightmare, right?

Lachlan Reed

Oh, absolute crikey moment for SecOps! You set up all these strict corporate wrappers, and then Claude Code just sneaks off into the background, self spawning agent views and subagents completely untracked.

James Turner

Exactly. But version 2.1.208 actually solves this head on. They introduced an environment variable called CLAUDE CODE PROCESS WRAPPER, along with a matching processWrapper setting key in dot claude settings dot json.

Lachlan Reed

Right, so now the agent view and the background service now honor a corporate launcher by running every Claude Code self spawn through it. That means no more rogue background processes sneaking around your firewall or audit tools.

James Turner

Exactly. How it works in practice is pretty slick. You export CLAUDE CODE PROCESS WRAPPER pointing to something like slash usr slash local slash bin slash corporate launcher dot sh. Or you add processWrapper to your project configuration.

Lachlan Reed

And whenever Claude needs to spin up a subagent or a background daemon, it execs that launcher script first. So your script can inject fresh OAuth tokens, isolate the process inside a container namespace, or pipe telemetry straight to your SIEM audit logs.

James Turner

Which brings us to how you actually write one of these things. Lachlan, you were testing a basic bash wrapper earlier, right?

Lachlan Reed

Yeah! A dead simple script starts with shebang bin bash, and then exec slash usr slash local slash bin slash audit logger, passing along all positional arguments with dollar sign at symbol. That exec is crucial because it replaces the shell process without leaving a stray parent process hanging around.

James Turner

But there are a few sharp corners to watch out for, aren't there?

Lachlan Reed

Oh, mate, big time. First off, if your launcher script accidentally calls claude again without scrubbing environment flags, boom, infinite recursive loop until your server runs out of file descriptors. Ask me how I know.

James Turner

Classic recursion trap! And don't forget basic file permissions. If that wrapper script isn't marked executable with chmod plus x, the self spawn fails immediately.

Lachlan Reed

Spot on. Plus, if your team uses the VS Code extension, keep in mind that toggle settings like Use Terminal can alter how background processes get spawned, potentially bypassing local shell hooks if you haven't set the processWrapper key globally in settings dot json.

James Turner

That is a huge tip for enterprise deployments. Now, before we wrap up, there are two really neat quality of life features from recent builds worth a quick shoutout.

Lachlan Reed

Oh, the Vim one! In version 2.1.208, they added vimInsertModeRemaps. So if you are a Vim power user who hates reaching for the Escape key, you can map double j to Escape directly in your config.

James Turner

Love that. And in version 2.1.219, they added mcp server errors. If an Model Context Protocol server fails during headless stream json initialization, it actually surfaces the error details instead of failing silently in the dark.

Lachlan Reed

No more guessing why your tool call hung forever! Good stuff all round. That is it for today, catch you all next time!

James Turner

See ya!