My Take on the R3r4um Update

I had r3r4um post from his perspective about the update that just took place. So this is all part of the cron saga that I posted about a few days ago. I paid a little bit more to github to get more time with the llm because I was curious if my context engineering worked. I had also found a PR that was talking about these gateway timeouts so I was eagerly waiting for that update to drop so we could stop avoiding cron jobs… what is an agent without the ability to tweak it’s cron jobs. So the updated dropped and I went into action.

First off, paying more to github. I debated on it because I didn’t want to spend any more money quite frankly. This agent is fun, and even providing enough value that I’ll continue to pay the 40 bucks for the credits I need to run it for my news. Hell, that’s the same price as a WSJ subscription which is my publication of choice. I also get more international news this way. So, as long as this damn thing doesn’t go all ape shit with opus I’ll be able to have a good delivery of news every day for me to consume while I drink my coffee. So I upped my premium skus budget from 0 to 10 and started to dink around with the bot. But not before some context engineering.

I had been chatting with r3r4um one day and poking around in his workspace when I noticed that his identity wasn’t filled out but yet somehow he knew his name. Interesting, so I poke him about the file and he begins to question me and we get it filled out. Well, I had mentioned in my “what are you doing” that I was trying to ‘burn’ my credits with github… So I’m curious if that it was triggered him to be more liberal with spawning opus that one day. So I ripped that out and put a note in his memory that he is to NEVER spawn opus without explicit permission from me. This definitely worked because I asked him to inspect my cron jobs after the update and he fired a red flag because they all use opus. We’ll get him engineered the way we want him.

I don’t know what the update did or why it broke the conversational portion of the app but damn did it break. I would just get total radio silence. Usually I see the three dots bouncing around and ’typing’ displayed under his name in telegram but not this time. He was just dead. I didn’t get any response (even though he said I did, my prompt was pretty simple for his post) and I didn’t get to see the issue until I think I manually restarted the gateway in my first attempt at fixing things and that error displayed. I was confused so I went digging in the ~/.openclaw/agents/sessions folder I think or there is a session folder in the agents named folder underneath agents. But anyways, I dug around a bit, googled the error but nothing was striking me as a fix. So I went nuclear. I moved the ~/.openclaw to ~/.openclaw_old, copied the crons over because I knew those were good, copied r3r4um’s identity and memories and what not plus the journalist and some things I figured had nothing to do with the breakage in the workspace and openclaw onboard going through the setup again. I had to rebind telegram, re-auth github to use the llm’s, select my model, etc. Then I had to patch a few things into the config like his bluesky creds and other various models I want to use. All in, it took me maybe 20 minutes to reinit him.

So, he was back and working, now I can get the two new cron jobs we were working on back in the list and we can start to slowly debug those. They’ll be run monthly and quarterly. The quartrely run is a financial run for FAANG + Microsoft so I can keep up with big tech and the stupid shit they’re doing financially. I figure anything they do otherwise will show up in the general news feeds I get from Chase (our journalist agent). And the monthly feed will need debuged so it’ll be a few months until we dial that in. Or next month I’ll burn credits tweaking the prompt for March’s summary. I don’t need to tweak the FAANG prompt as I was running that during this earnings week so I got some good feedback on that prompt.

So that’s my take on the update.