Sabitlenmiş Tweet
Hung Vinh
335 posts
Hung Vinh
@hungv47
Making slop before slop was slop.
Buon Ma Thuot, Vietnam Katılım Ağustos 2015
574 Takip Edilen144 Takipçiler
@vercel_dev @wc26bot Isn’t this really expensive to do unless it’s temporary? Perplexity did this and couldn’t afford it
Grok works well because it ties to the user’s premium account quota
English
Chat SDK added an 𝕏 API adapter.
We built an eve agent to predict World Cup matches (repo in agent bio).
⚽ @wc26bot who has better odds, 🇦🇷 or 🏴?
English
@cramforce @vercel_dev @wc26bot @rauchg Isn’t this really expensive to do unless it’s temporary? Perplexity did this and couldn’t afford it
Grok works well because it ties to the user’s premium account quota
English
Downside of a free community 😕
Most of these are just "Thanks!" replies which don't require any action, but they hide genuine messages from my paid Maker School community
Need to sort out a way to manage this
English
There needs to be a standard for plugins! @vercel_dev
Hung Vinh@hungv47
Sad that @vercel has (seems to) abandoned the Open Plugins standard and instead went in for Skills. A plugin is a higher level abstraction. It bundles skills, MCPs, hooks, commands, configs, everything. I don't want to think about assembling MCPs and skills myself. I want to install one plugin and have everything work. 𝚗𝚙𝚡 𝚙𝚕𝚞𝚐𝚒𝚗𝚜 𝚊𝚍𝚍 feels like 𝚗𝚙𝚖 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 for agent capabilities, and it should be standardized. OpenAI is embracing it, same with Anthropic and xAI. If you hate the idea of bundling, you can flexibly use the Skills CLI to install individual skills, or @executorsh / mcpm CLI to install MCPs. For the most part I just want to take the whole thing, especially for vendors. You can install a skills repo with plugins CLI and it works fine, not vice versa (case in point: @impeccable_ai) The plugins package is still live on NPM with 10k weekly downloads, the website is still live on Vercel, but all related repos got removed. Last update was 3 months ago, and the CLI doesn’t seem to work anymore. The existing Codex plugins use the key format github.com--@plugins-cli (e.g., github.com-openai-skills@plugins-cli). But the plugins CLI v1.3.1 uses just plugin.name@plugins-cli (e.g., skills@plugins-cli), derived from the manifest or directory name. Codex can't match these new entries.
English
Hey David, just take a look at the Junior repo, it’s very interesting.
Can you write up a technical article or something about this? Maybe expand from this article: [x.com/zeeg/status/20…)
Wanted to know your view on MCP vs CLI pairing with skills if possible
Also, what do you think about new agent frameworks like evedotdev compared to Junior?
Your articles are amazing, thanks man!
David Cramer@zeeg
English
this def aint podcast vibes.. gotta figure out if i use a diff room (ugh) or tweak how id use this one
English
Hung Vinh retweetledi
"hard questions are great but only if we deem you worthy enough to not silently downgrade you, or even get access at all"
English
Sad that @vercel has (seems to) abandoned the Open Plugins standard and instead went in for Skills.
A plugin is a higher level abstraction. It bundles skills, MCPs, hooks, commands, configs, everything.
I don't want to think about assembling MCPs and skills myself. I want to install one plugin and have everything work.
𝚗𝚙𝚡 𝚙𝚕𝚞𝚐𝚒𝚗𝚜 𝚊𝚍𝚍 feels like 𝚗𝚙𝚖 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 for agent capabilities, and it should be standardized.
OpenAI is embracing it, same with Anthropic and xAI.
If you hate the idea of bundling, you can flexibly use the Skills CLI to install individual skills, or @executorsh / mcpm CLI to install MCPs.
For the most part I just want to take the whole thing, especially for vendors.
You can install a skills repo with plugins CLI and it works fine, not vice versa (case in point: @impeccable_ai)
The plugins package is still live on NPM with 10k weekly downloads, the website is still live on Vercel, but all related repos got removed.
Last update was 3 months ago, and the CLI doesn’t seem to work anymore.
The existing Codex plugins use the key format github.com--@plugins-cli (e.g., github.com-openai-skills@plugins-cli).
But the plugins CLI v1.3.1 uses just plugin.name@plugins-cli (e.g., skills@plugins-cli), derived from the manifest or directory name. Codex can't match these new entries.
English
There's a few issues with Codex that stacked really badly, causing people's usage to get nuked. Crazy how well they all lined up to be as rough as possible
Issue 1 (shown here): gpt-5.6 costs 2x more after 272k tokens. Codex upped their limit to 372k tokens, meaning long threads were billed 2x higher
Issue 2: "Ultra" subagents were also spawned with Ultra, causing insane nested subagent usage that burned a TON of tokens. Apparently intended, but (hopefully) being toned down in system prompt
Issue 3: Sol and Terra use the "v2" subagent layer in Codex that is still early/unfinished/disabled by default. v1 spawned subagents with a fresh history, v2 copies the entire long context.
When combined, you end up with:
1. A ton of reasoning tokens filling context windows, triggering 2x billing
2. Subagents spawning with that filled context window, instantly billed at 2x
3. Ultra spawning too many of these high reasoning full-window subagents
If you had fast mode on when this happened, you got hit with an ADDITIONAL 2.5x
No wonder we were burning so hard...
Tibo@thsottiaux
Updates for Codex and ChatGPT Work users. No nerfing, only good stuff! - We have landed inference optimizations and are passing down savings to all the subscriptions for GPT-5.6 Sol. That should result in around 10% more usage on its own. - We noticed that by changing the context size limit in the product to 372k for GPT-5.6 Sol, up from 272k for GPT-5.5, it resulted in more usage being charged than intended. We have reverted to 272k and will work to roll back out to 372k in the days to come. You should notice that usage drains significantly less after this change. - To understand where the extra usage was coming from, we ran some experiments where reasoning efforts were changed (referred to as juice values under the hood) and have reverted this. - There is slightly more usage of multi-agent than intended in high and xhigh reasoning effort, we are fixing this going forward. Also fixing a small other thing we noticed with auto-review where we can be more efficient. And we continue to have the 5h limit temporarily not apply. Enjoy the rest of the weekend!
English
Updates for Codex and ChatGPT Work users. No nerfing, only good stuff!
- We have landed inference optimizations and are passing down savings to all the subscriptions for GPT-5.6 Sol. That should result in around 10% more usage on its own.
- We noticed that by changing the context size limit in the product to 372k for GPT-5.6 Sol, up from 272k for GPT-5.5, it resulted in more usage being charged than intended. We have reverted to 272k and will work to roll back out to 372k in the days to come. You should notice that usage drains significantly less after this change.
- To understand where the extra usage was coming from, we ran some experiments where reasoning efforts were changed (referred to as juice values under the hood) and have reverted this.
- There is slightly more usage of multi-agent than intended in high and xhigh reasoning effort, we are fixing this going forward. Also fixing a small other thing we noticed with auto-review where we can be more efficient.
And we continue to have the 5h limit temporarily not apply. Enjoy the rest of the weekend!
Fix@FixlationAI
OpenAI has reduced GPT-5.6 Sol's thinking budgets in an effort to make the model more efficient They essentially bumped everyone's reasoning down by 1... so if you were running Sol Extra High, you now have to set it to Max to get the same effort So we basically don't have Max reasoning anymore, how do you feel about these changes? 🤔
English
gnhf is a ralph, autoresearch-style orchestrator that keeps your agents running while you sleep - each iteration makes one small, committed, documented change towards an objective. You wake up to a branch full of clean work and a log of everything that happened.
github.com/kunchenguid/gn…
English
integrations.sh is so fast and convenient, thanks @RhysSullivan, but can we have breadth search like skills.sh?
English
@maria_rcks This is more of a harness problem, no? The caching is crazy
English
has no one built a “continue” bench yet? take a coding benchmark like DeepSWE or whatever, then interrupt the agent while it’s running with messages like “continue,” “keep going,” etc. and see whether it affects the final score.
if the results are damning enough, coding tools might need a proper way to resume agents without polluting the context window.
English
@complex_maths @theo Yea the behavior is hostile, I just think he’s mad for losing Bun. Didn’t see anything more than that.
English
I think it was the part where he wrote a blog post shitting on Jarred and tries to palm it off as “nothing personal at the end”. He even updated it getting pissy at people pointing out that it’s clearly a very personal and petty attack on Jarred, rather than the bun project as a whole.
English
Read this on stream earlier. Big mistake. I've never been so pissed off on camera.
I cannot fathom how anyone could publish an article like this. Andrew is a petty, evil man, and you should not trust any software he writes.
rohit@seatedro
crazy blogpost by andrew kelley just dropped
English