Hacker News Reader

Comment by jjcm | Hacker News Reader

Comparison of with/without:

Without the skill: https://non.io/video/diffui-product-vid-opus-only.mp4

With the skill: https://non.io/video/diffui-product-vid-with-skill.mp4

Also a disclaimer that there are several major playwright CVEs, so be careful with this / patch the installs you do have.

View on HN

Can you look up how many tokens both consumed?

In some cases, the point of a skill is to handle all context to a model, so that they don't need to look up on their own (sometimes filling up the context window with irrelevant stuff in their search)

Input:

Skill-video-studio: 334k

Vanilla Opus: 203k

Cache read:

SVS: 29M

VO: 17M

Output:

SVS: 152k

VO: 93k

Overall vanilla seems to use less, not more, but that's with a sample size of 1.

they look almost exactly the same to me. Granted, they are mostly video of the same product, but still, whats the major difference?

That's kinda the point I was trying to show. No disrespect to the author of the skill, but I think it's always worth evaluating what something looks like without a skill, since models these days are really good at doing stuff on their own.

Opus 5.5 and Astra are already very good (because of being RL trained at it) most of these. The way I see skills useful are: 1)If they are created to teach how to use a custom and new CLI or MCP 2) They describe gotchas, tips and tricks that are hard to find using simple search or potentially save some tokens for easier to find ones. The OpenAI and Anthropic teams already said to reduce the skills otherwise as they can also waste context space unnecessarily.

the major difference is annoying music

The localhost on top of the skill generated one adds that special handcrafted element everyone loves. /s

Reply on news.ycombinator.com