@thsottiaux for refactors mainly

The main issue i have with goals is that sometimes it just continues to do stuff for the sake of it.

Biased towards incremental steps that are not really needed and could be compounded, always looking for the next thinnest valuable slice, this increases the cycles quite a lot imo.

@thsottiaux Why the fuck you have never asked non-technical users what they want? Or once they expressed it you always ignored. #keep4o

@thsottiaux It's been pretty good for unattended work, but 5.5-xhigh tends to get into an ADHD loop of checking "one more thing". It drained roughly 70% of my weekly plan because I left it overnight. It was a fun learning experience, though. Now I include hard stopping gates in the objective

@thsottiaux needs a "randomly" option

@simpsoka Honestly, goal is pretty much the only way I work with Codex both at home and at work. It's just too good. I had Codex looping yesterday to fix some sloppy Opus code and it cleaned up so much, loooooooooool

i burned myself by wasting whole week usage by reckless usage of /goal.

So since then i started using it more intentionally, first spend some time+thoughts to create more detailed plan, with real goal posts.

basically if my goal is running longer than 2-3 hours, theres something way to vague in it

Had great time using it since this shift in mindset.

@thsottiaux To plug in real world outcomes into the goal. checkout https://github.com/Nimrobo/superdense

E.g Views, Conversion rate, signups, product analytics.

@thsottiaux My idea is to have a chat with the agent about what I want. I put it into plan mode to get a plan. Ask it for a meta prompt to do the plan for the /goal and then give the /goal back.

@thsottiaux its unusable now because of the terrible rate limits. and I’m on the Pro plan btw.

@thsottiaux tactical use only, codex is a glorified autocomplete that lies when you need it most.

@SIGKITTEN @thsottiaux Is that because you forget about it?

@thsottiaux if you give me the 10x i'll make it default 👀

@thsottiaux for well-defined tasks, always. just not for interactive chit-chats.

@thsottiaux Only some tasks can be specified enough, even then, it's hard to spec every obvious thing.

E.g. /goal Cypress -> Playwright migration was discussed & done, but tests put in wrong places as autistic GPT fixated on `legacy` word and put all tests in a single file named that.

@thsottiaux reset the usage and i'll let you know 😉

@thsottiaux with the current usage limits /goal is unusable even on the 20x plan

@thsottiaux I have been using it on some stuff with clear bullet point lists to achieve, but not anymore because it costs too much and I prefer to delegate to lower reasoning models clear detailed plans crafted from GPT 5.5 (x)high

@thsottiaux /goal is terrible.. Feels like it auto compresses context and continues for Infinity... Here is a better idea when i am about to reach the context limit: "drop all unnecessary noise, retain all important information when you are compacted, are you ready?" then /compact.🤣

@thsottiaux I’m afraid to use up too much quota now that the rate boost are over, it eats up usage quickly lol.