having all of herdr tui, herdr gui, orca, tern, rex installed is so confusing, each time i plan to start a new agentic work.
Going to keep only orca + rex for the next couple of weeks, less decisions to make.
Any one building a tool to share other tools people use inside enterprise? Like a hub, based on plug-in architecture, with scopes, permissions, context, tools, CDP support...etc
I developed many tools to boost my day to day job, only to find out others does similar stuff..
Am technical, am sad am not exploiting this tech as I should! Building many cool/throw-away experiments, and constantly learning, which has always been fun..
very weird thing happening rn
technical people are sad, they think their skills are now useless
non-technical people are desperately wishing they were more technical so they could use agents better
github stars often is not a good metric, too many repos hyped, not because they are useful but because are interested to use them, so they bookmark the repo
on the other hand, plenty of good repos with < 1K star out there!
the more i test these skillls the more i realize i should be as closer to vanilla LLM as possible, with exceptions for skills that fixes the repeatable tasks for my specific needs.
I have a huge mono repo (3Gb), I only work on a sub folder ~100mb, I prompted claude to do sparse checkout for only that subfolder whenever i create a worktree using @herdrdev, cd to it and run npm i and copy .env file. Smoothy π₯€
A thorough code review of a PR with, say, 100 changes can easily take an hour, I think a wiser use of that time is 10 min on a high-level review, then 50 min improving the AI setup, harness, or skills that led to those issues, fix the process not the outcome
I used to refine my prompts with another LLM or a skill like prompt-booster before sending them to the model, I donβt feel the need anymore. Frontier models have gotten so good at understanding intent that I can just explain what I want.
I think the agent will get it done in a couple of hours. But glad to see an improvement in LLM estimations overall, it used to estimate 2h work to 2 weeks :D