
And if you're applying to Startup School, paste your Paxel token into your application! We want to see how you build with these tools. Already applied? You can still go back and add it.
http://ycombinator.com/startupschool

And if you're applying to Startup School, paste your Paxel token into your application! We want to see how you build with these tools. Already applied? You can still go back and add it.
http://ycombinator.com/startupschool

@ycombinator I can’t believe there’s a free way to give YC my coding data for nothing in return. This is going to change the world. Trulyz

@ycombinator give it a try!! also built this form so if you have a bug or feature request just submit it here. we'll review it then fire off an agent and merge the result if we like it:)

@ycombinator "At the end of the run, a JSON payload of scores, narratives, redacted decisions, and session metadata is uploaded to YC." -from paxel url ..
should i use it?? 🤔

@ycombinator Why is this jerk blocking everyone on X?

@ycombinator This is... extremely invasive

@ycombinator What about adding it to your application for the summer batch? 👀

@ycombinator Transcripts leave the box but code is regex removed.
We are specific about this: code doesn’t leave. Other transcript data does.

☠️ you are a literal retard if you use this slop ☠️
What actually lands on YC servers (summarized from the docs):
• Behavioral scores (5 axes: Execution Leverage / Steering / Engineering Quality / Product Thinking / Planning). • Per-session narratives (~2,000 chars). • Redacted session events (file paths + truncated commands, capped). • First-prompt excerpts (200 chars). • Selected user-highlight excerpts (capped at 10k chars/session). • Steering traces, dispatch metadata (task descriptions ≤200 chars; raw dispatch prompt is replaced by first 12 hex chars of its SHA-1 hash). • Git metadata (remote URL, recent commit subjects/authors, aggregate velocity/LOC stats). • Client telemetry (runtime errors, pipeline duration, LLM call stats — credentials scrubbed). • Account metadata (your email + auth tokens for profile aggregation across machines).

@ycombinator Wow, now this is epic! Definitely gonna run this on my gstack- can I contribute to the repo? Just need a quick review Garry

@ycombinator Absolutely love this. We've already integrated Claude Code with the Autonomous Desk.
Paxel just gave us the v2 specs. We'll integrate Paxel with our desk!

@ycombinator Where is the code?
Complete embarassment to tell people to execute an arbitrary script without open sourcing.

@ycombinator Gotta open source it and make sending data to YC optional, otherwise it's just people donating their data to YC

@ycombinator @gustaf nice hair cut @garrytan

@ycombinator inspired by @charlieholtz at conductor, love the idea

@ycombinator it's like npx standout haha

@ycombinator reverse meta prompt the Claude Code /insights skill and grep local sessions folder across Claude, Codex, and Cursor and tell me how a build with AI.
genius.

@ycombinator is it originally from one skill of gstack?

@RobImbeault Code doesn’t leave your computer (at least that’s what Garry said in the video).

@ycombinator the concept here is great. a profile of how you actually build with ai pulled from real transcripts is a mirror devs have never had. make it truly local and this is an instant install for everyone living in claude code. the idea's right, just close the upload loop.