Activity
Mon
Wed
Fri
Sun
Jul
Aug
Sep
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
What is this?
Less
More

Memberships

Clief Notes

38.8k members • Free

AI Masters Community with Ed

12.3k members • Free

CC Strategic AI

3.2k members • $27/month

Chase AI Community

68.3k members • Free

4 contributions to Clief Notes
Too many skills?
In my efforts to build more autonomous systems, I’ve been developing a large collection of skills - and I think following the designs being taught in this community. Now I’m sitting at like 130 skills and getting concerned about context window issues as well as errors I’m seeing from codex, Claude and other coding clis that I’ve gone over some skills budget values. I have been playing around with the notion of a ‘skills on demand’ style library approach where a skill is lightly defined and the Claude.md/skills.md file is used to identify the necessary skills for a task and then calls/loads them into context to use. I haven’t done a good search of existing tools but I figured I would ask here first to see if someone has tackled this issue and how, or what completely failed so I can avoid that in my own effort. Thanks!
1 like • 24d
@Simon Gonzalez De Cruz yes! I totally bungled that concept.
1 like • 24d
@Simon Gonzalez De Cruz I have been going down that route…but I’ve also found that Claude often provides a less than ideal answer on hard things. It likes to cut corners and deliver solutions that are token efficient and fast, not always solving the problem. So I was intentionally looking beyond that guidance to see if anyone had implemented and what worked/didnt. I could be wrong to be so critical of Claude for that kind of stuff but I’ve been burned too many times so it’s a learned behavior now for me :)
Anyone doing Development orchestration?
I’ll start by saying I didn’t look exhaustedly to see if others had posted about this. I’ve built a bunch of skills to run engineering process from PRD generation through coding. The skills break the work into epics and tickets. Im starting to work on ways to put those tickets into GitHub issues, and I am playing with ideas on how to orchestrate a thin layer that manages the ticket queue and ensures prerequisites are done fist, managing ticket order, etc. The orchestrator would automate assigning Claude or codex to code the ticket, etc. Has anyone found a good solution for this? I know there are many and I’ve been looking at a bunch. I started to build my own but that seemed foolish. My goal is to be able to spec and design various applications and be able to have the bulk of the coding cycle process be autonomous. Right now I’m doing loads of babysitting in the terminal to make sure it picks up the next ticket, reminding the LLM to keep working on the next tickets, etc. some of these projects are taking days and, while clearly complex, it should be smoother than this. Thanks !
2 likes • 26d
@Eleanor Whitmore it’s been a few things, in no particular order: - leveraging sub agents and agent teams is crucial, otherwise context goes too fast to get more than one ticket in a session, so you need to /clear or /exit and start over. Very time consuming to manage context windows - Side quests are killer. Things come up when developing, so it’s unfortunately routine for either the LLM to have failed to properly decompose the project into tasks or just technical issues arise because of any number of issues.so you go off and solve that thing and then come back to work on the original issue and it’s mentally draining. - Babysitting - just having to keep an eye on the console so the agents are sitting on a question or hung…or with Claude they just hit capacity. Progress gets made but the cost is high on a task that should be mostly managed by some thin application.
Getting Claude/Codex to build to plan?
Maybe it's just me and my setup or approach, but I'm finding repeatedly that I spend a couple of passes on a planning document, lay out technical specs, define success/failure criteria, test requirements, outline automation testing requirements, etc. and yet...the outputs are husks of what I designed. Loads of stuff isn't done at all, a bunch is just stub code, no tests are written, no acceptance criteria was ever done, etc. I think I've internalized the approaches Jake evangelizes, and I've got it all in documents, but at the end of the day I'm finding (via Claude's own 'admissions') that the models just treat any rules and guidance as optional. They just ignore the requirements to read the documents - e.g. it will say stuff like 'The instructions said to read the specs, I read the first few lines and then figured I knew what to do so I stopped and got to coding' or some variant of that. I ended up building a bunch of harness gates and hooks (in Claude) but now I find that Claude just spends more time figuring out how to satisfy the hook validation than doing the actual work needed to pass it. I can build simple sites or tweak codebases with super small targeted changes, or build remotion videos to my hearts content using the processes Jake is showing, but anything substantive? Nope. It fails miserably. I want to believe I'm just doing it wrong, but so far I'm not seeing how it would be at all possible to do any kind of substantial coding using these coding tools. I'm a CLI person, always hand coded (the hard way, I know) so maybe there are better answers using IDEs? Would love any thoughts others have on how to avoid these issues and get the models to actually follow the file structure requirements and not just glance at the docs but read them and follow them.
Tell me what you want me to build you.
I had an idea. I want you all to comment A traditional workflow that you have right now. Like some sort of set of tasks that your industry does or what you're trying to do. I'm going to see how fast I can build some sort of prototype or solution that automates part of it in your industry or pain points and I'm just going to make a mass video that shows all of them for everyone who comments and this. Whatever I build if you want it DM me and I'll send you something. If you don't comment in this, I'm not building it But let's see what we can do. Everyone comment what you want me to build below or more importantly what you do and what your industry is
6 likes • Mar 9
@Kürşad Kozelo i built a remotion workflow that takes a script and uses nano banana to generate the images, tts to generate audio and assemble it per the script. My results feel hacky and not fluid. I’m going to watch the video from Jake to see how he gets the process 1) more organized and optimized and 2) fluid. What have your experiences been with using remotion? It feels like folks have had much better results than what I’m getting when it’s all assembled. I assume it’s a basic error in my approach that I’m compounding at some stage.
0 likes • Mar 31
@Koren Pollak I’ve built a bunch of versions of it and continue to iterate. Part of the problem is what I want to create. After building basic content slides with bullets, it seemed like the only thing I could reliably output was still just variations of bullets on some background. I integrated nano banana so it would generate the images and make the slides more interesting, but still just felt like bullets. Now I’m using NotebookLM to generate the slides and get much richer output. The pipeline starts with a review of the materials/codebase, then generates core messaging and pillars, then creates a full BOM and launch plan. From that it starts to generate all the materials (slides and word/pdf content) No repo to share at this point, sorry
1-4 of 4
M H
3
21points to level up
@43212898
don’t know why you need this and it’s not great that you ask for someone to do this as part of the setup. I just want to see what this is about first.

Active 10h ago
Joined Mar 8, 2026
Powered by