Claude Code hook and plan usage

Both are optional and only concern Claude Code. collied doctor checks each one.

The PermissionRequest hook

Without the hook, an approval alert shows the tool call as read from the screen. With it, the alert names the exact command or file. The hook only reports the call to collied; Claude Code's own dialog does not change.

Add it to ~/.claude/settings.json:

{
  "hooks": {
    "PermissionRequest": [
      { "hooks": [{ "type": "command", "command": "~/.cargo/bin/collied hook", "timeout": 5 }] }
    ]
  }
}

Plan usage

Claude Code passes the plan's 5-hour and weekly limits only to the status line command. To share them with collie, add this line to your status line script, after it reads its input into input (for example with input=$(cat)):

printf '%s' "$input" | ~/.cargo/bin/collied statusline >/dev/null 2>&1 &

It prints nothing, so your status line looks the same. It records the limits and each session's context window size, so the context ring uses the exact window. Everything else in the status line input is dropped.

If your statusLine.command is a single command rather than a script, point it at a small script that reads input=$(cat), runs the line above, then pipes printf '%s' "$input" into your old command.

Note

The limits exist only for claude.ai Pro and Max plans, after the session's first reply. collied doctor reports when the tap last recorded them.

The limits show in Usage and on the Apple Watch.