Amazing product; lots of feature requests still though!

I want 3 more than 4 or 5. 👍

It would give me insight into why sometimes costs explode even after I've recently done a compact.

Thanks for the kind words & feedback!

Be able to run some commands DURING a query/process/inference run, such as /cost

Tracking here: https://github.com/anthropics/claude-code/issues/535

Be able to pause and or update the query with fixed and/or additional information (e.g. typed/pasted the wrong thing need to update the inference)

After submitting, hit esc and tell claude what to do differently. Or double-tap esc to go back to the previous message + edit it.

You could show a running cost for the session so we don't need to run /cost ever

We considered this, but decided against it. We might consider adding this for verbose mode if folks feel strongly (please open a new issue).

Perma approve certain commands. I find myself approving the same commands over and over. You could also have an option that says approve non-destructive commands (e.g. ls/mkdir/yarn [build,start,run]/ cargo build/ cargo check, etc. and NOT rm).

This should be feeling a lot better with the most recent version. You can also manage the allowlist: https://docs.anthropic.com/en/docs/agents-and-tools/claude-code/overview#project-configuration.

Approve everything for session. Just give it something to do and let it do whatever. Most changes are very easily redone. Maybe don't rm -rf though? Which I'm pretty sure if already prevented. Example: I'm integrating to projects; there's tons of moving stuff around and trying to build over and over. I don't want to do this. Just give Code free reign. The goal is to get the AI doing as much stuff as possible

Just shipped shift + tab to auto-approve edits for a session. Looking at bash commands next -- it's much more complex to make sure commands are safe!

> You could show a running cost for the session so we don't need to run /cost ever

We considered this, but decided against it. We might consider adding this for verbose mode if folks feel strongly (please open a new issue).

I'm guessing the reasoning was, something something worry about users creating a negative association with prompting.

Here's what a certain LLM thinks of that (and it mirrors my thoughts on the matter)

---

Why users might want constant cost visibility, even though developers instinctively avoid it:

• Predictability and Control: Constant cost visibility allows users to form an intuitive sense of how each prompt or action translates into cost. Over time, this removes uncertainty and lets them confidently manage resources.
• Reduced Anxiety Through Familiarity: Initially, seeing costs constantly might cause discomfort—but repeated exposure actually reduces anxiety in the long run, because predictability removes the "surprise" or "shock" factor.
• Immediate Feedback Loop: Constant cost display creates a tight feedback loop between user actions and resulting costs, reinforcing the user's learning of efficient prompting habits.

A better design to meet this user preference would be:

• Optional Toggle:

Allow users to enable or disable constant cost visibility, so anxiety-prone users can avoid cost exposure, while others desiring predictability can leave it on.

• Subtle Real-Time Display:

Show a discreet but clear cost counter that updates immediately after each action, reinforcing the connection between usage and cost.

• Aggregate and Individual Cost Breakdown:

Display both the immediate prompt cost and cumulative session cost side-by-side, allowing users to see short-term and long-term implications at a glance.

These are good feature suggestions. I'm a big fan of the idea of showing the cost. I'm frequently having to check the console to see if I'm near limit because I don't want Claude to get interrupted in the middle of something. It helps me plan ahead to know how much credit I have left. It would also allow me to be more efficient with my prompting patterns to have that feedback.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.