Approval gates

How StoreFrame's default Claude session skips per-action approval, and when to opt out.

Claude Code normally prompts you before taking individual actions — running a command, editing a file, running git, etc. In the Hub Console tab, those prompts are turned off by default.

A Bash mode Console tab with `claude` running in standard mode

The default: skip permissions

New Claude tabs in the Console boot with the --dangerously-skip-permissions flag. Claude executes its planned actions without asking for per-step approval. The tradeoff:

ProCon
Faster — Claude doesn't wait on you between stepsLess explicit oversight; you see what happened after the fact
Better for autonomous tasks ("investigate and fix these exceptions")Worse for sensitive operations where you want to confirm before each action
Matches the "give Claude a task and walk away" workflowPotentially dangerous near irreversible operations if your prompt is too broad

For most exploratory and diagnostic work — reading logs, querying the database, running bin/magento commands, editing code — the default is fine.

Opting out

If you want explicit approval for each action, spawn a Bash tab in the Console and run claude yourself without the flag:

claude

Claude now prompts before each action. Confirm with y or reject with n. This is the standard Claude Code behavior outside of StoreFrame.

When to opt out

  • Working on production data you cannot recover (database mutations on live customer data)
  • Touching system files outside the project directory (/etc/, system services)
  • Pairing on something sensitive where you want a second set of eyes on every step
  • Unfamiliar territory where you'd rather approve each action before it runs

What Claude still confirms

Even with --dangerously-skip-permissions, Claude applies its own judgment before truly irreversible operations. Before running a DROP TABLE, deleting files outside the repository, or forcing a git history rewrite, Claude describes the planned action in plain English and asks for explicit confirmation. The flag skips the system-level approval UI; it does not remove Claude's own caution about destructive work.

On this page