Very much a beginner experiment, but I wanted to try making graphics into something you can interact with instead of just looking at.
I made the visuals in Midjourney, then brought them into Catnip to build out the interactive parts, tweaking the reactions as I went.
Would love to know what still feels rough, and what you’d add if you kept going with it.
Sinner's withdrawal was the first thing it caught, and the full pipeline's here:
Scout maps the field first: rankings, form, injuries, surface history, and what the open questions are
Planner splits the research into dimensions (hard-court form, injury history, Slam pressure-handling, draw paths)
Researchers run in parallel, one per dimension, each fetching its own sources
A validator gate runs after every stage: evidence files must pass a schema check, and claims must pin to source snapshots before the next stage starts
A reviewer checks for gaps and contradictions
Writer + stitcher assemble the report from evidence units
Citation renderer makes every claim traceable to a numbered source
Why the process mattered here
Two facts reshaped the draw before the analysis even started, and the pipeline surfaced both as structured evidence rather than vibes:
- Sinner (world No. 1, 44-3 in 2026, six titles including Wimbledon) withdrew with a right-knee injury. That removes the only player who dominated across every surface this year.
- Djokovic lost in the first round. The field's best big-match operator is gone.
The output
Men's: Zverev champion, Medvedev runner-up. Zverev is the highest-ranked fully healthy player, the only man with two Slam finals in 2026, and just won his first major (Roland Garros). His historic knock, not closing, is weaker than it has ever been. Medvedev has the best remaining hard-court resume (Brisbane, Dubai, Indian Wells final) but is 1-4 in Slam finals, which is precisely the runner-up profile.
Alcaraz is the hinge, not the pick: defending champion, but four months out with a wrist injury, parachuting into a best-of-five Slam. The pipeline classifies him as the highest-variance player in either draw.
Women's: Sabalenka champion, Gauff runner-up. Sabalenka's hard-court season is one of the best of the Open Era: the Sunshine Double, seven straight hard-court major finals, and 20 consecutive Slam tiebreaks won. Gauff is the only real threat (Cincinnati champion, 2023 US Open winner, home crowd), but the 2026 head to head tilts to Sabalenka.
The part I actually want to show you
Here's a claim from the report, the kind that single-prompt prediction never produces:
"Sabalenka's 20 consecutive Grand Slam tiebreaks won, breaking Djokovic's record."
That's not a vibe. It's a pinned, citable fact with a source snapshot behind it. Every material claim in the report works this way. If a prediction is wrong, you can go back and see exactly which evidence led there.
Skill pack is MIT open source: https://github.com/OpenSenseNova/SenseNova-Skills
I've been lying awake every night for almost three weeks now and I'm losing my mind. I've tried the whole routine, no screens, chamomile tea, blackout curtains, and none of it makes a difference once my brain decides it's not shutting off. Reading late night threads about how to fall asleep when you can't sleep just makes me more anxious because everyone seems to have some magic trick that never works for me. I'm exhausted during the day, snapping at people I care about, and I genuinely don't remember what it feels like to just close my eyes and drift off. Has anyone actually found something that helped when nothing else did, or am I just stuck like this forever?
The current collection covers:
- Excel and multi-file data analysis
- Deep research with citations and intermediate evidence files
- PowerPoint generation
- Infographic and image generation
- Academic, developer, financial, and social search
The most interesting part is how these skills can be chained together.
One of the included examples starts with a raw CSV containing memory-price data, cleans and analyzes the data, researches the reasons behind the price changes, cross-checks sources, produces a written report, and then turns the findings into a 16-slide presentation.
So the workflow is closer to:
raw data → analysis → sourced research → report → PPT
rather than simply asking a model to “make a presentation” in one prompt.
The repository is here:
https://github.com/OpenSenseNova/SenseNova-Skills
One additional note for anyone using the Excel or data-analysis skills: their open-source visual-generation model, SenseNova U1, can also turn the analyzed data into infographics. The results shown in the project examples look pretty solid. OpenSenseNova has also just made a free token plan available here:
Just joined the Dreame side with the X60 Max Ultra Complete.
Only finished unboxing and setup so far, but first impression is good. The packaging was neat, the setup process was simple enough, and the base station looks surprisingly nice in the room.
I’m coming from a much more basic robot vacuum, so I’m especially interested in the automated base features and whether they actually make daily cleaning feel more hands-off.
Will probably share an update after it runs for a few days. Curious if there are any settings I should change first.
I kept hitting the same wall with Claude: it's brilliant at doing things you ask it to, but it has no perception of what's happening in the world between your messages.
You have to keep poking it.
"Check if that game dropped in price."
"Look up whether anyone tweeted about our launch."
"See if that PR got reviewed."
It's always you who has to notice first.
So, World2Agent (W2A) is here: an open protocol + TS SDK that lets Agents subscribe to sensors, so it can notice things on its own and tell you.
Agents is great at doing things you ask. W2A lets it notice things you didn't ask about.
What you can do with it today:
These are all just sensors you plug in.
The fastest way to feel W2A is with Claude Code. In an active session, install the world2agent plugin:
/plugin marketplace add machinepulse-ai/world2agent-plugins
/plugin install world2agent@world2agent-plugins
/reload-plugins
Add a sensor — for example, Hacker News:
/world2agent:sensor-add @world2agent/sensor-hackernews
Restart Claude Code with the plugin channel loaded so sensor signals flow into your session:
claude --dangerously-load-development-channels plugin:world2agent@world2agent-plugins
What's inside:
Happy to hear what sensors people want next, Github repo's in comment