Current stack:
- Pi for everything.
- GPT-5.4 for everything code.
- Gemini-3.1 for design/brainstorming.
- Sonnet 4.5-no-thinking for openclaw.
- GLM-5 for parallel swarms.
- Opus 4.6 for everything else.
Currently Testing: Minimax-2.7.
I've come to really enjoy my little "Mana: Claude Code Companion" utility. Built for MacOS, it floats on top of your windows and reacts to Claude Code and opencode activity via hooks. Open sourced here: github.com/dep/mana-claud…
Extracted content. Here you go 🤘
### 1. Plan Node Default
- Enter plan mode for ANY non-trivial task (3+ steps or architectural decisions)
- If something goes sideways, STOP and re-plan immediately - don't keep pushing
- Use plan mode for verification steps, not just building
- Write detailed specs upfront to reduce ambiguity
---
### 2. Subagent Strategy
- Use subagents liberally to keep main context window clean
- Offload research, exploration, and parallel analysis to subagents
- For complex problems, throw more compute at it via subagents
- One task per subagent for focused execution
---
### 3. Self-Improvement Loop
- After ANY correction from the user: update `tasks/lessons.md` with the pattern
- Write rules for yourself that prevent the same mistake
- Ruthlessly iterate on these lessons until mistake rate drops
- Review lessons at session start for relevant project
---
### 4. Verification Before Done
- Never mark a task complete without proving it works
- Diff behavior between main and your changes when relevant
- Ask yourself: "Would a staff engineer approve this?"
- Run tests, check logs, demonstrate correctness
---
### 5. Demand Elegance (Balanced)
- For non-trivial changes: pause and ask "is there a more elegant way?"
- If a fix feels hacky: "Knowing everything I know now, implement the elegant solution"
- Skip this for simple, obvious fixes - don't over-engineer
- Challenge your own work before presenting it
---
### 6. Autonomous Bug Fixing
- When given a bug report: just fix it. Don't ask for hand-holding
- Point at logs, errors, failing tests - then resolve them
- Zero context switching required from the user
- Go fix failing CI tests without being told how
---
## Task Management
1. **Plan First**: Write plan to `tasks/todo.md` with checkable items
2. **Verify Plan**: Check in before starting implementation
3. **Track Progress**: Mark items complete as you go
4. **Explain Changes**: High-level summary at each step
5. **Document Results**: Add review section to `tasks/todo.md`
6. **Capture Lessons**: Update `tasks/lessons.md` after corrections
---
## Core Principles
- **Simplicity First**: Make every change as simple as possible. Impact minimal code
- **No Laziness**: Find root causes. No temporary fixes. Senior developer standards
The guy who created Claude Code ( @bcherny ) recently leaked how his team uses Claude.
One CLAUDE.md that you drop into your project.
Inside: past errors, conventions, rules - Claude reads it every session.
Boris uses this every day at Anthropic:
The stars have aligned, and the reviews are calling Passport: Hollywood a box-office smash! 🍿
Ready to get in on the blockbuster fun? Our newest course is now available on VR platforms, so hop in and get ready for your close-up!
🍊 It's been 90 days for me at @Cloudflare ...
👇 Here's what has surprised me
• There's very little ceremony or overhead when it comes to shipping, everyone is here to build
• Cloudflare is extremely high agency — there's alignment but it's just enough and most of it exists to unblock folks
• Management is very technical up and down the hierarchy, every leader / manager here talks in specifics
• Executive tier is extremely engaged, in various channels, very responsive, knows the details
• Startup mentality is pervasive; I thought we'd slow down after the Replicate acquisition but we've sped up
👇 Here's what I love
• Community. Reminds me a lot of my favorite parts of GitHub. Folks here are superfans of one another's work. They want to meet & hang and we're collectively excited when someone new joins the club.
• The primitives that we get to build on are excellent -- all built to be distributed at massive scale by default
• Hardware! We source and build our own hardware!
• We dogfood everything, you build "on platform", you have good stat sig day one!
• Innovation is funded everywhere -- "what if we built this on " is just another day at the shop
• We have some folks around that seem to think different/understand the internet in ways that I don't think is all that common
👇 Here's why I'm long Cloudflare
• Tenure at Cloudflare is remarkably long as is the % of folks who have boomeranged back, both very bullish signals
• It's a sleeper hit in the developer platform space, yes WAF and yes Cloudflare One are big businesses but we are just getting started & we are cooking in developer platform
• Every team I talk to has something in the works that they're extremely excited to ship and then another thing right behind that
• When I ping folks in my network to talk about Cloudflare, they are immediately responsive
• The product has a lot of rough edges -- folks know, they're working on it -- it only gets better from here
But don't just take it from me! I asked three colleagues what their favorite things were, here's what I got.
• "The way Cloudflare empowers engineers at every level of the org to have huge impact"
• "You can basically work on anything at cf, talented engineers, infra moat so its not just software, cool research and innovation that you don't really get anywhere else"
• "Cloudflare has a pretty unique tech stack with a really powerful infrastructure. We're working on some really innovative things that no one else is doing. There is a lot to do -- pretty much anyone can see something needs to be done and go do it."
All of that said -- I might be dm-ing some of you soon to work on some really challenging MLE stuff here at Cloudflare soon. Answer me if this sounds like your kind of thing 😜
@paulo_kombucha I had a few generations fail or render a static image for the video. Would it be possible to be refunded some credits to account for those?
My new app is live!
Introducing masco.dev, a new way to build your own mascot, powered by VideoBGRemover
I’m also using it for my own product marketing, and I made some cool animations and logos with it
@dep You export the video with hevc .mov. It is already possible with Masko and then you can use it on MacOS. It is always a bit tricky to use and depends on the platform. In a MacOS, use AVPlayer configured for transparency
Readout 0.0.8 is live: new session transcript search tab, new tool usage tab, skill and agent customisation (add/edit/remove), cost projections, session handoffs, sidebar customisation, and most importantly, a cute new splash screen.
→ readout.org