SETTINGS.JSON PERMISSIONS CONFIGURATION

FILE: .claude/settings.json (checked into repo; all team members inherit these rules)

{
  "permissions": {
    "allow": [
      "bash:read",
      "bash:grep",
      "bash:find",
      "bash:git:status",
      "bash:git:log",
      "bash:git:diff",
      "bash:npm:test",
      "bash:npm:run:dev",
      "read:/**",
      "edit:/**",
      "write:src/**",
      "write:tests/**"
    ],
    "require_approval": [
      "bash:npm:install",
      "bash:npm:uninstall",
      "bash:npm:publish",
      "bash:git:push",
      "bash:git:reset",
      "bash:rm",
      "bash:mv",
      "delete:/**",
      "write:.env**",
      "write:package.json",
      "network:fetch",
      "network:post"
    ],
    "deny": [
      "bash:sudo",
      "bash:chmod",
      "bash:chown",
      "read:.env**",
      "read:.github/workflows/**",
      "delete:./**"
    ]
  }
}

WHAT THIS DOES:

ALLOW (no prompt): Read any file, grep for symbols, run tests, check git status/history, edit source files, start dev server. AI works fast without interruption.

REQUIRE_APPROVAL (must confirm): Install dependencies, push to remote, delete files, modify .env files, make network calls. Human decision gate prevents mistakes.

DENY (blocked always): Sudo commands, file permission changes, reading secrets, accessing CI workflows.

THE TRAP AVOIDED: -dangerously-skip-permissions flag removes all gates. Instead, this config lets AI be useful AND safe by gatekeeping risky operations only.

TEAM SETUP: Commit settings.json to repo. Include .env in deny list. Each developer inherits the same rules when they clone. No per-developer configuration drift.

DEVELOPMENT FRICTION: Reading takes 0 seconds (allowed). Running tests takes 0 seconds. Git push takes 2 seconds (human confirms). Deleting takes 2 seconds (human confirms). Trade-off: 2x approval prompts for 100x more autonomy on daily tasks.

VERIFY: Run claude permissions list to see what's currently configured.