标签：#low

Introdução Integrar Apache Airflow com .NET 10 não significa portar o orquestrador, reescrever DAGs em C# ou executar o runtime Python dentro da aplicação. A solução correta é manter o Airflow responsável por criar, agendar e monitorar workflows e fazer o serviço .NET consumir sua API REST pública. O serviço autentica, dispara um DAG Run com parâmetros, guarda o identificador retornado e consulta o estado até receber success , failed ou canceled . Essa separação preserva o papel de cada tecnologia e cria um contrato claro entre a aplicação transacional e a plataforma de dados. Neste guia, eu vou implementar esse fluxo de ponta a ponta usando .NET 10 , HttpClient , autenticação JWT e a API /api/v2 do Apache Airflow 3.3 . O exemplo não se limita a um POST : ele gera um dag_run_id rastreável, serializa conf corretamente, reutiliza o token até perto da expiração, renova a credencial após uma resposta 401 , aplica timeout ao monitoramento e propaga cancelamento. Também vou expor a integração por uma Minimal API, para que outro sistema possa iniciar o processo sem conhecer os detalhes do Airflow. O nome atual da plataforma da Microsoft é .NET 10 , e não “.NET Core 10”. A marca “.NET Core” foi usada até a versão 3.1; desde o .NET 5, o produto unificado passou a se chamar apenas .NET. Essa diferença não altera o código, mas evita confusão ao procurar documentação, imagens de container e pacotes compatíveis. O cenário prático será um serviço de pedidos que solicita a execução do DAG etl_vendas . A configuração enviada contém a data de referência e um identificador de correlação. O Airflow continua executando tarefas Python, SQL, containers ou jobs distribuídos; o C# apenas controla o ciclo de vida da execução pela fronteira HTTP. ℹ️ Informação: no Airflow 3, os endpoints públicos estáveis ficam sob /api/v2 . Rotas internas de UI não são um contrato de integração e podem mudar conforme o frontend. Pré-requisitos Para acompanhar o exemplo, você precisa do .NET 10 SDK , do Dock

Building an operating system is a project I have had my eyes set on ever since I discovered free will in the realm of programming. Years ago, I did a reasonable amount of research, paying extra attention to the subject during my computer science degree and I was able to understand Operating System Theory and how it works from first principles but I never really got around to building one. I had only flimsy reasons for not embarking on it like "why build one when there are tons of working ones out there? The theoretical knowledge is enough" . More recently, I am ignoring the need to not re-invent the wheel for the joy of programming. So if you are interested in also rebuilding stuff because you can, join me on this series as I document how I am going to be building kluster. kluster is in its infancy and the direction is not clear but the one certain thing is that I will be building it entirely in Rust, save some assembly instructions and a linker script and I will be explaining every single line of code along the way. It will also be designed to target the raspberrypi 4 & 5, on qemu and on real hardware respectively. This is an opportunity for anyone who wants to see how Rust works at the lowest of levels to hop on and join the ride. Note that this series will be your biggest lesson on delayed gratification because we will write a lot of code before we even get to see anything meaningful on screen but I will foreshadow what you can get by the end of part 3 if you are patient enough: {{ image(src="/images/os-part3-result.png", alt="Part 3 Results OS Dev") }} You can also clone the source code for part 1 from Github and follow along. Project Setup First things first, let us setup the foundation of the project. I'll be straight with you, I love Rust and I enjoy using the Rust ecosystem in its entirety so I will stay true to that and use it as obsessively as any true Rustacean; I won't hold back. Without doubt, all the dependencies we need are freely available as long as

Jeremy Edberg & Qian Li discuss why external orchestrators decrease reliability and how to use your existing database for durable execution. They share how DBOS Transact uses standard tables, SKIP LOCKED queues, and unique primary keys to manage complex, fault-tolerant AI workflows with minimal latency, all without the operational overhead of separate distributed systems. By Jeremy Edberg, Qian Li

There's a class of bug reports that keeps coming back in the NocoBase community, especially in the Chinese-language forum: "all my times are off by 8 hours" or "dates show up as the day before." China is UTC+8, so the shift is 8 hours there. I run my instances at UTC+9, and sure enough — my shift is 9 hours. Whatever your offset is, that's the size of your shift. That pattern is a strong hint that this isn't random corruption. It's a mechanism. I set up NocoBase 2.x against both PostgreSQL and MySQL and measured what actually gets stored and how it gets reinterpreted, until the mystery had a concrete answer. Test setup: NocoBase 2.0.51 and 2.1.23 (official Docker images) × PostgreSQL 16 and MySQL 8.4. All data written and read through the REST API, with the server timezone controlled via the container's TZ environment variable. I'm deliberately ignoring the browser-side rendering here — this is about what the server stores and how it interprets it. Background: 2.x has four datetime field types NocoBase 2.x collections offer four datetime-ish field types ( official list — though several of the per-type detail pages still say "To be added", which is exactly why I measured instead): Type What it's for Datetime (with time zone) Absolute instants — event start times, logs Datetime (without time zone) Wall-clock times you want preserved as-is Date only Birthdays, due dates, anniversaries Unix timestamp System integration Measurement 1: what each type actually stores I imported "2026-07-12 09:00" via xlsx and looked at the raw values in each database (identical on 2.0.51 and 2.1.23): Field type PostgreSQL MySQL Datetime (with TZ) timestamptz → 2026-07-12 09:00:00+09 ( an absolute instant, offset included ) DATETIME → 2026-07-12 09:00:00 ( wall clock only — no offset information ) Datetime (without TZ) timestamp → 09:00:00 DATETIME → 09:00:00 Date only date → 2026-07-12 date → 2026-07-12 The first row is the whole story. The same field type — "Datetime (with time zone)" — i

GitHub redesigned GitHub Issues navigation using a client-side architecture that combines caching, predictive prefetching, and service workers to reduce perceived latency. The approach uses IndexedDB, in-memory caching, and background synchronization to serve data faster. GitHub reported instant navigation improvements from 4% to 22%, with latency reductions across multiple navigation By Leela Kumili

Guidelines, not rules. Here's the difference — and why it matters. What is SOLID? SOLID is a set of software design guidelines — not hard rules, but principles that guide how we organize our code. The goal is simple: as your codebase grows and your team scales, things should get easier to change, not harder. SOLID is what makes that possible. Five principles. One goal. Let's walk through each one with real code. S — Single Responsibility Principle A class, function, or method should have one and only one reason to change. The Violation class Bird : def __init__ ( self , name : str , bird_type : str ): self . name = name self . bird_type = bird_type def make_sound ( self ): # two jobs — deciding the type AND making the sound if self . bird_type == " parrot " : print ( " Squawk! " ) elif self . bird_type == " eagle " : print ( " Screech! " ) elif self . bird_type == " owl " : print ( " Hoot! " ) else : print ( " ... " ) make_sound() has two responsibilities — deciding which bird type it is AND making the sound. That's two reasons to change. Add a new bird? Touch make_sound() . Change how sounds work? Touch make_sound() again. Two different reasons, one method. SRP violated. The Fix from abc import ABC , abstractmethod class Bird ( ABC ): def __init__ ( self , name : str ): self . name = name @abstractmethod def make_sound ( self ): pass class Parrot ( Bird ): def make_sound ( self ): print ( " Squawk! " ) class Eagle ( Bird ): def make_sound ( self ): print ( " Screech! " ) class Owl ( Bird ): def make_sound ( self ): print ( " Hoot! " ) # Usage birds = [ Parrot ( " Polly " ), Eagle ( " Sam " ), Owl ( " Oliver " )] for bird in birds : bird . make_sound () Now each class has one responsibility. Parrot.make_sound() only changes if parrots change how they sound. Nothing else touches it. O — Open/Closed Principle A class should be open for extension but closed for modification. SRP and OCP go hand in hand. When you fixed SRP in the Bird example above — you also fixed OCP.

Glow is targeting a new class of endpoint risks created by the rapid adoption of AI agents and developer tools inside enterprises.

Picking the right deep learning library shapes almost everything about a computer vision project, from how fast you can prototype a model to how painful it is to ship one into production. Two frameworks dominate this decision today: PyTorch and TensorFlow. Neither has definitively won, but the split between them has become clearer than it was five years ago, and understanding that split is the fastest way to stop guessing and start building. Why the Choice of Framework Still Matters It's tempting to think framework choice is a solved problem — just pick whatever's popular and move on. But vision work has quirks that make the library underneath your code more than a technical footnote. Custom data augmentation pipelines, non-standard loss functions for tasks like instance segmentation, and the need to export models to mobile or edge devices all behave differently depending on the ecosystem you're in. Market data backs up the idea that this is still a genuinely contested space. TensorFlow holds a larger footprint in enterprise deployment, with roughly 37% market share and tens of thousands of companies using it in production, largely thanks to TensorFlow Serving, TensorFlow Extended, and TensorFlow Lite running across billions of devices. PyTorch, meanwhile, has become the default in research settings, with a majority of recent computer vision papers shipping PyTorch reference implementations first. Job postings mentioning PyTorch have also edged ahead of TensorFlow in recent hiring data, reflecting how much prototyping and applied research work now happens in that ecosystem. PyTorch: The Researcher's Default PyTorch's dynamic computation graph is the feature people mention first, and for good reason. Because the graph is built as your code runs, you can set breakpoints, inspect tensors mid-forward-pass, and change model behavior conditionally without recompiling anything. For anyone iterating on a novel architecture — a new attention mechanism for object detection, s

What Happened On July 20, 2026, cybersecurity firm Group-IB disclosed a new espionage implant dubbed HollowGraph that hijacks compromised Microsoft 365 mailboxes to run a command-and-control (C2) channel hidden inside calendar events. The malware attaches encrypted files to calendar entries dated May 13, 2050 — far enough in the future that a mailbox owner would never scroll to them — and retrieves operator instructions from the same dead drop. All traffic moves through the Microsoft Graph API, making the activity indistinguishable from legitimate M365 usage. At least 12 systems have been infected, with three actively communicating with the threat actor between June 3 and July 9, 2026. The indicators point to a targeted espionage campaign focused on Israeli organizations . Technical Analysis HollowGraph is a lightweight .NET DLL that supports only two commands: GET and SEND . To receive tasking, it queries the compromised mailbox's calendar for an event titled in the format "Event ID: <7-char-taskID>", downloads the attached file, and decrypts it using RSA and AES-256-GCM. To exfiltrate data, the implant creates a new calendar entry titled "Boss{..}ID{..}" and uploads stolen files encrypted with the attacker's public RSA key. The Group-IB research team described the mailbox calendar as a "covert dead-drop," with HollowGraph retrieving commands from events scheduled within a fixed one-hour window between 22:00 and 23:00 UTC on the far-future date. The hybrid encryption scheme uses separate RSA key pairs for inbound and outbound channels, keeping them cryptographically isolated. A second, unencrypted channel runs over DNS tunneling . HollowGraph refreshes its Microsoft Entra ID (Azure AD) credentials by querying IPv6 AAAA records from the attacker-controlled domain cloudlanecdn[.]com . Each returned IPv6 address yields 14 usable payload bytes, which the malware assembles and decodes as UTF-8 text to update its logAzure.txt configuration file — a file masquerading as a

Stack Overflow's question volume has been falling since ChatGPT went public in November 2022 ( OpenAI ). The site that trained a generation of developers, and most of the AI tools those developers now use, is slowly emptying out. In October 2023, Stack Overflow laid off 28% of its staff ( Stack Overflow Blog ). CEO Prashanth Chandrasekar framed it as a restructuring toward profitability. Everyone in the industry understood the real cause. Traffic was down. The thing causing it was sitting in every developer's browser tab. This is not another "AI killed Stack Overflow" piece. That take is everywhere and it misses the actual problem. The interesting part is the feedback loop, and it points somewhere uncomfortable for the AI industry itself. The conventional story, and what it misses The popular version goes like this. Developers used to paste error messages into Google and land on a Stack Overflow thread. Now they paste the same error into ChatGPT, Claude, or Copilot and get a direct answer. Why click through to a forum, risk a condescending comment, and wait for a human when a model answers in two seconds? That part is true. It explains the traffic drop. It does not explain why the people building the AI should be worried. The seed corn problem Here is the part most coverage skips. Every large language model trained on internet text consumed a huge amount of Stack Overflow. The site's archive of voted, edited, human-reviewed answers is one of the highest-quality programming datasets in existence. It is the reason an AI can answer your Python error at all. Now run the loop forward. AI tools answer questions directly. Developers stop posting on Stack Overflow. The archive stops growing. The next round of models trains on a corpus that is increasingly old, increasingly stale, and missing everything that happened after 2022. When you train an AI on data generated by another AI, quality degrades. Researchers proved this formally. Shumailov and colleagues showed that model

This is the workflow I follow before I use AI agents to implement any feature or bug fix. 🧭 Requirements/Specification ↓ Design/Architecture ↓ AI Code Generation ↓ Human Review ↓ Build & Static Analysis ↓ Testing & Validation ↓ Defect Resolution ↓ Security & Compliance Review ↓ Release ↓ Production Monitoring vs Claude Code ↓ Implements feature ↓ Codex QA Agent ↓ Runs application ↓ Tests happy path ↓ Tests edge cases ↓ Tests error handling ↓ Produces QA report This will resolve the self-review bias, confirmation bias, or AI-to-AI bias. 1️⃣ Understand Before Writing Code Before touching any code, I try to understand what I'm building and why . I usually start by reading: specs/<module>/<TICKET>-<slug>.md plan/<module>/<TICKET>-<slug>.md status.md Then I review the project conventions: specs/CONVENTIONS.md specs/conventions/core-porting.md Finally, I read the existing implementation (entities, services, mappers, etc.) so my changes follow the existing architecture instead of introducing a new style. 💡 Pro-Tip Good code fits into the codebase. Great code looks like it was always there. 2️⃣ Plan the Change Once I understand the requirements, I identify which architectural layers are affected. I always respect the dependency order: Schema / Entities / DAOs ↓ Mappers / DTOs ↓ Service Layer ↓ Application Layer ↓ Controllers I don't jump ahead of dependencies. If a change is complicated or ambiguous, I document the approach before writing code. --- ## 3️⃣ Write the Code While implementing, I follow the repository's rules. Some examples: | Rule | Detail |---|---|---| | DTOs | Generated from `schema.yml` — never handwritten | | Status values | Sourced only from the Core Porting specification | | Traceability | Every ported behavior includes a source citation | Citation formats I use: - `← Source <path>` - `← PS §...` - `← BR-###` Beyond repository rules, I also try to: - ✅ Match existing naming conventions - ✅ Keep comments minimal and meaningful - ✅ Make small, focused chang

Building an AI agent with n8n is the fastest, cheapest way to turn a large language model into a useful worker — if you stay within its sweet spot. The honest truth, informed by the custom agents we ship, is that n8n carries a well-scoped agent further than most people expect. An LLM node, a few tool/webhook nodes and a trigger are all you need. This guide walks you through that exact workflow and, just as importantly, names the precise moment n8n stops cutting it and a custom build must take over. What You Need Before You Start You'll need a running n8n instance (self-hosted or cloud) and API keys for the services you want to integrate. Grab a Gemini or OpenAI key from their respective developer consoles — n8n's official AI agent builder documentation lists the full compatibility. The quick-start template also gives you a one-click import to see an agent's skeleton immediately. How to Build an AI Agent with n8n: The Core Workflow The core is a chain of nodes: a trigger wakes the agent, an LLM node reasons, and tool/webhook nodes take action. That's the entire pattern. Here's how to assemble it. Set the trigger Drag a Webhook node onto the canvas if you want the agent called via HTTP, or a Schedule node to run it periodically. For our example, we'll use a webhook that receives a customer question. Add the LLM node Attach an OpenAI Chat Model (or Gemini) node. In the node's parameters, craft a system prompt that scopes the agent. For a support bot, something like: You are a helpful support agent for our SaaS product. Use the tools provided to answer questions. If you don't know, say you need human help. This prompt is the boundary of the agent's autonomy. Keep it specific — vagueness leads to hallucinations. Attach tool and webhook nodes Here's where n8n shines. Drag a Function node to run custom JavaScript (e.g., querying a database) or a HTTP Request node to call an external API. Wire them as "tools" by connecting them to the LLM node's tool output. In the LLM node

The Problem Real411 is a South African platform where citizens report digital harms: misinformation, incitement, hate speech, and harassment. When someone submits a complaint, it needs to be reviewed by multiple people, assessed against legal criteria, and resolved with a public verdict. The process must be transparent, auditable, and fair. I joined this project early and worked on it extensively over a long period. A senior solutions architect consulted on the database schema design. There was a cloud person who helped with parts of the infrastructure. Other coworkers contributed at different stages. I spent most of my time on the API layer and the frontend components. This article covers the architecture decisions I worked with, what I learned from the senior architect's design choices, and how the system evolved. The Status Machine Most applications model status as a column on a table. You update the value and the old state is gone. That works for simple workflows but fails when you need to know not just where a complaint is now, but how it got there and who made each decision. The senior architect who consulted on the database design suggested an append only status log. Instead of a single status column, the complaint_status table records every transition as a separate row. Each row has the status code, the user who made the change, a timestamp, and optional notes. The current status is derived by querying the most recent row. I implemented this pattern across the API layer. Every status transition became an insert operation rather than an update. It took some adjustment to shift from mutable state to event sourced state, but the benefits were immediate. Auditing became straightforward. The state machine also became easier to implement because each transition is a simple insert with a business logic check, not a conditional update. The schema has seventeen status codes covering the full lifecycle: received, claimed, under assessment, pending secretariat review,

New study confirms 2024 "momentum flux theory" on how angular momentum of water flows drives rotation.

A team lead announces that the team will start using AI-assisted development. Everyone nods. Nobody asks what that actually means on Monday morning. Some times ago I was in that position. A project I was working on needed to start using AI-assisted development, and the team was new to it. Nobody had rules written down for an agent to follow. Nobody had skills defined for it to load. There was no shared idea of how this should work inside our specific repo. Someone had to go first. That someone was me. The rules worked because I built them for one repo I spent time curating a set of rules and skills for that project. Not generic ones. I shaped them tightly around how that repo was actually structured, its conventions, its layout, the things a new engineer usually has to learn by asking around. I wanted an agent working inside that codebase to already know what a human teammate would have picked up in the first two weeks. I gave a demo. It landed well. Well enough that it got shared further across team, as something other teams could learn from. I gave the demo again. Same reaction. Then a few developers reached out for the actual rules and skills files. I said sure, and then I actually looked at what I would be handing them. The problem showed up the moment other people wanted in It was not copy-paste-able. The rules referenced folder names, module boundaries, and patterns specific to one repo. Handing them over as-is would have meant handing over advice that was wrong for their project, dressed up as a shortcut. So I told them to use it as a reference. Look at the structure, understand the reasoning, adapt it to your own repo. That is correct advice. I watched people nod at it and then quietly missing it. I was solving the wrong problem the whole time I had been thinking about this as a documentation problem. Write good rules, explain them well, let people copy the idea. What I actually had was a generation problem. The rules that worked were the ones rendered speci

When a series of regional outages forced a rethink of a multi-region AWS API, the team discovered that an obstacle to global failover was hiding in plain sight: a pre-flight discovery call baked into every client session years earlier as the only available option. This article describes what it took to remove it, and what the rollout actually cost. By Suresh Gururajan

In the previous post I argued that agents are great planners and DAGs are great executors . This one is the practical follow-up: when you actually sit down to wire several LLM calls together, what tool do you reach for? Because the moment one prompt's output feeds the next, you've built a workflow — whether you call it that or not. download transcript → summarize → translate (tool) (LLM) (LLM) That tiny pipeline is already the whole problem in miniature: a non-LLM step (fetch a YouTube transcript), then a model call, then another model call that depends on the first. Run it as one giant prompt and you lose visibility; split it into steps and you gain debuggability — at the cost of more calls and more state to manage. The naming trap Half the confusion is vocabulary. The same idea ships under a dozen labels: Name What it whispers Chain sequential, output → input Pipeline stages, data flowing through Flow branches and conditions Workflow general orchestration Agent workflow the model also decides The word sets expectations. "Chain" promises a straight line; "agent workflow" promises the thing might re-plan on you mid-run. Pick the label that matches how much autonomy you're actually handing over — calling a deterministic two-step pipeline an "agent" only invites disappointment. The real choice: library or orchestrator? There are two families of tools, and they solve different problems. LLM-native chaining libraries — LangChain , LlamaIndex Workflows , Azure Prompt Flow , or visual layers like Flowise . These understand LLM-specific concerns out of the box: prompt templating, passing context between steps, token budgets, streaming, retries on a flaky model. General orchestrators — Airflow , Prefect , AWS Step Functions , Azure Logic Apps . These treat each LLM call as just another task in a DAG, and give you the heavyweight reliability machinery: durable state, scheduling, checkpointing, audit trails, human approval. The rule of thumb that falls out of the last post: F

I am interested in agentic coding for the same reason I care about good engineering process in general: I want work to move forward in a way that is inspectable, repeatable, and resilient once the task gets messy. A lot of AI-assisted coding still feels like improvisation. You ask for something, get a result, adjust the prompt, try again, and hope the useful reasoning is still somewhere in the scrollback. That can work for tiny edits. It gets much less convincing when the task starts touching architecture, tests, review, or pull requests. What I want instead is a workflow where the model helps me think and execute, but inside a structure I can inspect afterwards. I want artifacts, gates, and something I can resume tomorrow without reconstructing the entire mental state from memory. That is why I use po8rewq/agentic-skills . It gives me a practical way to do agentic coding as an engineering workflow rather than as a long sequence of chat turns. A task moves through requirements, architecture, implementation, checks, review, and pull request creation. Each stage leaves something I can read, verify, and challenge. What makes this interesting to me The interesting part is not just that there is a CLI. Plenty of tools have a CLI. What matters to me is that it turns AI-assisted coding into a staged system: requirements force the task to become explicit architecture makes risks visible before code is written implementation happens against a plan instead of against a vague prompt checks and review happen as part of the flow, not as an afterthought runs are resumable, so interruptions do not destroy context That changes the feel of the work quite a bit. Instead of asking "what should I prompt next?", I am usually asking "what stage is this task in, and what should exist before I move on?" Where this really clicked for me was when I noticed I was spending less energy trying to preserve context in my head and more energy evaluating actual outputs. What the repository actually

Why I Choose Lovable for Building Full-Stack Applications with AI Over the last year, AI-assisted software development has evolved from generating code snippets to building complete web applications. We've all seen tools like Cursor, Claude Code, GitHub Copilot, Replit Agent, Bolt, and many others enter the market. Each has its strengths, but after experimenting with several of them, I keep coming back to Lovable whenever I want to build a new web application from scratch. This isn't a sponsored post—it's simply the workflow that has worked well for me. If you're interested in trying Lovable, you can use my referral link below. Disclosure: new users receive additional signup credits, and I receive referral credits if you sign up through it. Referral: https://lovable.dev/invite/AQ02SOZ Why Lovable Stands Out Most AI coding assistants help you write code. Lovable helps you build an application. Instead of focusing on individual functions or files, it takes a higher-level approach where you describe what you want, and it generates a complete full-stack application that you can continue refining. A typical workflow looks like this: Idea │ ▼ Describe the application │ ▼ Lovable generates • Frontend • Backend • Database • Authentication • API integration │ ▼ Preview instantly │ ▼ Connect GitHub │ ▼ Iterate and Deploy Unlike traditional no-code platforms, you're not locked into a proprietary editor. Lovable supports GitHub synchronization, native Supabase integration for authentication and PostgreSQL-backed data, and deployment options ranging from Lovable-hosted apps to your own infrastructure. Why I Keep Choosing Lovable After building several side projects, these are the reasons I continue to use it. 1. Rapid idea-to-production workflow The biggest productivity gain isn't AI-generated code. It's reducing the number of decisions needed before users can interact with your application. Instead of spending hours creating project structure, authentication, routing, database

Why Workflows Need a Dedicated Evaluation Framework Traditional software testing covers code correctness. Workflows add two layers of uncertainty: LLM output is non-deterministic : the same input can produce different results across runs Cross-step dependencies : a Phase 3 problem may only surface at Phase 7, making the debugging chain long Without an evaluation framework, every workflow change requires a full end-to-end run: slow, expensive, incomplete coverage. Three-layer testing decomposes the problem. Three-Layer Evaluation Structure Layer 3: End-to-end tests (Workflow level) Full pipeline from trigger to completion Test cases: eval/cases.yaml Metrics: completion rate, Phase 4 avg rounds, gate trigger rate Layer 2: Integration tests (Phase level) Cross-step data flow is correctly passed Cross-phase routing logic fires correctly Layer 1: Unit tests (Step level) Each subagent's output matches its output contract No real LLM calls — validates JSON schema only Test priority: Layer 1 should be the most numerous and fastest — catches contract violations in seconds. Layer 3 is the slowest and most expensive — run it only when changes affect the main pipeline. Layer 1: Step-Level Unit Tests Unit tests verify that subagent output files match the declared schema. No real LLM calls needed. # tests/unit/test_phase3_output.py import json from pathlib import Path def test_analysis_output_schema (): """ Phase 3 output must conform to analysis_final.json schema """ output = json . loads ( Path ( " test_fixtures/phase3/analysis_final.json " ). read_text ()) assert " passed " in output assert isinstance ( output [ " passed " ], bool ) assert " confidence " in output assert 0.0 <= output [ " confidence " ] <= 1.0 assert " root_cause " in output assert isinstance ( output [ " root_cause " ], str | type ( None )) assert " evidence " in output assert isinstance ( output [ " evidence " ], list ) # on failure, error field must be present and non-empty if not output [ " passed " ]: ass