I feel so lucky to be around successful entrepreneurs like @joypbuilds! In the Tech Startups discord server we create spaces for people like Joy to share their learnings when going viral on Korean Twitter.
Check us out and submit an introduction at https://t.co/gcfYUViAdp
@ranman@bcherny@techgirl1908@karpathy Are you using Beads as your memory layer? Curious to what other dependency graphs people are finding success with.
@bcherny@trq212@lydiahallie Are ya'll aware of how hard it is to get Claude Code to activate Skills?
AFAIK the entire point of skills is that it allows your agent to fetch context based on the current conversation, but its REALLY HARD to get them to activate. To the point where you basically have to explicitly tell them "go activate this skill", which at that point you might as well be using a slash command.
For example, the Anthropic provided frontend-design skill says "Use this skill when the user asks to build web components, pages, or applications" but its only activated if you explicitly tell it to run the skill!
It seems like the main Claude website doesn't have this problem because if I say "go create a document" it'll load the docx skill immediately. CC needs this fix imo.
I just created a skill of:
---
name: blog-knowledge
description: whenever you are working on anything related to my blog, such as files in apps/website/blog/ or the user mentions keywords such as "blog" "article" or anything similar, you MUST load this skill
---
<content>
but no matter what I do, I can't get it to activate when I work on my blog.
and I have only found two methods to get Claude Code to activate the skill on its own (without telling it explicitly):
1. Ask "What skills are available?" at the start of the session (and this is still pretty flaky)
2. a UserPromptSubmit hook that calls a bash script which injects custom instructions called "MANDATORY SKILL ACTIVATION SEQUENCE..." (221 tokens every time) which works great, but is extra context I don't want.
Scott Spence wrote an entire article where he set up a testing framework to evaluate different methods to fix this skill activation problem, worth a read: https://t.co/l07trYapoQ (screenshot attached)
@idanmasas Sorry to hear you're feeling behind in life, that sounds really uncomfortable.
I am really happy to hear that you're improving marketable skills like software engineering, that's worth $$.
It sounds like you're leveling up your intuition through these experiences.👍Keep it up.
Why do doctors offices get a pass for making you wait an hour for an appointment they would've canceled if you showed up 15 minutes late and charged you a fee?
Seeing tweets like this makes me so proud of https://t.co/gcfYUViAdp
Builders posting substantive updates each month, community centered around building, not wantrepreneurship.
@alldataffm @pmitu I felt the same way about this, so I have a discord community that is 8k strong where you must post substantive monthly updates to be considered a builder.
@idanmasas Imagine what lessons you'd learn by being in the trenches of OpenAI when they weren't worth anything, up until today.
You'd watch all of the small mistakes, and small victories that led to the conclusive we all see.
I'm working to scale an awesome discord community for indie hackers. We're 6,000+ strong, with no spammers or scammers. 😌
Check us out: https://t.co/gcfYUVj82X