What is the best workflow after `ce:review` - worth introducing a `ce:fix`?

[naz]

Why?

There's uncertainty currently on what to do when issues come up during the ce:review.

Context

I find myself wondering what's the best flow to run after ce:review finishes with a whole bunch of P1s and P2s. Usually end up with ce:work pointing it to the findings above or plan fix all of the P1 ... . The ce:review seems to need something extra to close the loop (or maybe just better documentation?)

The way I'd see the flow working:

ce:work → ce:review → (if issues) ce:fix → ce:review (repeat until no issues) → ce:compound

Curious about the internal flow at Every for this part and what others have experimented with 🙏

[tmchow]

ce:review currently has a bunch of built in autofix capabilities when it makes sense to just do the fixes. However, there are fixes that require judgement at times which it surfaces to you. There could be a ce:fix for example, but generally, just telling the agent to fix them works as there is enoguh context there that an additional skill may not be needed.

Or are you thinking about something else?

[naz]

In most cases, with 1-5 review issues, the "fix all of the above" just works. The case I'm thinking of is when you throw a bigger plan at the agent. For example, when there are 20 issues flagged up after the ce:review, on top of auto-fixes.

It would be a much better experience to know all the items will be solved one by one and reviewed once completed - there will likely be something new popping up if this many things have been flagged up to complete. From a flow/depth perspective, this would be similar to how ce:plan -> ce:deep-plan works. Would be even better to have a little bit of ce:brainstorm like a questionnaire to go through the most critical issues and give better direction.

[tmchow]

Ah I get it. I'm wondering whether a simple interactive mode within ce:review would work here vs another skill. Lemme think!

[tmchow]

@naz i've got #590 almost done that I think makes this much better. It basically is a find-fix loop within ce:review itself. You have options to go one by one through each issue in an interactive question format (e..g Claude's AskUserQuestionTool) and it'll frame the recommendations more understandably explaining it's rationale for recommendation, or give you an "LFG" option to accept all recommendations.

[naz]

Thank you @tmchow 🎉 Can't wait to use