headroom.walls.sh · permissions

Claude Code permissions

Stop Claude Code from asking permission for every command — configure allow and deny rules once in settings.json.

Why permissions exist

Claude Code can run shell commands, read and write files, and call external APIs. By default it asks for confirmation before any action that could have side effects — a bash command, a file write, a network request. This is safe but noisy once you've decided to trust Claude Code in a given context.

The permissions block in settings.json lets you pre-approve (or pre-block) specific patterns so Claude Code doesn't interrupt you for things you'd always allow.

The permissions block

{
  "permissions": {
    "allow": [
      "Bash(git *)",
      "Bash(npm run *)",
      "Read(**)",
      "Edit(**)"
    ],
    "deny": [
      "Bash(rm -rf *)",
      "Bash(sudo *)"
    ]
  }
}

Each entry is ToolName(pattern). Deny takes precedence over allow — if an action matches both, it's blocked.

Tool names

Tool	What it covers
Bash	Any shell command. Pattern matches the full command string.
Read	Reading a file. Pattern matches the file path.
Edit	Editing an existing file. Pattern matches the file path.
Write	Writing a new file. Pattern matches the file path.
WebFetch	Fetching a URL. Pattern matches the URL.
WebSearch	Web searches. Pattern matches the query.
mcp____	MCP server tools. Pattern: `mcp__servername__toolname`.

Pattern syntax

Pattern	Meaning
*	Matches anything within a single path segment or word
**	Matches anything including path separators — use for file paths
git *	Any git subcommand and flags
npm run *	Any npm run script

Common allow patterns

Git (all operations)

"Bash(git *)"

npm / yarn / pnpm

"Bash(npm *)",
"Bash(yarn *)",
"Bash(pnpm *)"

Build tools

"Bash(make *)",
"Bash(cargo *)",
"Bash(go *)",
"Bash(swift *)"

Testing

"Bash(pytest *)",
"Bash(jest *)",
"Bash(rspec *)",
"Bash(npm test)"

File operations (all files)

"Read(**)",
"Edit(**)",
"Write(**)"

File operations (project only)

"Read(src/**)",
"Edit(src/**)"

Web fetching

"WebFetch(*)"

Common deny patterns

"Bash(rm -rf *)",
"Bash(sudo *)",
"Bash(curl * | bash)",
"Bash(chmod 777 *)"

Where to put permissions

Two scopes — pick based on trust level:

File	Use when
~/.claude/settings.json	You trust Claude Code with git/npm/read globally — same rules everywhere
.claude/settings.json	Project-specific tools (Docker, kubectl, deploy scripts) — commit with the repo

Project-level rules apply on top of user-level. Both allow lists merge; deny takes precedence across both.

A practical starting config

{
  "permissions": {
    "allow": [
      "Bash(git *)",
      "Bash(npm *)",
      "Bash(make *)",
      "Bash(cat *)",
      "Bash(ls *)",
      "Bash(echo *)",
      "Bash(which *)",
      "Bash(pwd)",
      "Read(**)",
      "Edit(**)"
    ],
    "deny": [
      "Bash(rm -rf *)",
      "Bash(sudo *)"
    ]
  }
}

This eliminates most prompts for typical development work while blocking the most dangerous patterns.

--dangerously-skip-permissions

For fully automated / headless use (CI, scripts), pass --dangerously-skip-permissions to skip all confirmation prompts:

claude --dangerously-skip-permissions --print "Run the test suite and report failures"

This bypasses ALL confirmation prompts including destructive operations. Only use in sandboxed environments, containers, or CI where the blast radius is bounded. Never in an interactive session on your main machine.

Add the statusLineHook while you're here

The permissions block lives in the same ~/.claude/settings.json as the statusLineHook that powers Headroom's live usage display. If you're editing settings.json to fix permissions, it takes one more line to get your Claude Code session and weekly usage in the menu bar:

{
  "statusLineHook": "cat ~/.claude/headroom-usage.json 2>/dev/null | jq -r '"CC \(.sessionUsagePct|floor)%·\(.weeklyUsagePct|floor)%"' 2>/dev/null || echo 'CC --%'",
  "permissions": {
    "allow": [
      "Bash(git *)",
      "Bash(npm *)",
      "Read(**)",
      "Edit(**)"
    ]
  }
}

Headroom reads that file and shows your session (5h) and weekly (7d) usage as a live % in the menu bar — color-coded before a limit stops you.

brew install --cask patwalls/tap/headroom

→ Full settings.json reference
→ statusLineHook setup
→ MCP server configuration