Zukko Blog
Guides to Claude Code, Cursor, Codex and the Claude and GPT APIs: setup, token savings and troubleshooting.
Prompt Injection: What It Is and How to Protect a Bot and an Agent
5 min · · development
How prompt injection works through user input and third-party documents, and which measures limit the damage: data isolation, tool permissions, key limits.
Your Own Neural Network Proxy on Node.js: Keep the API Key Off the Frontend
4 min · · development
A small Express server between your frontend and the model API: the key stays on the server, and models, request size and rate are capped.
DeepSeek API in Python: Requests, Streaming, JSON and Error Handling
4 min · · development
Ready code for DeepSeek V4.1 Flash and V4 Pro on the OpenAI SDK: a request, streaming, JSON parsing and careful error handling.
Prompt Engineering Examples: 12 Ready-Made Templates for Work
4 min · · development
Twelve prompt templates for everyday tasks: what each one says, why it is built that way, and what a single API call costs.
Gemini API in Python: requests via the OpenAI SDK and OpenAI format
5 min · · development
A working Gemini API example in Python with no extra library: change base_url in the OpenAI SDK, add streaming and JSON, and compare with google-genai.
Claude Code YOLO: Permission Modes and Working Without Prompts
6 min · · агенты
Claude Code has no mode literally called YOLO, but it has several permission modes and flexible rules. How to stop clicking Yes without losing files.
Chatbot memory: how to store conversation history without paying twice
6 min · · development
The model does not remember earlier messages: you send the history with every request. We look at how dialogue cost grows and what keeps it in check.
OpenAI SDK or LangChain: when a framework helps and when it gets in the way
5 min · · development
One scenario written with the plain SDK and with LangChain: how much code, where debugging is easier and how to keep control of token spend.
Anthropic SDK for Python: base_url, key and a first request via a gateway
5 min · · development
The official anthropic package for Python just needs a different address. We cover base_url, the key, streaming, token counting, caching and errors.
Translating subtitles with a neural network: SRT without broken timecodes
6 min · · автоматизация
The main risk when translating SRT with a neural network is a broken file structure. A scheme where timecodes never reach the model, and a Python script with validation.
Claude Code: a checklist of best practices for daily work
6 min · · development
Habits that make working with Claude Code predictable: a short CLAUDE.md, a plan before edits, small tasks, checking results and keeping spend under control.
Claude Code security: a checklist before running the agent on a project
6 min · · development
The agent reads files and runs commands, so security settings belong before the first launch. A checklist: permissions, secrets, sandbox, a separate key with a limit.
Editing texts with a neural network: proofreading, style and cutting
5 min · · development
Different kinds of editing need different prompts. Wording for proofreading, style and cutting, how to get edits with explanations and how not to lose the author's voice.
An MCP server in Java: a minimal example with one tool
6 min · · development
We write a minimal MCP server in Java with one tool, register it in Claude Code, and learn to find the cause when the agent doesn't see it.
ANTHROPIC_API_KEY and ANTHROPIC_AUTH_TOKEN: where to get a key and where to set it
5 min · · API keys
Which variable the SDK needs, which one Claude Code needs, where to get a key and how to set it on Windows, macOS and Linux without it reaching git.
A neural network for Lua and Roblox: writing Luau scripts with AI
6 min · · development
Without guidance a model may confuse Lua with Luau and suggest outdated Roblox APIs. How to phrase the task, check the code and connect an agent to your project.
A database for RAG: where to store knowledge — vector DB, search or context
6 min · · development
RAG does not require a vector database. Four ways to store knowledge, the cost of a query for each, and a working SQLite example.
A neural network for writing scripts: Bash, PowerShell and Python
5 min · · development
A model writes scripts for routine fast, but running them blindly is risky. How to state the task, demand a safe dry-run mode and check the result before launch.
LangChain agents: tools, the loop and capping spending
5 min · · агенты
We build an agent with create_agent and two tools and run it on a model through an OpenAI-compatible gateway, then cap the steps and count tokens.
OpenAI API documentation in plain terms: Chat Completions by parameter
5 min · · development
A short reference to the Chat Completions endpoint. Each parameter with an example, the response and usage format, gateway-specific notes, and the cost of one reply.
Claude Code Router: do you need it if the gateway speaks the Anthropic format
6 min · · агенты
Claude Code Router appeared to point Claude Code at other providers' models. If the gateway already accepts the Anthropic format, much of that is one variable.
AI for game programming: Unity, Godot and browser games
6 min · · development
An agent writes scripts but does not see the engine editor. How to structure a game project for AI, what to hand the agent, what to do by hand, and what it costs.
Dify in Docker: installation, first-run setup and connecting models
6 min · · development
Install Dify with Docker Compose, create the admin account and connect Claude, GPT and DeepSeek models on one key. Plus what to do when containers will not start.
Refactoring code with AI: small steps and tests
5 min · · development
An agent can rewrite a module in a minute and break it just as fast. How to organize refactoring: tests first, one kind of change at a time, a commit per step, rollback.