Following-up on my previous article about Agent Skills on Gemini CLI, this time I'm showing how to create your own skills and share a few more that I recently created. #gemini #geminicli #agentskills
danicat.dev/posts/202602...
📣 The 185 mentoring organizations for #GSoC2026 are live!
Student or new to open source, now is the time to start exploring project ideas. 💻✨
🚀 Applications open March 16
🚀 Accepted Contributors April 30
🚀 Coding Starts May 25
opensource.googleblog.com/2026/02/intr...
#GSoC #OpenSource
She was built with ADK, Vertex AI Memory Bank and Vertex AI RAG, among other things. I'll soon publish about the whole process!
This is why I'm building my own. I'm calling her Sharon, but her codename is "OpenPaw" as paws are cuter than claws. :)
Sharon is just a prototype but she is already changing my life. One of her best moments was sending me to sleep because I was working on her overnight. lol
One of my personal frustrations is that I was never important enough to have an Executive Assistant. Having someone to manage your appointments, activities, priorities, etc., sounds like a dream, specially coming from someone with ADHD. #adk #gemini #agents
One of my favourite uses of AI is to work with images. In this week's blog I wrote about how I'm using Genkit and Nano Banana Pro to restore old family photos. Please check it out!
danicat.dev/posts/202602...
#genkit #golang #nanobanana #gemini3
How do you measure AI agent success? @danicat83.bsky.social built Tenkai - a framework to deploy and benchmark your agents.
github.com/danicat/tenkai
Curated in this week's #notonlyswift
E.g. I was just writing a genkit flow in Go and the agent was not very good at it so it took a lot of trial and error. At the end I asked "please create a 'genkit-go' skill documenting how to properly create a project in genkit go". #genkit #geminicli #antigravity #agentskills
A small tip for those working with Gemini CLI, Antigravity or any other coding agent that support Agent Skills: whenever you are doing something new, and you finally reached a point where the agent managed to do a working implementation of what you requested, stop and ask it to create a skill .
There seem to be interest in specialist servers for other languages beyond Go, potentially creating an “idiom” or template for what a language-aware MCP should look like. What do you think? Where would you like to see this research go next? #mcp #golang #fosdem2026
How do you scale AI agent expertise without drowning in context bloat?
In my latest article, I dive into **Agent Skills** — a new feature in Gemini CLI that represents a shift toward declarative engineering. Read more about modular, shareable AI workflows here:
🔗 danicat.dev/posts/agent-...
Of course, when you branch out too much, at some point you need to start refining and consolidating. So a phase of rapid expansion (using Antigravity) is often followed by focused work to refine and publish it (using Gemini CLI). In some sense, it is similar to the double diamond pattern in design.
For example, when creating Tenkai, my experimentation framework, I was often split between frontend and server upgrades, but also tweaking godoctor (the thing that was being tested). I could do all of these changes in parallel thanks to the agent manager.
Antigravity works better for me when I'm in brainstorming mode or working with web applications. Agent manager mode allows me to work in projects that have multiple "moving parts" at the same time.
It takes away the distractions and I can pay close attention to what the model is doing. It also blends more nicely when I working on stuff that is inherently terminal based, e.g., if I need to constantly run commands on the terminal to validate my development, as when I'm developing an MCP server.
I'm now using both Gemini CLI and Antigravity for long enough to start forming some opinions. The terminal UI in Gemini CLI is really helpful when I want to work in **focused** mode.
#gemini #geminicli #antigravity
I haven't been mindblown by Gemini in a while, then today the cli decides to do this... Is this what the kids call these days "karma farming" or "flexing"? 😆
#gemini #svg
Next week , on February 1st, I will be at FOSDEM talking about my development process for MCP servers using Go. I'm also showcasing the experimentation process I have been developing for the past few weeks. If you are around don't be shy and come say hi!
fosdem.org/2026/schedul...
#golang #oss
One of the things I never really adapted well after moving to the UK is the darkness during the winter. It totally messes up with my biorhythm. My sleep schedule is completely messed up, so I'm working crazy hours to compensate. I'm so glad my job allows this sort of flexibility
New post on my blog! This time I'm talking about how to use science and statistics to put some order into the chaos of non-deterministic behaviour in coding agents. Can you really say your best practices are really the best? Now I can (with 95% confidence 😁)!
danicat.dev/posts/202601...
Looking forward to attending the AICamp meetup in London tomorrow! I'll be speaking about Agent Development Kit (ADK).
www.aicamp.ai/event/eventd...
Taming Vibe Coding: The Engineer's Guide danicat.dev/posts/202512...
#gemini #vibecoding #lifeatgoogle
Building FrameFit: AI-Powered macOS App in 3 Hours
Written with #Gemini and speedgrapher (created by @danicat83.bsky.social). I had to give it a try 👏. @developers.google.com #GDE
www.eliostruyf.com/building-fra...
Being interviewed to write an article is definitely a nice way to think about writing a new content.
@danicat83.bsky.social, speedgrapher is a nice project! Great work! 👏
github.com/danicat/spee...
How to Create a Diagnostic Agent With Agent Development Kit danicat.dev/posts/202510...
Someone at @ardanlabs.com shared this.
yzma is the new Go package for calling llama.cpp without CGo that I introduced at my @golab.io talk today. Check it out!
github.com/hybridgroup/...
#golang #llamacpp #llm #vlm #slm #tlm
Getting ready to setup for my @golab.io keynote "That Machine Always Lies".
#golab2025 #golab #golang #tinygo
“Util packages are a sign of defeat” @konradreiche.com at @golab.io
Totally agree! Naming is hard, but getting names right can dramatically improve software design, as often one thing leads to another. #golang
First speaker of the day at @golab.io is @danicat83.bsky.social with "The Gopher's Craft in the Age of AI"
#golab #golang
