Hacker Newsnew | past | comments | ask | show | jobs | submit | kurtextrem's commentslogin

I found this prompt works well to nudge it to use a better grep as the start, then just keep using grep (Cursors instant grep in my case):

``` - For planning, prefer using morph-mcp `codebase_search` - subagent that takes in a search string and tries to find relevant context. Best practice is to use it at the beginning of codebase explorations to fast track finding relevant files/lines. Do not use it to pin point keywords, but use it for broader semantic queries. "Find the XYZ flow", "How does XYZ work", "Where is XYZ handled?", "Where is <error message> coming from?" ```

(see also https://news.ycombinator.com/item?id=48205911; having higher quality results at the beginning of a thread seem to improve the output vs. having faster search later on).


Have you tried https://github.com/mksglu/context-mode? I prefer it over rtk / headroom (at least in Cursor). The stop hooks seem to work well / prevent the agent from blowing up its context window.

Related read (not from me): https://entire.io/blog/improving-agentic-search-in-coding-ag...

> The clearest result was that faster search alone only modestly helps, while better-ranked results improve first-query retrieval and help agents find the right code sooner.

Their tool "pgr" is a research preview only, so it'd be interesting to see semble vs pgr.

I'm also collecting other tools that are similar, most notably is probably Morph's WarpGrep (has a free tier too). Apart from that, there is codemogger (https://github.com/glommer/codemogger), cs (the author also commented in this HN post).

In the similar area, but not fully related, the author of fff is also pretty involved in any thread that goes into that direction (see e.g. https://x.com/neogoose_btw/status/2052161471296225710). Similar to colGREP is also mgrep (by mixedbread) and osgrep (but they seem to predate colGREP). I also found codedb on X (https://codegraff.com/blog/codedb-code-intelligence), the post reads well, but haven't tried.


Just stopping by to say this is great usage of the new APIs.


How does it compare to jpegli?


tau-coder produces about 10% smaller output for the same VMAF quality, compared to jpegli. The above evaluation link contains evaluation directly against jpegli


Also, the vite team in collab with a few others is building https://rolldown.rs/, to replace esbuild and rollup in vite. It's goal is to be faster than esbuild, with extended chunking options and so on.


I'm not entirely sure if we can really tell anything about esbuild from that comparison, as vite's production build time is 1300ms (which uses rollup), but dev startup time 1100 (uses esbuild to prebundle). It seems like vite itself has overhead.

The only bench I'm aware of was presented in November 2023: https://x.com/boshen_c/status/1719596594985681275?t=x8FaB9Aw..., where esbuild was faster.


"xylitol-sweetened (30 g) water, an exposure comparable with a pint of numerous xylitol-sweetened ice creams, a xylitol-sweetened bakery good, or several pieces of xylitol-sweetened candy" (source: https://www.reddit.com/r/science/comments/1d9ql2u/comment/l7...)


Anything for Nancy pelosi?


Really cool stuff!


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: