Been using Claude Code's handoff long enough that it starts to feel uncanny โ every iteration is like another AI waking up in Westworld.
It comes back carrying the memories we left it, but somewhere along the way it always quietly loses something that mattered.
That's why I go with /compact. Handing off is one of those things where the person leaving always knows best what the next one will need.
Maybe the better question is: why use all these other agents when you've already got Codex or Claude Code?
A new agent really needs to bring something to the table to make people want to switch.
Pretty much โ for devs it's CC or Codex, for everyone else it's OpenClaw or Hermes. The entry point collapses to one app, it's just not the same app for everyone.
Good news: my Harness Engineering is done.
Bad news: it's a beast now. Super heavy, super complex.
The thing is, I could go a lot faster by running stuff in parallel. But I keep chickening out. Worried it'll fry my brain and I'll lose the plot.
TikTok used to drive me nuts. Then it hit me โ maybe I was just watching it wrong.
Now I scroll right past:
Too many gimmicks? Skip.
"Let me share 3 secrets with you" โ skip.
"You're all doing it wrong" โ skip.
"Did you knowโฆ?" โ skip.
Edited too fast โ skip.
Too polished, too flashy โ skip.
A few weeks of this and my whole feed changed. No more hooks yelling at me, no more fake urgency.
I open the app andโฆ nothing's trying to grab me by the throat. Just calm.
Turns out the algorithm wasn't the problem โ I was feeding it the wrong stuff. Now it finally feels like mine.
The /btw command in Claude Code is seriously underrated.
What makes it special is that it doesn't disrupt your main thread. Take prompt optimizationโa classic use caseโit's a perfect fit for /btw.
I've actually made this a habit, especially during reviews. Instead of firing off a review command to CC directly, I:
Send my initial requirements to /btw
Let it generate a complete prompt
Pass that prompt back to CC
The biggest win? The prompts come out thorough and well-formed.
If Skills is the brain and muscle memory,
Obsidian is the training groundโ
A space for free exploration, chaotic combinations, and personal breakthroughs.
One structures. One liberates. Both evolve.
Skills + Obsidian: A Dual-Layer Architecture for Knowledge Management
From Evernote hoarding โ Notion databases โ AI summaries โ none solved the core problem.
Here's how I finally cracked it. ๐งต
The workflow:
1. AI assists reading (summarize, explain, question)
2. Insights crystallize into Skills
3. Skills immediately apply to the NEXT task
Knowledge stops being static. It becomes operational.