Fullstack that reaches production tested, secure and observable.

• 

  TaskForge

  Orchestrator that turns a request into a task graph, isolates each task in its own Git worktree and runs them in parallel waves, with human review and merge at the end. The planner is read-only, and no agent can publish on its own.

  1. Pressure and constraint

     An agent writing to a shared repository is a risk: a solo publish has to be impossible, not merely unlikely.

  2. Decision

     Split planning from execution, keep the planner read-only, isolate every task in its own branch and worktree, and leave review and merge to a person.

  • professional
  • TypeScript
  • Node.js
  • Zod
  • Git worktrees

• 

  StechClub

  Education platform delivered end to end: public area, company SSO, learning tracks, video served through signed S3 links, an application quiz and an admin panel. Used today by 15+ mentees.

  1. Pressure and constraint

     The company already had its own SSO, and course video cannot be served straight from the bucket.

  2. Decision

     Reuse corporate SSO instead of reimplementing auth, address media by hash under separate prefixes, and hand out links through an authenticated API.

  • professional
  • React
  • Vite
  • TypeScript
  • NestJS

• 

  Conn-Pod

  A library that splits AI work across different models: one executes, another verifies. Each role has declared read and write scope, bounded retries and an evidence receipt tied to the version. Every task runs in its own Git worktree, under a sandbox.

  1. Pressure and constraint

     The model that writes code cannot be the one that approves it. And an agent with disk access is attack surface: the boundary has to be verifiable, not agreed upon.

  2. Decision

     Split executor and verifier by model, run each task in its own worktree and keep writes with the host, which then checks that only the declared paths changed. A subscription is the default profile; paid API usage is opt-in only.

  • personal
  • TypeScript
  • Node.js
  • Claude Code
  • Codex CLI

1. October 21, 20261 min readin Portuguese

   Zoologic com Luna e Haiku

   O Zoologic ganhou autenticação, módulos de gestão e deploy usando principalmente GPT-5.6 Luna e Claude Haiku.

2. October 14, 20262 min readin Portuguese

   49 arquivos acusados, nenhum com defeito

   O linter da minha base de notas acusou 49 arquivos de reescrever o próprio histórico, e nenhum deles tinha defeito.

3. October 7, 20262 min readin Portuguese

   Nenhum modelo substitui outro em silêncio

   O modelo que fazia o papel de sênior no meu projeto pessoal ficou sem saldo no meio de duas tarefas, e as duas pararam com um erro 429.

1. Context

   Before writing code, I try to understand who depends on the system and what has already failed in it.

2. Constraint

   Deadline, team and existing debt are part of the problem, so the solution has to fit the constraint that actually exists.

3. Decision

   I write down what I chose, what I dropped and why, so the next person does not have the same argument again.

4. Evidence

   Tests, metrics and review close the task. If I cannot show the scenario where it works, it is not done yet.

If one of these cases looks like a problem of yours, write to me.