Good point, in this case I didn’t 😅 But since the issues were all on public repos I could have used the CLI without authenticating

This is the great part of coming back to a side project after 4 months of nothing

Wow, that's actually quite a bit worse than what I've seen people post. Could this be the smaller 2.5 Flash model instead of 3 Pro?

I still feel like CLIs are not the right UX even for most SWEs 🤔

Claude Code has a web interface though!

Hello Paige, it’s great to have you here 😊

Does it know about options that were explored but ultimately decided against? I saw that deciduous supports this but I was wondering how it would know based on commit messages

That's it!

Repo with usage instructions and more implementation details is here: github.com/nimobeeren/s...

Also: words that appear in the official solution only ever appear in one place in the puzzle grid. Unfortunately we can't filter them out upfront because they could still be part of the spangram, which can be a concatenation of words (though the full spangram never appears more than once either).

I also found some interesting "hidden rules" about the game. Despite being legal, you never _need_ a strand to cross itself or another strand.

Example: the word LIME in this image from the tutorial would never actually appear in a solution, because it requires crossing itself.

4. Rank solutions found in phase 3 by semantic similarity between words in the solution and the theme. Compute embeddings for all words and the theme, then score each solution by the average pairwise cosine similarity between its words and the theme.

3. Filter covers found in phase 2 to those containing a spangram. If the cover has more strands than the word count specified in the puzzle, try concatenating adjacent strands to form the spangram.

2. Find all ways to cover all cells of the grid exactly once using a subset of the words found in phase 1. This is an exact cover problem solved with a backtracking algorithm that uses the MRV (Minimum Remaining Values) heuristic: always branch on the cell with the fewest covering strands.

1. Find all strands in the grid that spell dictionary words. Starting from each cell, recursively take a step in all directions (using DFS), stopping if there is no word in the dictionary which starts with the strand so far.

Cool, thank you! I flew into Salt Lake City once so I’ve probably seen it from the air

Where is this?

Bumped the solve rate up to 20% by avoiding timeouts!

github.com/nimobeeren/s...

Cool! I guess I knew about the official Claude Code extension in Cursor/VS Code, but last I checked it wasn't very good

Oh interesting, I didn't know third-party tools were building on top of Claude Code.

Regarding results, have you compared it to alternatives or are you satisfied enough to just stick to what you know?

I’m increasingly doing the latter, e.g. I’m using Cursor + Opus for everything now