Guides

Guides to agentic work

Skills, hooks, routines, subagents and MCP, explained with the setup steps and recipes you need, plus the craft of prompting and evals.

Guides

Agentic ways of working

The building blocks for handing work to an agent and keeping control: what each one is, when to use it, and how to set it up.

Agentic ways of working · Beginner · 18 min read

The agentic way of working: an operating model for teams

How a team works with coding agents every day: give them the right context, constrain them with hooks, permissions and CI gates, verify every change, and roll the model out over 30 days without losing control of what ships.

Read the guide

Agentic ways of working · Intermediate · 15 min read

Skills: teach an agent a procedure once, reuse it everywhere

How SKILL.md works, every frontmatter field Claude Code reads, where skills live, which tools read the open Agent Skills format today, how to write descriptions that trigger, and how to test a skill. Three complete skills to copy.

Read the guide

Agentic ways of working · Intermediate · 19 min read

Hooks: guardrails that run every time

Claude Code hooks run your scripts at fixed points in a session and can block, rewrite or add context. Every current event, matchers, the five handler types, exit codes and JSON output, precedence, debugging, security and seven recipes.

Read the guide

Agentic ways of working · Intermediate · 21 min read

Routines: scheduled agents that open pull requests

Claude Code routines run a saved prompt in the cloud on a schedule, an API call or a GitHub event. Setup, environments, connectors, prompts that run unattended, three real routines, and local options: /loop, cron with claude -p, and GitHub Actions.

Read the guide

Agentic ways of working · Advanced · 16 min read

Subagents: delegate work to isolated agents

Custom subagent files and every frontmatter field, when isolation or parallelism pays, background agents, worktree isolation, four orchestration patterns with prompts, what it costs and how to cap it, and subagents in the Claude Agent SDK.

Read the guide

Agentic ways of working · Intermediate · 18 min read

MCP: connect agents to your tools and data

How the Model Context Protocol works in its 2026-07-28 revision, how to add servers to Claude Code, Claude, ChatGPT, Codex and other tools, how to build one with the official SDKs, and how to keep a server from becoming a back door.

Read the guide

Agentic ways of working · Beginner · 15 min read

Context and memory: what your agent knows, and when

Where to put instructions so agents follow them: CLAUDE.md levels and imports, AGENTS.md across tools, auto memory and app memory, compaction and clearing, and a short team template you can test.

Read the guide

Agentic ways of working · Intermediate · 19 min read

Permissions and safety: let agents act without losing control

Permission modes, allow, ask and deny rules, managed settings, the Bash sandbox, secrets, network egress and prompt-injection defences for Claude Code and similar agents, with safe default configurations you can copy.

Read the guide

Agentic ways of working · Intermediate · 15 min read

Plugins: package and share your agent setup

A Claude Code plugin bundles skills, agents, hooks, MCP servers and commands into one install. plugin.json and marketplace.json, install scopes, pinning versions to a commit, rolling a setup out to a team, and how to vet what you install.

Read the guide

Guides

Craft

The skills underneath: writing prompts that hold up, and testing them like code.

Craft · Beginner · 30 min read

Prompting, from easy to hard

A seven-level ladder from a clear ask to prompts tuned against data. Each level has a before and after, a prompt to copy, an exercise and a way to measure the gain, tied to current vendor docs.

Read the guide

Craft · Intermediate · 24 min read

Experiments and evals

Prove an AI change helps before and after it ships: cheap checks first, a golden dataset with a holdout, judges calibrated against people, small-sample statistics, CI gates, A/B tests, agent evals and today's tools.

Read the guide

Craft · Advanced · 18 min read

Agent ops: see, control and improve agents in production

Trace agent runs with Langfuse, Laminar, LangSmith, Phoenix and OpenTelemetry, control them with permissions, hooks, budgets and kill switches, and improve them with trajectory evals, prompt caching and model routing.

Read the guide

The top of every guide is free. Join free to read the rest.

Join free