Unclecode (Hossein)

🎉 Crawl4AI 0.7.7 is live, Self Hosting is the KING and QUEEN 🤴 🫅 Self hosting just leveled up. This release turns @crawl4ai into a real platform, not a library. You get a monitoring dashboard baked into Docker, open slash dashboard and you see the entire system breathing in real time, CPU, memory, active requests, browser pool state, errors, cleanup cycles, all streaming every two seconds over WebSocket. There is a full Monitor API now, everything exposed as simple REST. Trigger cleanup, kill frozen browsers, pull performance metrics, plug into Prometheus, wire it into your own infra without fighting the black box problem that every crawler suffers from. The smart browser pool went through a serious upgrade. Permanent, hot, cold pools that actually learn your usage pattern and cut memory use by an order of magnitude. Infra feels lighter, faster, cleaner. Critical fixes landed too, async extraction stalls, viewport quirks, DFS traversal edge cases, all hammered down. This release closes a loop. Crawl4AI now behaves like a self hosting engine with proper observability, privacy first, predictable cost, and operational clarity. You get the power without the enterprise tax. Next steps stay ambitious, hosted API, distributed crawlers, self contained binaries, and a path toward cheap AI infra for everyone. This version is a foundation, not a peak. Massive respect to the 56k plus who starred the project, shipped patches, tested weird edge cases, and supported the vision. OSS only works when the community actually shows up. 🔗 GitHub: github.com/unclecode/craw… 💬 Discord: discord.gg/jP8KfhDhyN 📖 Release Notes: github.com/unclecode/craw… Special thanks to sponsors @am_tambe, @vinayprabhu, @capsolver And to the individual supporters Martin Sjöborg Romek Rozen Kourosh Kiyani Max Bodewes Everyone who filed a bug, fixed a line, or pushed the platform forward left fingerprints on this release. You made 0.7.7 real.

I put a lot of heart into my technical writing, I hope it's useful to you all. 📌 Here's a pinned thread of everything I've written. (much of this will be posted on the Claude blog soon as well)

@trq212 You are one of the very very few people I enabled the notifications to catch up with ur articles and posts. I enjoyed them, specially lesson learned from building claude code, promot caching, and file system. Good job 👏 keep it up plz

Amazing, this is big! Now CC is a real assistant, finally we can send/receive events to/from a running CC! That make a CC run for 24 hrs in a two-ways interaction! I call it “agent-server” or in this case claude-server! Need to build a management layer top of this, spawn multiple CCs with channel enabled, and … These running CCs can become the ai assistant of any running apps! Every app easily has CC powered agent! I have to cook something for crawl4ai with this

We just released Claude Code channels, which allows you to control your Claude Code session through select MCPs, starting with Telegram and Discord. Use this to message Claude Code directly from your phone.

As of now, do we have ability to unload a skill knowledge from context when we dont need that skill anymore? I understand this disturbing promot caching probably, but i think this worth of the cost. If this is possible then context is shrinking dynamically when a skill is not needed and basically skills become like an OS native processes. This allows infinite skills, garbage out unused skill, and many more things…

Keep them all in environment variables, then in skill simply refer to them using their name, or let skill knows it can use “cat” un linux to read then when it needs then. U can manage ur secrets in .env or .env.skills which u can tell skill to load em first, or keep it in os env variables, this is best specially when u run ur app in a container where all secrets already passed to container and they are available at os env variables level

@trq212 How to handle sensitive keys in Skills though?

@heyNaitik @trq212 Cron job is better, it doesn’t trigger or add latency to current CC, I even made a script for this, uses haiku and in result it manage the skills folder, it does many things, including detect unused skills

@trq212 One of my fav hacks has been to give skills a “last used” date, and have my Claudemd file audit all unused skills every 2 weeks. Keeps the skills list fresh and prevents it from bloating or overlapping skills!

👀

@trq212 A quick question, a named session, u make a plan and approve to start it in a new session, is this a total new session (so I can resume to both separately) or overriding the previous one?

A few end of week ships: You can now set effort to 'max' which reasons for longer and uses as many tokens as needed. This will spend your usage limits more quickly so you have to activate it per session. Hit /effort to try it.

1 million context window: Now generally available for Claude Opus 4.6 and Claude Sonnet 4.6.

@bcherny Is this available in agent sdk?

Released today: /loop /loop is a powerful new way to schedule recurring tasks, for up to 3 days at a time eg. “/loop babysit all my PRs. Auto-fix build issues and when comments come in, use a worktree agent to fix them” eg. “/loop every morning use the Slack MCP to give me a summary of top posts I was tagged in” Let us know what you think!

@bcherny @vasuman Alsp not forget ultrathink command is back

👋 we recently changed the default effort to medium to better balance intelligence with speed, based on median usage we were seeing in the wild. You can change it anytime in the /model selector if you prefer low effort (faster) or high effort (more intelligence). The setting is sticky and will persist for your next session. If you’ve changed effort to high already, can you double check you’re on the latest version, and that your CLAUDE.md’s (in /memory) don’t have confusing/conflicting instructions? That’s the most common cause. [deleted a previous post where I accidentally said default is high, it’s early and I haven’t had coffee..]

They lobotomized Claude in Claude Code, probably because they're losing money on every call "Ads are coming but not to Claude" I don't give a shit. Bring ads to Claude if it means I get to spam 200 IQ Opus again

I see this post went viral, however This chart shows job postings, not actual hiring. What kind of SWE jobs are these? • salary levels? • seniority? • actual responsibilities? AI tools like Claude Code may simply allow companies to hire cheaper juniors who supervise generated code. That can increase the number of postings while lowering the cost per role. Tasks that previously required experienced engineers may now be done by more junior workers using AI tools. So quantity may rise while total payroll stays flat, or even drops. In other words, more jobs posted does not necessarily mean stronger demand for high-skill engineers or higher wages. Without data on compensation, seniority, and actual responsibilities, this chart proves very little.

@JohnnyNel_ These r new skills software engineers should practice. Like we set up all our rules, restrictions and guardrails as you mentioned, then we start development with tools such as CC. I see in close future we teach such things in universities as main core courses!

@unclecode Claude reorganized my entire database at 2am once... now I always set guardrails before AI touches production code.

If you use claude code on infra and want to be in the yolo mode, which u shouldn’t, at least PLZ USE HOOKS!! Specifically PreToolUse which fires before any tool call (Bash, Edit, Write). Hook receives event data, and you control what happens or NOT happens next via exit codes.

This chart lacks the data needed for these conclusions. Are these actually SWE roles, or people checking Claude Code output? Are they juniors companies now hire cheap because AI tools let them extract more work per person? The spike shows no salary info. We don't know if demand increased because companies found cheaper labor, or if actual SWE income went up. Need: average salary, skill requirements, whether ads demand real engineering or just Claude Code operation. Due lack of data one take is: this spike is people who couldn't code before, still can't, but can now build and ship with Claude Code. That demand evaporates once Claude Code handles security, testing, and deployment automatically. Few years max.

Citadel Securities published this graph showing a strange phenomenon. Job postings for software engineers are actually seeing a massive spike. Classic example of the Jevons paradox. When AI makes coding cheaper, companies actually may need a lot more software engineers, not fewer. When software is cheaper to build, companies naturally want to build a lot more of it. Businesses are now putting software into industries and tools where it was simply too expensive before. --- Chart from citadelsecurities .com/news-and-insights/2026-global-intelligence-crisis/

@lydiahallie Done ✅ for Singapore ans Malaysia 🤞🤞 btw I received the OSS support many thx 🙏

Want to host Claude meetups in your city? We'll cover the funding, send swag, and give you monthly API credits for your demos. You also get access to pre-release features and a private slack with the team! Go apply 💛

@claudeai Done ✅ for Singapore and Malaysia 🤞🤞

We're launching Claude Community Ambassadors. Lead local meetups, bring builders together, and partner with our team. Open to any background, anywhere in the world. Apply: claude.com/community/amba…

Multiple times a day, I need this and instead I do temporary copy/paste then double escapes type the new task! I just wish it doesn’t restore automatically, till I press ctrl + S again, becoz this remove the whole benefit, I may wanna continue conversation a few more turns then restore the stashed prompt.

Claude Code Tip of the Day: Prompt Stashing Long prompt, short detour? Control + S stashes your draft so you can fire off a quick question first. Your stashed prompt restores automatically after you submit.

GPT-5.4 + Codex just dropped and many developers ask: “Should I move from Claude Code to Codex?” My answer: don’t migrate. Use both. After 2 years using Claude Code to maintain Crawl4AI (61k⭐) and now adding Codex, I found a workflow where they amplify each other: Claude thinks. Codex executes. Claude attacks the result. Full workflow ↓