5 Claude Code Settings Worth Knowing
I spent months re-approving the same handful of curl commands for this blog's content pipeline before I sat down and read Claude Code's settings reference end to end. Most of what actually changes day-to-day life with it isn't a slash command you stumble into — it's a handful of keys in settings.json that nobody points you to until you go looking. Here are five that earned a permanent spot in mine.
Let your editor validate the file for you
Claude Code publishes an actual JSON Schema for its settings files. Point your editor at it and a typo or a misplaced key turns into a red squiggle instead of a silent no-op:
{
"$schema": "https://json.schemastore.org/claude-code-settings.json",
"permissions": {
"allow": ["Bash(npm run lint)", "Bash(npm run test *)"],
"deny": ["Read(./.env)", "Read(./.env.*)"]
}
}
VS Code, Cursor and anything else that understands JSON Schema will now autocomplete keys and flag values the schema rejects. It can lag a release or two behind the CLI, so don't panic over a warning on something genuinely new — but for the 95% case it turns a guessing game into autocomplete.
Permission rules merge across files — they don't replace
I assumed for a while that a personal ~/.claude/settings.json and a team's committed .claude/settings.json were "last one wins." They're not, for list keys like permissions.allow: Claude Code concatenates them. Your project's shared rules keep applying no matter what you add in your own file, and your own Bash(git *) doesn't need to be re-declared by every teammate. The exceptions are a short list of keys that hold an ordered sequence instead of an unordered set — fallbackModel and modelPicker among them — where the highest-precedence file wins outright instead of merging.
Strip the AI fingerprint from commits, if that's what you want
The attribution key controls the Co-Authored-By trailer Claude Code adds to commits and the attribution line it adds to PR descriptions. Setting a field to an empty string removes just that one; there's a third field for the session link specifically:
{
"attribution": {
"commit": "",
"pr": "",
"sessionUrl": false
}
}
This isn't about hiding anything — it's a legitimate call plenty of teams make about commit history hygiene, and it's a documented, one-time setting rather than something you have to remember to ask for on every commit message.
A one-line status line beats a dashboard tab
The status line runs any shell command you give it against JSON session data on stdin and prints whatever comes back. You don't need a script file for something simple — jq alone is enough to surface context usage without alt-tabbing anywhere:
{
"statusLine": {
"type": "command",
"command": "jq -r '\"[\\(.model.display_name)] \\(.context_window.used_percentage // 0)% context\"'"
}
}
It re-renders on the interface's own events by default; add a refreshInterval only if you're showing something time-based like a clock, since that's the one case the event-driven updates won't catch on their own.
Run /status before you assume a setting didn't take
The single most useful debugging habit I picked up: when a setting doesn't seem to apply, run /status before touching the file again. It lists exactly which settings files loaded for that session. Precedence runs managed settings, then a --settings flag, then project-local, then shared-project, then user — so a value you set in your personal file can be silently overridden by something your team committed, and /status is faster than re-reading four files to find out which one won.
None of these are exotic — that's the point. They're the kind of thing you'd only find by reading the reference doc start to finish, which almost nobody does for a tool they already use every day. If you've only ever touched /config, the actual JSON file underneath it has more in it than the menu lets on.