Andrey

189 posts

Andrey banner
Andrey

Andrey

@onsails

Angel Investor. @the_HaqqNetwork Chief Tech. Patient and curious.

शामिल हुए Eylül 2012
2.4K फ़ॉलोइंग1.1K फ़ॉलोवर्स
Sterling Crispin 🕊️
Sterling Crispin 🕊️@sterlingcrispin·
Claude 4.6 is a good programmer but writes insanely severe bugs constantly, it won't catch them all in audits, nor will other claudes You need codex 5.4 auditing every commit 4+ times. If you don't believe me, try it. I have an /auditcodex skill for it github.com/sterlingcrispi…
Sterling Crispin 🕊️ tweet media
English
183
102
1.9K
239.3K
Andrey
Andrey@onsails·
@hasantoxr Now implement cron and channels and go after openclaw
English
0
0
0
86
Hasan Toor
Hasan Toor@hasantoxr·
LangChain just open-sourced a replica of Claude Code. It’s an MIT-licensed framework that recreates the core workflow behind coding agents like Claude Code but in an open system developers can inspect and modify. It is called Deep Agents. I spent a bit of time looking through the repo and it’s actually a pretty helpful reference if you’re trying to understand how these coding agents are structured. Here's what's inside: → Planning tools for breaking down tasks → File system access for reading, writing, and editing code → Shell command execution with sandboxing → Sub-agents for handling complex work in parallel → Auto-summarization when context gets too long Another useful aspect is that it’s model-agnostic, so you can plug in different LLMs and experiment with building your own coding agents on top of the same structure. If you’re exploring agent frameworks or just curious how tools like Claude Code work under the hood, this is a pretty good repo to bookmark. Link in the comments.
Hasan Toor tweet media
English
73
304
1.5K
173.3K
Andrey
Andrey@onsails·
@ashpreetbedi I'm using Agno to build something similar but multi-agent. Might consider replacing my own agents with Pal. How to extend Pal without forking it? Add mcps, tools etc.
English
0
0
1
633
Andrey
Andrey@onsails·
@Shimarble2050 That's a great way of putting my longread in a single tweet without "read more"! Be a human, indeed. Algorithms are for computers.
English
0
0
1
15
Shimarble
Shimarble@Shimarble2050·
@onsails bizarre world. write for the sake of writing and sharing thoughts, not the algo nor online persona. I’m human before I’m a profile… online is a place I visit not where I live!
English
1
0
1
22
Andrey
Andrey@onsails·
It bothers me a lot, that everyone says you need to be consistent with social media presence. Think about it: your relevance depends on you feeding the algorithm. Consistently. It's a tax you pay to the social media for the right to be alive in the noosphere. Our posts become a digital heartbeat. Break that beat – and you are out of the scene. What happened to individualism, the main human achievement of 20th century? It didn't disappear. It got repackaged into a "personal brand". You are marching together, thinking about the next post while writing today's one. You are convinced that you express yourself while all you do is just putting your feet to the ground at the same tempo as everyone else does. Ironically, that's the opposite of individualism, personality or genuine human authenticity. You dilute yourself in the mass of "public opinion" while "the public" forgets about the post you published yesterday, being hungry for today stories. Publishing is important! It had always pushed human nature to the new limits since the invention of typography. But don't be a sheep walked to a pasture by a social norm every morning. Be an anomaly. Never in human history a great man followed the common sense. Wishing you a great Friday night. If this tweet gets zero engagement, thats not a bug, thats the thesis.
English
1
0
2
100
Andrey
Andrey@onsails·
@shaundadevens HIP-4 will also contribute to perps via arb and hedge
English
0
0
2
401
shaunda devens
shaunda devens@shaundadevens·
Even if HIP-4 captured 100% of Polymarket's volume, the line item would add just ~5% to HL revenue. Believe this highlights that: a) The scale of perp markets (including HIP-3) is massive. b) HL is undervalued at $7B vs Polymarkets $10b recent round.
shaunda devens tweet media
English
47
45
623
47.1K
Andrey
Andrey@onsails·
code-simplifier integration: If @anthropic's official plugin is installed, review-loop runs it first — specifically on files changed between your branch and target. Cleaner code going in = fewer review cycles = faster merges. claude plugin install code-simplifier@anthropic-official
English
1
0
1
45
Andrey
Andrey@onsails·
review-loop v3.8.0 🔄 Two big changes: → Migrated to Claude's new Task tools — proper dependency tracking between review iterations instead of flat todos → Optional code-simplifier pre-pass on your branch diff before reviews begin github.com/onsails/cc
Andrey tweet media
English
1
0
2
193
Andrey
Andrey@onsails·
Rewrote my review-loop (wiggium-style) using new task system. It became much more consistent now when I can leverage blocking tasks! github.com/onsails/cc Also before there was just one TodoWrite tool which writes all tasks at once, so in order to report them on the fly, claude needed much more context. Now it’s more lean.
English
0
0
0
429
Alex Finn
Alex Finn@AlexFinn·
And just like that Ralph Wiggum is dead Claude Code can now create its own project tasks and manage itself This is the next step towards Claude being a 24/7 autonomous agent Lesson from this: spend more time on the planning phase. Have Claude build as many detailed tasks as it can. The more time you spend on this, the more time you'll save later having to prompt Claude, because it will just be able to manage itself for hours
Thariq@trq212

x.com/i/article/2014…

English
187
165
3.3K
497.2K
Andrey
Andrey@onsails·
Also detects your target branch automatically. Checks for existing PR first, falls back to git history. Never assumes master/main.
English
1
0
0
33
Andrey
Andrey@onsails·
Built a Claude Code skill that reviews your code in a loop until nothing's broken. Not one pass. Minimum 4 iterations. Here's why that matters:
English
6
0
3
162
Andrey
Andrey@onsails·
Spending 16 hours per week in Claude Code planning architecture is worth more than 100 AI-authored commits per day. Consider tokens as investments rather than fuel.
English
0
0
5
149
Andrey
Andrey@onsails·
@claude_code Control what's shared: small config to include or ignore specific items globally or per project. Keep one source of truth for the team. Repo & install instructions: github.com/onsails/ccsync
English
0
0
2
93
Andrey
Andrey@onsails·
@claude_code What it gives: share your tuned setup to any project in seconds, see changes before applying, approve fast, avoid copy-paste. Two-way flow keeps everyone aligned: project improvements come back to your personal setup.
English
1
0
3
109
Andrey
Andrey@onsails·
I mentor my teams/partners to adopt agentic coding with @claude_code. Many tried AI coding and stopped – default setup isn't enough. My tuned agents/skills/commands 10x dev experience. Problem: they live in my personal setup, but projects need them too – and team edits should come back. Solution: ccsync – two-way sync personal ↔️ project.
Andrey tweet media
English
1
0
7
241