Home›Practice
Practice
Practice
Hands-on guides and first-hand notes on agent frameworks, skills and AI tools.
Total 217
Running Claude Code safely in a dev containerRun Claude Code inside a dev container so its commands stay in an isolated, reproducible environment. Covers setup, keeping sign-in across rebuilds, organization policy and the risks of skipping prompts.Using Claude Code with Amazon Bedrock or Google Cloud's Vertex AIRun Claude Code on your own AWS or Google Cloud account: set up Amazon Bedrock or Google Cloud's Vertex AI, configure credentials and environment variables, pin model versions and fix common errors.Code review with Claude CodeSet up Claude Code to review pull requests: learn what the GitHub Code Review service checks, who can turn it on, and when to run /code-review locally instead.Getting started with Claude Code in the desktop appUse Claude Code without a terminal: open a local project in the desktop app's Code tab, send a first request, review the changes and choose a permission mode.Running Claude Code tasks in the cloudClaude Code on the web runs tasks in cloud VMs that keep going after you close your laptop. Learn to start tasks from the terminal, pull sessions back, share them, and auto-fix pull requests.Using Claude Code with ChromeClaude Code can drive your signed-in Chrome browser through the Claude in Chrome extension to test local pages, read console errors, fill forms and extract data. Here is how to connect and troubleshoot it.Claude Code agent teams: several sessions working togetherAgent teams let several Claude Code sessions split up work, message each other and share a task list. Learn how to enable a team, manage teammates, and when not to use one.Claude Code output styles: changing how Claude respondsClaude Code output styles change Claude's role, tone and response format for a whole session. Learn the four built-in styles, switch between them, and write your own style file.Customizing the Claude Code status lineA guide to the Claude Code status line: write a small script that reads session JSON, wire it into settings, and fix a blank bar.Switching and configuring models in Claude CodeHow to choose and switch models in Claude Code: what each alias points to, the priority of /model, --model and ANTHROPIC_MODEL, and how to adjust effort and extended thinking.Claude Code costs: checking usage and cutting spendHow Claude Code charges by API tokens, how to read the /usage session block and plan breakdown, how teams see spend, and the main habits that cut token usage.Claude Code checkpointing: rewinding mistakesClaude Code checkpoints file edits before each turn. Use /rewind to restore code, conversation or both, summarize long sessions, and see which changes checkpoints cannot undo.Claude Code interactive mode: shortcuts and tipsCommon Claude Code interactive-mode shortcuts and tips: interrupting, queuing messages, running commands with !, asking side questions with /btw, and fixing Option-key shortcuts on macOS.Claude Code CLI reference: commands and flags by taskA grouped cheat sheet for Claude Code's command line: starting and resuming sessions, one-off prompts in scripts, and controlling permissions, models and spend, plus what to do about flags missing from --help.Claude Code settings.json: scopes, precedence, and troubleshootingWhere each Claude Code settings file applies, how precedence resolves a key set in several places, session-level overrides, and how to check with /status and claude doctor when a change does not take effect.Claude Code permissions: allow, ask, and deny rulesHow Claude Code permission rules work: allow, ask, and deny evaluation order, what 'don't ask again' saves, how Bash wildcards match, and why Bash rules are not a security boundary.How Claude Code loads CLAUDE.md and auto memoryHow Claude Code's two memory systems work: CLAUDE.md files you write and auto memory Claude writes, where each is loaded from, when it loads, and how to check why an instruction did not take effect.How to write and use Skills in Claude CodeWhat Claude Code Skills are, where to save them so each session loads them, how the SKILL.md frontmatter controls invocation and tool approval, and how to troubleshoot a skill that does not trigger.Connect Claude Code to an LLM gatewayHow to point Claude Code at an LLM gateway that your organization runs: set the base URL and credential, let administrators distribute and pin the settings, and see how gateway billing differs from a claude.ai subscription.Building a search app with embeddingsEmbeddings turn search from keyword matching into meaning matching. You will build a semantic video search that returns answers with timestamps, learn the chunk-and-embed pipeline, and set up Azure OpenAI for it.Building a chat applicationFrom a first API call to a production chat app: tell chatbots from AI chat applications, pick SDKs over hand-built plumbing, design the UX quirks AI introduces, and track the metrics that keep quality up.RAG and vector databasesRAG grounds an LLM on your own documents: chunk them, embed them into a vector database, retrieve the closest chunks per query, and generate answers from them. You will build a small RAG chatbot end to end and know how to evaluate it.Getting started with function callingFunction calling makes an LLM return structured JSON describing a function to run, bridging chat and your code. You will build a course-search flow end to end: define tools, parse the call, run the API, and answer the user.Advanced prompting techniquesSeven prompting techniques you can stack: few-shot, chain-of-thought, generated knowledge, least-to-most, self-refine, maieutic, plus temperature control. You will pick the right mix for a task and make outputs predictable.