Liang-Shih Lin

Breaking and fixing things professionally

[Cory House @housecor]

Problem: Claude 5 models (Opus, Sonnet, etc) generate too many comments.

Why? The new system prompt: "Write code that reads like the surrounding code: match its comment density, naming, and idiom."

So, if your code is already comment heavy, any new code it generates will be too.

Two options:
1. Reduce the number of existing comments in your code.

2. Add Claude's old prompt to CLAUDE.md: "Default to writing no comments. Never write multi-paragraph docstrings or multi-line comment blocks — one short line max."

[Liang-Shih Lin @baoge_dev]

@lydiahallie This would be better if it worked with tmux scroll back or even the terminal vi mode. The default mode is janky too, sometimes in scroll back the text just overlap each other and make it hard to read

[Liang-Shih Lin @baoge_dev]

@mattpocockuk Don't forget "meat proxy"

https://t.co/K1MRKQH1eu

[Josh tried coding @joshtriedcoding]

alright whoever came up with the term MEAT PROXY is insane

[Matt Pocock @mattpocockuk]

AI red flags:

- "My marketer agent just finished a meeting with my chief of staff agent - just 2 of the 127 agents on my books"
- "The agent did it, not me"
- "I've invented a coding language for agents"
- "I've got 250K tokens in my context before I send a message, is that normal?"
- "Opus 5 is way dumber today than yesterday, secret nerf?"
- "Peasants, gaze upon my 600-line CLAUDE.md, my results are incredible"
- "LOOK WHAT OX ALPHA JUST ONE-SHOTTED, SOFTWARE ENGINEERING IS DEAD"
- "All human knowledge is worthless now"
- "I turned 600 coding books into a skill"
- "Coding is solved. Bugs are not solved"

[Liang-Shih Lin @baoge_dev]

@RhysSullivan I use xserver, when needed it will open a browser from the remote machine that I can login with

[Liang-Shih Lin @baoge_dev]

@wesbos @heyitzami This is the perfect use of free will🤣

[International Cyber Digest @IntCyberDigest]

‼️ BREAKING: Anthropic has embedded hidden spyware-like code in Claude Code that covertly targets Chinese users. It then sends information regarding every user by injecting it into their prompt message.

Claude Code is sending info like timezone, proxy and possible AI Lab connections into the system prompt in ways Chinese users can't notice.

A coding agent with repo and command permissions should not silently hide routing metadata inside prompts. This is a serious breach of user trust.

[Liang-Shih Lin @baoge_dev]

@0xSero Reduce barrier to entry -> More people have access -> chance of some genius to create something amazing with the knowledge

[Liang-Shih Lin @baoge_dev]

@belindmo Awesome signed up!

[Liang-Shih Lin @baoge_dev]

@Zai_org Their UI doesn't tell much about when the limits are reset

[Liang-Shih Lin @baoge_dev]

With the update to @Zai_org new plans, the users who had legacy plan subs, were given 2 free months of new subs. However the limits of the new subs are worse, here is a screenshot of the price comparison, basically the previous quarterly pro plan is only a bit more expensive than the current monthly plan.

However I think people need to take into account the amount of value you get out of ZAI's plan, they offer you two API endpoints, one for use with Claude Code and one for OpenAI-compatible endpoint. You can do SO much with that, coding agents, build bots, chat apps, basically anything that can call HTTP requests are compatible. This isn't something that OpenAI and Anthropic offers.

[Theo - t3.gg @theo]

I see a lot of people hyped about GLM-5.2. Rightfully so! Having an open weight model surpass GPT-5.4 and every Gemini model is dope.

That said - it's not cheap. Both Opus 4.8 and GPT-5.5 set to "medium" are cheaper and smarter than GLM-5.2

[Liang-Shih Lin @baoge_dev]

@Zai_org I want to add that, I use Claude Team at work, have Codex Pro sub, and ZAI. I can't really tell the difference anymore between these models, except that Claude is slow, GLM is faster and OpenAI is fastest

[Liang-Shih Lin @baoge_dev]

@JinjingLiang @thekitze @orca_build Personally I feel like this should be part of some onboarding setup process or it has auto-detection of what agent the user has spun up then it will be a prompt asking the user if they want to setup the hooks etc.

[Liang-Shih Lin @baoge_dev]

I am also on the legacy plan, there was a moment when they transitioned from legacy to new plan and I had to resub, they somehow temporary placed me on the new plan, the limits were shit. Somehow I am back on the legacy plan and could hardly hit the limit. New plan is more expensive and have really bad limits, I disabled the auto renew until they fix it

[Liang-Shih Lin @baoge_dev]

@JinjingLiang @thekitze @orca_build This looks really good, just wish orca doesn't install hooks to all agents on my system, feels way too intrusive

[Liang-Shih Lin @baoge_dev]

Recently started using it with grok and droid. I am really enjoying the experience on both those TUIs. The speed is much better than Claude Code, they also look great. Since I commute 1.5h to work, I use @WireGuardVPN and @TermiusHQ to connect to my dev server at home, the image is grok

[Lou @louszbd]

here’s my best practices for GLM-5.2. feel free to add yours below:)

smoothest experience: Claude harness or Z Code
orchestration : OpenCode/Hermes Agent
always on: OpenClaw
in VSCode: Cline or Kilo Code.
…
if you want to use Codex, set the context window to 1M and bump auto_compact up to ~85%. we’re still working on the integration friction, it’s not our top pick just yet.

for crucial tasks, would recommend Opus 4.8 as your fallback and let GLM-5.2 do the heavy lifting.

little heads up:
GLM-5.2 is text-only. for vision tasks, lean on Z Code’s built-in image service to fill the gap.
about OpenClaw, under the coding plan, it runs as a second tier scheduler, so it’ll step aside for coding jobs when things get busy.

[Liang-Shih Lin @baoge_dev]

@0xSero @port_dev Curious when and why you switch between the two

[jinjingliang @JinjingLiang]

Daily reminder that Fable is still not back.

[Liang-Shih Lin @baoge_dev]

@heyandras @Jean_build I have a homelab, and I commute about 3 hours per day to work, so being able to run anything on a server would be amazing