The fenbs blogAll articles
Page 14 of 48
- Claude Code on Windows: Native or WSL, and How to Install EachClaude Code runs natively on Windows without WSL, or inside a WSL distribution. Which to pick, the exact install command for each, where Windows keeps your settings and sign-in, and the errors Windows users actually hit. · 7 min read
- Claude Code Commands Cheat Sheet: Slash Commands and CLI FlagsThe slash commands, CLI flags and keyboard shortcuts you will actually use, grouped by job and one line each, with a link to the guide that goes deep on every group. · 7 min read
- How to Create a Claude Skill: SKILL.md Template and Five RulesBuild one skill end to end: the folder, the SKILL.md frontmatter, a description that makes it trigger, bundled scripts, testing that it fires when it should and not when it should not, and sharing it in Claude Code and the Claude apps. · 7 min read
- Claude Code Plugins, Skills, Subagents and Commands: Which to UseA skill is instructions, a subagent is a separate worker, a command is the old form of a skill, and a plugin is a package of the others. What each one is, how they compare side by side, where hooks and MCP servers fit, and a short decision flow. · 6 min read
- CLAUDE.md Examples: Four Real Files and What Each Line DoesFour annotated CLAUDE.md files, for a web app, a library, a monorepo root and your personal ~/.claude/CLAUDE.md, with the reason each line earns its place and the kinds of line that should never be there. · 7 min read
- Claude Code Best Practices: 25 Tips From Daily UseTwenty-five habits that make Claude Code sessions shorter, cheaper and easier to trust, grouped under planning, context, review, permissions and recording the work, each with a link to the guide that goes deeper. · 7 min read
- How to Reduce Claude Code Token Usage, and See What You UsedWhere to see what a Claude Code session used, what actually spends the tokens, and the settings and habits that cut usage without making the work worse. · 7 min read
- Auto-Approve in Claude Code: Auto Mode, Allow Rules and RisksEvery Claude Code permission mode, how to write allow, ask and deny rules that approve the safe commands and nothing else, when the skip-permissions flag is acceptable, and a starter settings file. · 8 min read
- Claude Code for Beginners: Your First Hour, Step by StepA first-hour walkthrough of Claude Code: install it, sign in, open a repository, ask a question, make one small change, read the diff, commit it, and know what to learn next. · 7 min read
- Claude Code Desktop App vs CLI vs VS Code: Which to UseThe desktop app, the terminal CLI and the VS Code extension run the same Claude Code engine and read the same configuration. What each one adds, what each one lacks, who each suits, and how to move a session from one to another. · 7 min read
- Claude Code vs Cursor: Which Fits the Way You WorkOne is an agent that started in the terminal, the other an editor built around its agent. Where each runs, how each plans, which instruction files each reads, how you approve and undo its work, and who each suits. · 8 min read
- Claude Code vs Codex: Two Coding Agents ComparedAnthropic’s Claude Code and OpenAI’s Codex each run in a terminal, an editor, a desktop app and the cloud. Where they differ: the instruction file each reads, how each decides what needs your approval, subagents, MCP and pull request review. · 7 min read
- The Best AI Coding Agents, Compared by How You WorkThere is no single best AI coding agent, only the one that fits where you work, how much you let it do alone and how you review what it did. Six questions to ask, then Claude Code, Codex, GitHub Copilot, Cursor, Gemini CLI, Windsurf and Aider measured against them. · 7 min read
- Installing Codex CLI on Windows: Native, WSL or the AppCodex runs on Windows three ways: the CLI natively in PowerShell, the CLI inside WSL 2, or the Codex view of the ChatGPT desktop app. Which to pick, how each installs, how the Windows sandbox works, where your settings live, and the errors Windows users hit. · 7 min read
- Codex App vs CLI vs IDE Extension: Which Codex to UseCodex is one agent with several front doors: the CLI in your terminal, the extension in your editor, the Codex view of the ChatGPT desktop app, Codex cloud, and review on GitHub. What each is for, what they share, and who each one suits. · 6 min read
- ChatGPT Agent Mode: What Replaced It, With 12 ExamplesChatGPT agent mode has been retired, and its job now belongs to ChatGPT Work and its cloud browser. What agent mode was, how Work takes a task today, what it stops to ask you, twelve example tasks with the checkpoint in each, and the risks that have not gone away. · 7 min read
- ChatGPT Projects Best Practices: Files, Instructions and MemoryScope a ChatGPT project to one body of work, choose its memory setting on purpose, keep its files current and few, write instructions a stranger could follow, and save what the chats decide back into the project. · 6 min read
- copilot-instructions.md Examples and a TemplateThree annotated .github/copilot-instructions.md files, for a TypeScript web app, a Python service and a .NET API, plus a path-specific .instructions.md file, a template to copy, and the lines that should never be in any of them. · 7 min read
- GitHub Copilot CLI: Install, Sign In and Your First Agent SessionGitHub Copilot CLI is Copilot’s agent in your terminal. How to install it on Windows, macOS and Linux, sign in, run a first session safely, decide what it may do without asking, and where it fits beside Copilot in your editor and the cloud agent. · 8 min read
- AGENTS.md Examples: Templates for Python, TypeScript and MonoreposAnnotated AGENTS.md files for a Python project, a TypeScript app and a monorepo root with a nested file, how long each should be, what never goes in one, and which coding agents read it. · 7 min read
- Azure DevOps MCP Server: Connect Claude, Copilot or CursorMicrosoft’s official Azure DevOps MCP server comes in two forms: a hosted remote endpoint and a local npm package. Which one your assistant can use, how sign-in works, how to narrow the tools, and how to keep it read-only. · 7 min read
- GitLab MCP Server: Setup, Tools and MCP vs glabGitLab builds its own MCP server into GitLab itself. Where it lives, who has to switch it on, how the OAuth sign-in works in Claude Code, Cursor and VS Code, which tools it offers, and when an agent is better off with the glab CLI. · 7 min read
- Slack MCP Server: Setup in Claude, Cursor and ChatGPTSlack runs its own MCP server, and it reaches your assistant through clients that carry a registered Slack app. How to connect Claude, Claude Code, Cursor and ChatGPT, what your workspace admin has to approve, which scopes each tool needs, and what an assistant should never post on its own. · 7 min read
- Claude Desktop MCP Config: The JSON File, With ExamplesWhere claude_desktop_config.json lives on macOS and Windows, the exact mcpServers format, working examples for npx, uvx, node and Python, environment variables, Windows paths, and why remote servers now go in as connectors instead. · 7 min read
- Claude Desktop MCP Server Not Showing Up: A Fix ListWhen a server in claude_desktop_config.json never appears, fails to connect or shows no tools, the cause is almost always one of a handful. A fix list by symptom, where the logs are on macOS and Windows, and how to test the server without Claude. · 8 min read
- Which Apps Support MCP? The Major MCP Clients ComparedClaude, ChatGPT, Codex, Cursor, VS Code, Copilot CLI, Gemini CLI, Devin Desktop and Zed all speak MCP, but not in the same way. What each one’s own documentation says about local and remote servers, sign-in, and support beyond tools. · 8 min read
- MCP Inspector: How to Test and Debug an MCP ServerThe MCP Inspector is the reference tool for poking at an MCP server without an AI in the way. How to run it, connect it over stdio or Streamable HTTP, sign in, read the traffic, script it, and get past the errors people hit first. · 8 min read
- MCP Transports: stdio vs Streamable HTTP, and What Happened to SSEThe current MCP specification has two standard transports, stdio and Streamable HTTP. The old HTTP with SSE transport is deprecated, yet SSE still appears inside Streamable HTTP. What each one is, when to use which, what the requests look like, and how to move off the old one. · 7 min read
- Building an MCP Server in Python With FastMCPA working MCP server in Python with two tools, run over stdio and HTTP, tested in the MCP Inspector and connected to Claude Code. Plus the naming trap: FastMCP in the official SDK is now called MCPServer, and the standalone FastMCP is a separate project. · 8 min read
- MCP vs CLI for AI Agents: When a Command Line Beats a ServerA coding agent with a shell can use a command-line tool instead of an MCP server, and often should. The general trade-off: discovery, whose credentials, context cost, structured results, sharing, remote use and audit, and how to use both. · 8 min read
- What Is Agentic AI? A Plain-Words Definition With ExamplesAgentic AI is AI that is given a goal and works toward it by choosing and taking actions, not just producing an answer. Where the term came from, what people mean by it, eight examples, and what it is not. · 7 min read
- Agentic AI vs Generative AI: The Difference in Plain WordsGenerative AI produces something for a person to use. Agentic AI uses tools to do something, several steps in a row. The difference side by side, the same job done both ways, why the line blurs, and what it changes for a team. · 7 min read
- How to Build Your First AI Agent, Step by StepThe smallest useful AI agent has one goal, two or three tools, a loop, a stop condition and a person who approves the risky step. Here it is in about sixty lines with the Claude Agent SDK, then how to test it and give it a task list to work from. · 9 min read
- Why AI Agents Fail in Production: Nine Causes and FixesMost agent failures are not the model being stupid. They are vague tasks, crowded context, poor tools, long chains, no finish line, no tests, the wrong permissions, agents talking past each other, and nobody checking. Nine causes, what research says about them, and the fix for each. · 8 min read
- Indirect Prompt Injection: How It Reaches AI AgentsAn AI agent does not need to be talked into anything by its user. It only needs to read something an attacker wrote. Where that text comes from, how an attack unfolds, why a better prompt will not stop it, and the controls that do limit the damage. · 8 min read
- How to Evaluate AI Agents: Metrics, Test Sets and Human ReviewAn agent that worked in the demo can still fail one run in three. What to measure, how to build a small test set from your own tasks, which grader to use for what, and a weekly loop a small team can keep up. · 7 min read
- AI Agent Observability: Traces, Logs and the Record People ReadWhen an agent takes forty steps to finish a task, you need to see all forty when something goes wrong. What to capture, the OpenTelemetry conventions for it, what Claude Code and the agent SDKs already export, and why a machine trace is not the record your team reads. · 7 min read
- What Is Spec-Driven Development? A Practical GuideSpec-driven development means writing down what you want and why before an AI agent writes any code, then planning, splitting into tasks and building against that document. What a good spec holds, the loop, the tools that formalize it, and when it is more ceremony than the job needs. · 7 min read
- Claude Cowork: 15 Example Tasks and PromptsFifteen jobs you can hand to Claude Cowork today, each with a prompt you can adapt and the point where Claude stops to ask you: files, research, documents, spreadsheets, email and calendar, and work that runs on a schedule. · 7 min read
- Claude in Chrome: What the Extension Can Do and How to Stay SafeClaude in Chrome lets Claude read, click and type in your browser, using the accounts you are already signed in to. What it can do, how its site permissions and approval modes work, what Anthropic says about prompt injection, and the habits that keep it safe. · 8 min read
- User Story Template: The Format, and a Card You Can CopyThe As a / I want / So that format, where it came from, the INVEST check in six lines, and a card template you can paste into any board, with a version for when an AI assistant will do the work. · 7 min read
- User Story Examples, With Acceptance CriteriaTwenty user stories across e-commerce, SaaS admin, a mobile app, an internal tool, an API and a building job, each with the acceptance criteria that say when it is done. Then five weak stories rewritten. · 10 min read
- Acceptance Criteria Examples: Given/When/Then and ChecklistsThe two ways to write acceptance criteria, a checklist of rules and Given/When/Then scenarios, when each one fits, twenty examples across common features, the mistakes that make criteria useless, and who writes them. · 9 min read
- Acceptance Criteria vs Definition of Done, With ExamplesAcceptance criteria say when one piece of work does what was asked. The definition of done says what every piece of work must meet before anyone calls it finished. One worked story shows both, and how they look when an AI assistant does the work. · 7 min read
- Definition of Done: Examples and a Checklist TemplateFour definitions of done you can copy, for a software team, a small web team, a content team and work done by AI assistants, plus a blank checklist template and a short way to agree one without a week of meetings. · 7 min read
- Bug Report Template: What to Include, With ExamplesA bug report needs six things: a title that names the problem, the environment, numbered steps, expected and actual results, evidence, and severity. A template to copy, a bad report rewritten, a version an AI assistant can act on, and a shorter form for people outside the team. · 7 min read
- Kanban vs Scrum: Which Suits a Small TeamScrum plans work in fixed Sprints with set roles and events. Kanban manages a continuous flow and limits how much is started at once. What each guide actually asks for, a side-by-side comparison, and a decision guide for a small team. · 7 min read
- Release Notes Template and Ten Good ExamplesRelease notes tell the people who use your product what changed and what, if anything, they need to do. A structure that works, a template you can copy, the difference from a changelog, and ten short examples. · 7 min read