Hans L’Hoest @hanlho.com · 29/09/2026This is where they could have mentioned that Huggingface could defend itself from the OpenAI hack exactly because of GLM 5.x as the closed models didn’t allow them.. 1536
Hans L’Hoest @hanlho.com · 22/09/2026Good read on the internals of Opencode by Kit Langton. And there is the postscript … read at your own peril! 010
Hans L’Hoest @hanlho.com · 17/09/2026Got access to Siri AI. Asked first 'in vogue' question. "what is Jev the new model by typesafe.ai" not bad. My next question: will this stay free? There must be some part here where Apple will start charging for it. 000
Hans L’Hoest @hanlho.com · 07/09/2026Mistral models usage on Openrouter. GLM 5.2 at the top by a wide margin. Medium 3.5, their latest general purpose model is not doing great. openrouter.ai/provider/mis... 010
Hans L’Hoest @hanlho.com · 04/09/2026Vibe Code allows you to choose if the file systems needs to roll back as well when you traverse your session history. That's different from Pi where the file system remains untouched (the better default). Claude Code used to always rewinded the file system (not sure if that's still the case). 000
Hans L’Hoest @hanlho.com · 02/09/2026The model was already there, now you can use GLM 5.2 in Vibe Code hosted in the EU: 000
Hans L’Hoest @hanlho.com · 17/07/2026Kimi K3 got released, also on Opencode Go. Lets hope something can be done on the usage later on when open weights are released? 011
Hans L’Hoest @hanlho.com · 05/06/2026Adversarial prompting also works for reviews: ask the LLM to form an hypothesis on what could be wrong in the code base and let it report with test scenarios that expose the breakage. I packed it as a skill but mostly use it directly, `adversarial-review`: codeberg.org/hanlho/agent... Example: 100
Hans L’Hoest @hanlho.com · 18/05/2026Results from a long-horizon study. Anthropic and Grok, surprisingly weirdly confirming what I would expect them to be: on both ends of the spectrum. www.emergence.ai/blog/emergen... 000
Hans L’Hoest @hanlho.com · 12/05/2026LLMs' performance are often categorised. One category is 'tool calling', I was not sure what it meant and when to use it so I did some AI-assisted research and turned into an infographic. Also available here: hanlho.com/p/when-to-us... 010
Hans L’Hoest @hanlho.com · 11/05/2026Understandable but I wasn’t looking forward to this. Pi is renaming all its packages. I pin extension versions after security review, so I’ll need to re-review them again at some point. (Remember, they run in-process with Pi’s full privileges.) 000
Hans L’Hoest @hanlho.com · 03/05/2026A year ago I started adding below signature to my emails because it felt like at least doing something about people carelessly handling data. I wonder what others have tried to protect their data in emails. (And is workable, pgp is not so much unfortunately.) Feedback welcome! 100
Hans L’Hoest @hanlho.com · 02/05/2026Was casually checking what alternatives there are to Classlist that are privacy conscious. Auto correct split the name but regardless, recommending as a first an application that promotes integrating well with a Google product will not be high on my list. 000
Hans L’Hoest @hanlho.com · 29/04/2026I am a user of Opencode Go and Codex plans. With Deepseek V4 coming out I asked if I should replace Kimi with it. Then asked to turn it into an infographic. Maybe it helps someone else too. (GPT 5.5 is more expensive than Opus these days. I works very well, not using it as default.) 000
Hans L’Hoest @hanlho.com · 24/04/2026I put together a Jujutsu footer extension for the Pi coding agent. After a few days of usage: works pretty well. Jujutsu Footer Extension for Pi Coding Agent: hanlho.com/go/jj-footer... 020
Hans L’Hoest @hanlho.com · 14/04/2026First Pi extension: show Jujutsu information in TUI (Fully vibe coded. Shelving under: works, low risk.) 000
Hans L’Hoest @hanlho.com · 30/03/2026Plan Mode ... No harm done but this is why I started running in Safehouse 000
Hans L’Hoest @hanlho.com · 26/03/2026If you installed or upgrade LiteLMM on March 24th all your secrets may have been stolen. (Is it me or should their TLDR at the top be more explicit about that?) docs.litellm.ai/blog/securit... 000
Hans L’Hoest @hanlho.com · 23/03/2026Asked my coding agent to write after a file to ~/dev Instead, it created ~/dev/1774107553.txt in my working directory. I think I am not going to ask to clean that up ... 000
Hans L’Hoest @hanlho.com · 02/03/2026Just for fun, or in case you need inspiration naming your agents, here is the list of names Codex pulls from when creating an agent: github.com/openai/codex... 000
Hans L’Hoest @hanlho.com · 27/02/2026If code is written more and more by LLMs, the focus seems to be shifting to creating guardrails so agents can validate their own work. Heard in: The Pragmatic Engineer - Mitchell Hashimoto’s new way of writing code share.snipd.com/episode/d792... 000
Hans L’Hoest @hanlho.com · 03/02/2026With every skill becoming also command across coding agents quickly, typing “/” gets unwieldy fast. I like Codex’s split: use “$” for skills/custom commands only; “/” should only show its built-in commands. 000
Hans L’Hoest @hanlho.com · 20/01/2026Too many models on Opencode cleaned up: (Snapshot, will probably be different in a month.) 100
Hans L’Hoest @hanlho.com · 15/01/2026So this morning the Claude Code plan is again blocked from Opencode. Anthropic is doubling down on this one. What is interesting is that they specifically seem to be targeting Opencode. For example Droid and Cursor do still work. xcancel.com/ryanvogel/st... 110
Hans L’Hoest @hanlho.com · 12/01/2026Why is Claude Code asking me every day to accept "Updates to Consumer Terms and Policies"? Hoping I may one day accept the default option? 100
Hans L’Hoest @hanlho.com · 09/01/2026Anthropic blocked Claude Code credentials from being used in Opencode, and presumably other CLI agents. Ouch. 100
Hans L’Hoest @hanlho.com · 06/01/2026Nice. Makes me think of a 'Reflect' phase I have in one of my commands that I use when implementing tasks. Clearly on a smaller and local scale. Your command seems to expand on that by 'bulk reflecting' on PRs (and taking into account pipeline checks). 010
Hans L’Hoest @hanlho.com · 05/01/2026My current LLM agent architecture to try and remove duplication from commands, skills and agents. And CLI agents, in this case between Claude Code and Opencode. 100
Hans L’Hoest @hanlho.com · 25/12/2025Opencode now also supports skills. Even Claude Code compatible. 011
Hans L’Hoest @hanlho.com · 16/12/2025Where has this been for all my CLI life: github.com/charmbracele... 110
Hans L’Hoest @hanlho.com · 04/12/2025Also, Zed has been supporting rainbow _indents_ for a long time. Below is how you configure those and what it looks like: 010
Hans L’Hoest @hanlho.com · 25/08/2025ccstatusline: very helpful to get insight in Claude Code's context statistics. Most useful for now: how much time is left in the 5h window (Block Time)? github.com/sirmalloc/cc... 010
Hans L’Hoest @hanlho.com · 12/12/2024Socio-technical architecture in a nutshell. (source: Domain Modeling Made Functional by Scott Wlaschin) 020
Hans L’Hoest @hanlho.com · 21/11/2024Ran into this same post on BlueSky and Twitter/X which surprised me quite a bit. Comments : 41 vs 25 Likes : 31 vs 40 110
