Allowing work without approvals
The three degrees between “ask about everything” and “ask about nothing”, how bypass is turned on and what it gains and costs you.
By default the Agent asks before it writes to a file or runs a command. On the phone that shows up as work standing still, waiting for your tap. That is deliberate — but there are situations where it is more of a nuisance than a help.
Three degrees, not two
Between “ask about everything” and “ask about nothing” there is one more degree, which often gets overlooked:
- Ask every time
- The default. Every write and every command waits for approval. The safest and the slowest.
- Standing approval for a given kind of action
- When the Agent asks for approval, you can choose that it should stop asking about this kind of action. It applies to that kind, not to everything — you get running tests out of the way, deleting files not. It can be taken back at any time in Settings.
- Without approvals (bypass)
- The Agent does not ask at all. It is turned on in the Server's configuration, not from the phone — a deliberate decision at the computer, not a tap on the tram.
How to turn bypass on
In the Server's configuration file, the one you hand to it through --config:
{
"no_approvals": true
}It applies from the next start of the Server. The phone notices by itself — it stops asking.
What you gain
- Longer tasks finish without you. This is precisely the case where you give out work and walk away — with a question after every command the Agent would wait until you looked.
- Tests, formatting and the build run in one go, instead of stopping at every step.
- Less tapping on the phone for things you would always allow anyway.
What you lose
- The brake. The Agent runs with your rights and without approvals it can do anything on the computer that you could from the command line.
- The moment to think. The question “may I run this command?” is often the place where you notice that the Agent understood the task differently than you meant.
- The overview of what is coming. You still see the steps, but as done, not as a proposal.
When to turn it on and when not
Turn it on in a project that is in version control and whose loss would not hurt you — your own experiment, a rewritable prototype, a project that can be thrown away and cloned again. There bypass is exactly what makes coden a tool you do not have to sit with.
Do not turn it on where the Agent touches production data, deploys, or works in a directory holding things that exist nowhere else. The line between that and the previous paragraph is not technical but yours.
A practical compromise: bypass on, the project in git and a commit before every bigger task. Getting back is then one command away — and that is cheaper than approving every step.
Stuck somewhere other than what is here? Write to support@coden-app.com.