@KostyaAI stated in the tweet that Spider's interface isn't the most interesting part. What's really fascinating is that it refuses to guess. Every instruction is mapped to claims, owners, files and approvals, and missing context gets flagged instead of being made up. Better agents may come from better context graphs, not necessarily longer prompts.

This Spider tool is shared by @defileo. He wrote: "Holy hell, someone got a spider to read my prompt... I'm never writing prompts by hand again." He originally thought his prompt was clean and had been running it for weeks. Then he had Spider read through it from a stranger's perspective: 181 words, scored 37 on the first pass.

Here's the detailed process:

- Every word is tagged as one of: claim, owner, file, approval, or ambiguous

- Ambiguous words go to his "second brain", which brings back 45 links he had already decided on previously

- 12 unresolvable terms were flagged for him to handle

- 0 terms were guessed

The final score was 82, and he didn't rewrite a single line.

He noted that the prompt was never bad to begin with—it was just disconnected from the things he already knew. Spider pulls in that existing context, so he doesn't have to type it all in manually.

This perfectly confirms KostyaAI's judgment: better agents may come from better context graphs, not necessarily longer prompts.

A similar principle appears in another guide for Claude Opus 5.5 by @defileo. He recommends adding three files to a project: who.md, voice.md, now.md, and explicitly lays out a rule: "Never make up numbers, dates, prices, names or citations. If you don't have it, say 'I don't have this information' and tell me where I can get it." Another rule states: "If a question has a definite correct answer, just give it directly. Don't give me three options when one is clearly correct."

All of these point to the same direction: instead of letting AI guess, it's better to give it clear context and allow it to say "I don't know".

One netizen commented: "Getting 37 on a prompt you've run for weeks is really humbling." Maybe we should all let Spider read our prompts too.