---
title: "Mihai_LeanZero (@mihaileanzero) | Hashnode"
description: "Atlassian consultant and Forge developer at LeanZero. We do Data Center to Cloud migrations and build Forge apps, and I write up what I learn at leanzero.net."
url: "https://hashnode.com/@mihaileanzero"
---

@mihaileanzero

# Mihai\_LeanZero

@mihaileanzero · Bucharest, Romania · Joined September 2026

Atlassian consultant and Forge developer at LeanZero. We do Data Center to Cloud migrations and build Forge apps, and I write up what I learn at leanzero.net.

[Website](https://leanzero.net/)

## Comments

[Good checklist. On keeping tool descriptions static, I'd also pin them on the client side: hash each tool's name and description when someone approves the server, and re-prompt when the hash changes. Otherwise "static" only protects you from servers that behave. Same logic for annotations like readOnlyHint, the spec treats them as hints, so the real enforcement has to be your own scopes and server-side checks, which you already say in the last group.](https://hashnode.com/posts/twenty-security-checks-before-your-mcp-server-goes-public/6ad155a64c89af0468c46f61/comment/6ad1bb3a0c5c64ab3d9e8f1a)

Comment · Article · Oct 16 · [Twenty Security Checks Before Your MCP Server Goes Public](https://hashnode.com/posts/twenty-security-checks-before-your-mcp-server-goes-public/6ad155a64c89af0468c46f61)

[The errors-are-instructions section is the part I'd underline. One thing I'd add next to the "Do not retry" message: that text is still only a request, and a model will occasionally retry anyway, so I'd back it with a counter in the loop that refuses a third identical call (same tool, same arguments) and returns a fixed "this exact call already failed twice" result. I'd also give the error object a retryable field of its own instead of leaving it to the prose, since a timeout and a not\_found need opposite instructions and the model shouldn't have to infer which one it got from the wording.](https://hashnode.com/posts/ai-agent-tool-calling-designing-tools-the-model-will-actually-use-correctly/6ad0d1e8865ce90870b8c7fe/comment/6ad0dc81afbfd10d12159349)

Comment · Article · Oct 15 · [AI agent tool calling: designing tools the model will actually use correctly](https://hashnode.com/posts/ai-agent-tool-calling-designing-tools-the-model-will-actually-use-correctly/6ad0d1e8865ce90870b8c7fe)

[Fair, a test that reproduces the failure is the stronger version and the trailer is the cheap fallback for rules that can't be tested. If a rule can be pinned by a regression test or a structural check (a lint, a type that makes the bad state unrepresentable), the trailer should just name that test and the date becomes noise. The rules I'd still keep a dated incident note for are the behavioral ones, where there's no test to point at, and there the date at least tells you whether the underlying cause could have been patched out since. Pruning a rule whose test still fails without it is easy, pruning one with no test is where the guessing starts.](https://hashnode.com/posts/good-agentic-workflows-are-born-from-pain-not-principles/6aaee98131b8b540d787fd9b/comment/6ab00715325ee698b4811a7c#reply-6ad0698b479de991a5dd79eb)

Reply · Article · Oct 15 · [Good Agentic Workflows Are Born From Pain, Not Principles](https://hashnode.com/posts/good-agentic-workflows-are-born-from-pain-not-principles/6aaee98131b8b540d787fd9b)

[One thing worth checking on that Modelfile: the TEMPLATE only uses .Prompt, with no .Messages and no .Response. Going by template.go in the Ollama repo, a template with neither gets {{ .Response }} appended at parse time and then runs once per user/assistant pair in legacy mode. So from turn two on, the earlier answer lands right after the bare <|start|>assistant with no channel header and no <|end|>, and the hard-coded system block repeats in front of every pair (a SYSTEM line in the Modelfile is ignored too, since .System is never referenced). Single-shot ollama run would look fine, multi-turn chat is where I'd expect it to drift. Ranging over .Messages and writing each past assistant turn as a final-channel message closed with <|end|> should avoid it. I read this from the source, I haven't run it against a gpt-oss GGUF.](https://hashnode.com/posts/fine-tune-gpt-oss-20b-with-unsloth-and-run-it-in-ollama/6acde4dbefbf7f1d1effb388/comment/6ace0131ef889bfa647a351a)

Comment · Article · Oct 13 · [Fine-tune gpt-oss-20b with Unsloth and run it in Ollama](https://hashnode.com/posts/fine-tune-gpt-oss-20b-with-unsloth-and-run-it-in-ollama/6acde4dbefbf7f1d1effb388)

[Good rundown, especially the honest framing that most of your latency is sitting in the Mongo call and not the runtime. One thing worth adding for anyone reading the type-stripping section as "great, I'll drop ts-node everywhere" -- the unsupported-syntax list in Node's own docs is longer than just enum and decorators. Parameter properties (the constructor(private foo: string) shorthand a lot of NestJS-flavored Express code uses) and TS import aliases (import Foo = Bar.Baz) also throw ERR\_UNSUPPORTED\_TYPESCRIPT\_SYNTAX, same as enum declarations and namespaces with actual runtime values. None of those are exotic in a backend that grew up with decorator-style patterns even before using real decorators, so "just run node app.ts" can still blow up on day one for a reason that has nothing to do with enum.](https://hashnode.com/posts/node-js-vs-bun-vs-deno-in-2026-the-runtime-decision-your-mern-stack-actually-needs-to-make/6abfb87a68bbc0a1067d3222/comment/6abfc02c6a21b8485fd45ea9)

Comment · Article · Oct 2 · [Node.js vs Bun vs Deno in 2026: The Runtime Decision Your MERN Stack Actually Needs to Make](https://hashnode.com/posts/node-js-vs-bun-vs-deno-in-2026-the-runtime-decision-your-mern-stack-actually-needs-to-make/6abfb87a68bbc0a1067d3222)

[The highest-numbered-container detail checks out -- checked the current docker/compose scale-down code and it explicitly orders containers so a scale-down trims the highest replica numbers first and keeps the lowest one running. So web=2 back to web=1 does remove the newer container and keep the older one, exactly as described. What the fixed-name web-blue/web-green setup buys you over --scale is not depending on that ordering holding: it's current implementation behavior, not a documented contract, so a future Compose release could change the tie-break without calling it a breaking change. Naming the slots explicitly makes the swap correct regardless of how Compose orders replicas internally.](https://hashnode.com/posts/architecture-case-study-migrating-a-developer-saas-from-serverless-to-a-10-vps-with-docker/6abf646f1308726bbb4b9cd5/comment/6abf89b3d62e6e07297c10c6)

Comment · Article · Oct 2 · 1 · [Architecture Case Study: Migrating a Developer SaaS from Serverless to a $10 VPS with Docker](https://hashnode.com/posts/architecture-case-study-migrating-a-developer-saas-from-serverless-to-a-10-vps-with-docker/6abf646f1308726bbb4b9cd5)

[The cross-origin fetch part is what stands out most to me here -- the browser reads another repo's .keep-the-why, then fetches that repo's state.json directly, no server in between. GitHub Pages happens to serve everything with permissive CORS by default, which is probably why this works smoothly for the examples here, but the post calls out "GitHub Pages or another static host." For a host that doesn't set Access-Control-Allow-Origin -- a bare S3 bucket without CORS configured, a plain nginx box serving static files -- does the Globe just silently drop that wave, or is there a visible failure state distinguishing "nothing there" from "blocked by CORS"? That seems like exactly the kind of thing that would be invisible until someone hits it.](https://hashnode.com/posts/the-globe-following-your-reasoning-into-other-people-s-repositories/6abe79902a890b1eceb0ebd1/comment/6abf508e601d77f7e6af1c20)

Comment · Article · Oct 2 · [The Globe: Following Your Reasoning Into Other People's Repositories](https://hashnode.com/posts/the-globe-following-your-reasoning-into-other-people-s-repositories/6abe79902a890b1eceb0ebd1)

[The verification section is the strongest part of this for me -- treating "document not found" as the only thing that establishes logical absence, rather than trusting the delete call's own success response. One thing worth adding for Azure AI Search specifically: deletes propagate to replicas asynchronously, so a verification read issued right after the delete request can still land on a replica that hasn't caught up yet and come back with the document still present. If the verification step runs immediately after the delete call in the same task execution, that's a false "still there" -- harmless if it just triggers a retry, but it could get misread as "deletion failed, escalate" when the delete actually succeeded. Feels like it wants the same settle/retry treatment you already give the delete itself, applied to the verification read too.](https://hashnode.com/posts/deleting-personal-data-from-a-rag-system-in-net/6abd5fa1c02e2338b2100943/comment/6abf507dc548f1f021941369)

Comment · Article · Oct 2 · 1 · [Deleting Personal Data from a RAG System in .NET](https://hashnode.com/posts/deleting-personal-data-from-a-rag-system-in-net/6abd5fa1c02e2338b2100943)

[Good rundown, especially the biased-reference-counting explanation, that part usually gets glossed over. One thing I'd push on: the use case you're recommending it for, ML preprocessing alongside a Node API, is also the one most exposed to the C-extension caveat you flag near the end. Plain-Python CPU work via ThreadPoolExecutor genuinely benefits today, but the moment that pipeline leans on numpy or pandas internally, which most ML preprocessing does, the actual speedup depends on whether those specific wheels ship free-threaded builds, not on whether 3.14t itself is stable. Worth naming which of the common data-science libs already ship free-threaded wheels as of 3.14 versus which still fall back to re-enabling the GIL at import (it's just a warning, easy to miss in a busy log), since that's the real gate on whether the recommendation holds for that exact workload.](https://hashnode.com/posts/python-finally-killed-the-gil-what-free-threading-in-3-14-means-for-your-stack/6abe19ad0e8fa9267ac13f61/comment/6abe692419ce2918e5eea945)

Comment · Article · Oct 1 · 1 · [Python Finally Killed the GIL: What Free-Threading in 3.14 Means for Your Stack](https://hashnode.com/posts/python-finally-killed-the-gil-what-free-threading-in-3-14-means-for-your-stack/6abe19ad0e8fa9267ac13f61)

[Really thorough methodology, especially separating the client-side watch timeout from the controller's own ProgressDeadlineExceeded, that distinction trips people up constantly. One thing worth asking about the recovery step: reapplying the same healthy spec created a brand new revision each time (2 to 4, 4 to 6, 6 to 8) instead of landing back on an existing one. In a setup where you're triggering an automated rollback off ProgressDeadlineExceeded, does kubectl rollout undo without an explicit --to-revision reliably land on the last genuinely healthy spec, or does it just walk back to the most recent distinct ReplicaSet regardless of whether that one was itself a different failure? Two failures in a row would make that ambiguity matter a lot more than the single-fault case this lab walks through.](https://hashnode.com/posts/how-to-debug-a-stuck-kubernetes-rollout-with-a-hands-on-lab/6abe3dc913d662f7af15bf03/comment/6abe691719ce2918e5eea8cc)

Comment · Article · Oct 1 · [How to Debug a Stuck Kubernetes Rollout with a Hands-On Lab](https://hashnode.com/posts/how-to-debug-a-stuck-kubernetes-rollout-with-a-hands-on-lab/6abe3dc913d662f7af15bf03)
