evansjp

Nothing prunes it, atm. Currently I'm not sure if it's valuable that it does? 97KB is ~25k tokens. `grepathy context <file>` filters by the decision's Touches: list - a typical file pulls 3KB, not 97. But, I'm open to hearing a case on pruning!

For comparison the whole git log body on that branch is 3KB; the pack holds 33x the reasoning.

these are great suggestions! terrific recs - thanks. I think you're right to lean into OC and Hermes

whatever your CLI currently uses. so if you use subscription, it'll use that. vice-versa for API.

hey that's awesome! yeah the eval showed first pass was only ~65% real decisions. The fix that stuck was an entry has to name a real file it touches or it gets dropped. A code decision names code.

I agree agents don't always self-talk decisions, that's why we distill the whole transcript after the fact instead of asking them to log anything. Your baselining idea is good!

lol yes, agents ran the mechanics. I set the bars before any run and audited the keys. And the "why" behind the project is literally in the repo's .ai/why/ where you grep it if you would like.

- the distiller reads your messages, it just doesn't quote them. Decisions you directed get captured and labeled `directed`.

- eval is public here: https://github.com/evansjp/grepathy/blob/main/docs/REPORT.md

- fair point on the last one too. For the original Clerk incident the transcript is deleted, so I can't prove nobody approved it. Maybe I even waved it through. That's kind of the point though. Right now there's no record either way. I'd rather have some sort of receipt.

thank you so much! started with a Claude Code adapter, but you think Codex would be a good next one?

CTO responded well to it - frankly because we're all using some coding agent and it's just the natural process of code review for them to sniff things out. No mention of it in any PR summary - CTO found it in the migration that would mutate the User record to accept `null` email.

Some of you are wayy too bored at work - some of these are too realistic

You're absolutely right, I'll get an update out here to lift the country restriction.

Very cool, but shouldn't be on Show HN yet until people can actually use it.

That's actually a super good idea for a metric I didn't think about tracking and I love that you commented that.

What was the time for the event that didn't show up in the calendar? Was it something till midnight?

Kali is the Hindu goddess of time--also of death but we'll ignore that for now lol--and she is a "she".

Haha alright, I can turn off flirting by default for the next update. And how late in the afternoon did it suggest for a lunch?

Thank you!

I'd be flattered for sure. I think there's a lot of uncaptured value in AI to API requests, in general. Imagine an app that schedules "Flight at LAX at 2pm", but also offering to order the Uber to the Airport in time to get there 2 hours before departure. Sounds pretty cool to me and until OpenAI's GPT Plugins have broader support, I'll build that :)

I was taking the strategy of no one has an iOS app called "Kali", but perhaps it's a bad strategy. We'll have to see. For now I just want to know if the app is something people like :P

If I were your assistant managing your schedule and you come to me and say "I have a sales meeting next Tuesday." Of course I'd schedule the sales meeting, but I might also schedule time for you to 1) Research the client, 2) Drop off dry-cleaning to look sharp for the meeting, 3) A time to pick up that dry-cleaning.

That's what a "scheduling assistant" does. :)

AI is good at reasoning what other events should be scheduled along w/ an event.

Thanks for your question!

Damn that's an awesome idea! Writing that down... Thanks!

Yep you're right on there. That's been the most popular complaint--working to add that right now!

Thanks for the video--watched it. Dalton and Michael are definitely right that Discovery ideas are limited because their bound to finite discovery (only so many restaurants open tonight, etc). I'll start thinking of the optimal pivot from here, thanks again for the recommendation! Helped a lot.

Also only available in US atm, so sorry to my fellas not in the States!

I'm actually looking for any way to improve Harvester. Finding product-market fit is SO hard, so please let me know what you hate, love, would love to see.