This is your new model season reminder to have an agent:
1. Review your traces
2. Improve any skills, config files, and agents files pertaining to recent failures.
3. Remap how it orchestrates the models based on the new ones out, if at all.
1
32
And yet it did
teapawt.org/baney75/status/2121983…
If you told me that a model that was only 1.7B parameters was capable of producing something like this a week ago, I wouldn’t believe you.
1
28
If you told me that a model that was only 1.7B parameters was capable of producing something like this a week ago, I wouldn’t believe you.
The future is making skills for smaller local models to use, this will greatly cut costs and ensure your limits are only be used for your hardest tasks.
people sell clipping courses. Claude Code runs the whole workflow as /clip
VOD in, captioned 9:16 clips and copy.md out
it's 3 files in your project:
.claude/skills/clip/
> SKILL.md
> moments.md
> scripts/cut.py
[SKILL.md]
· the 8 steps in order: transcribe, pick, check stills, crop, cut, caption, copy, ffprobe
· disable-model-invocation: true, so it runs only when you type it
· allowed-tools: whisper, ffmpeg, ffprobe and python run without permission prompts
[moments.md]
· the prompt that picks 20-60 s moments and skips spoilers. loaded only when step 2 needs it
[scripts/cut.py]
· one ffmpeg call per moment: facecam on top, gameplay below, 1080x1920
then it's one line:
/clip vods/stream.mp4
all 3 files are in the replies
the skill is the shell. CLAUDE.md, the rules check and the earnings log that make it work are in my gta clipping pipeline guide ↓
Running both is amazing opus 5.5 to sol 6.1 I’m still using the credits from when they were going to make the cuts to limits.
I've improved Claude's output a lot by having it use Codex to review its job, then act on findings until Codex gives the green light.
I heard about that a month ago from @dhh here on X, but I was a bit skeptical if it would meaningfully improve outcomes.
I added a simple instruction on my global CLAUDE.md so it uses Codex for reviews, ignoring plain simple things to keep things fast. Same rule in reverse for Codex.
In the last week or so Claude used it 85 times, 69 of those with actual issues that were fixed.
If you only use Codex for reviews, a 20$ plan is enough so I really recommend this for anyone coding with AI.
Claude is a great engineer, but Codex is more autistic and better at PR reviews.
1
56
Local models have been called many poor names. LFM3-1.7B Designed this, look how far we have come. Local is the future. @huggingface @0xSero
Stand with Jensen.
#standwithjensen #bringjensenback standwithjensen.com/
The Righteous man Falls many times but gets Back Up.
5
97
We need to make an effort to make old gaming consoles run omarchy, and hyper optimize certain games from steam.
Omakade 1.19 is out for Omarchy, and it's the biggest update yet. 77 pull requests and 37 changes you may notice.
Game Mode now handles Steam with just a controller. Steam's first-launch prompts, like a EULA, show up on the Starting screen, and the controller turns into a pointer so you can accept them from the couch.
If Steam signs you out mid-session, Game Mode offers Restart Steam and usually gets you signed back in.
Switch games from Lichii are in the library now too, with their play time.
There's also a new Text size setting, clearer focus on every theme, big libraries that stay smooth while covers load, safer save restores, and an app that's 80 MB smaller.
github.com/btsouth/omakade
1
64
I made a skill so smaller models can create great 3D models. So much so that a 1.7B model went 39/40 on a blind test. Wow!
100% Omarchy is great on old hardware, save your old computers from the dump!
Recently setup @omarchy on a decade+ old home Mac that moved at a glacial speed despite deleting all data, apps and fresh macOS.
Still navigating WiFi drivers, but overall I cannot believe it's the same machine. Feels light, fast and breezy, + installation itself took <5 mins!
3
94
I do feel like a lot of the work being done is repetitive, but as models approve I think skills will matter less and less and become more and more personal preference.
371 new GitHub repos are tagged claude-code.
What are they?
68% extend it: skills, plugins, MCP servers, hooks
13% wrap it in UIs and session tools
6% configs and templates
11% are projects built with it
Developers are building around the tool, not with it.
Check the blog for more details 👇👇👇👇👇
34













