Everything in one place
Guides
Complete the 15-minute first task, then use the practitioner pathway to progress from your own work to a tested handover. The other entries support each stage. Every source link goes to a primary source.
Start here
-
Use a chat tool that you already have to turn fictional meeting notes into a checked volunteer update.
The exercise gives you a complete input, a clear output and a deliberate error to find before you trust the result.
Prompt engineering overview, Claude docs ↗Prompting best practices, Claude docs ↗
-
Five practice stages from a checked task of your own to a handover that another person can use without live help.
Each stage creates visible evidence of work, checking and transfer. The route does not treat tool use as certification.
How Anthropic teams use Claude Code ↗Agent Skills overview, Claude docs ↗
-
One or more plain-text files, in the CLAUDE.md or AGENTS.md style, that hold what an organisation would otherwise re-explain to every AI tool.
Every session starts informed, and the context is versioned, reviewable and owned by the organisation rather than by one person's chat history.
CLAUDE.md files, Claude Code docs ↗AGENTS.md, the open format ↗
-
Prompting basics
The smallest useful skill: give context, give an example, and ask for a format.
Most poor results come from a thin request, not from the model. Ten minutes on this changes what every later tool returns.
Prompt engineering overview, Claude docs ↗Prompting best practices ↗
-
What not to paste into a tool, how to check what comes out, and which decisions stay with a person.
The tools are confident when wrong, and organisations hold personal data. Both need rules before habits form.
Guidance on AI and data protection, ICO ↗Anthropic usage policy ↗Anthropic privacy policy ↗
Next
-
Claude Code, Codex and Cursor: an agent that reads and edits your files and runs commands, instead of a chat box you paste into.
This is where AI work stops being copy-and-paste. It is also the tool class that most needs a context layer and clear permissions.
-
An open standard that lets an AI tool connect to your calendar, files, Slack, database or other systems through one kind of plug.
Most useful work needs the tool to see real data. MCP is how that connection is made without a custom integration each time.
What MCP is, modelcontextprotocol.io ↗Connect Claude Code to tools via MCP ↗
-
Automations that run on their own
A scheduled or triggered job that does a recurring task without a person starting it: a weekly report, a daily triage, a sync when a file changes.
The gains from a session fade if a person still has to remember to run the thing. A job on a schedule keeps working after everyone has moved on.
Run prompts on a schedule, Claude Code docs ↗GitHub Actions documentation ↗Zapier help centre ↗
-
Reusable instruction packages an agent loads when a task matches, with optional scripts and reference files bundled in.
A method written once as a skill is used the same way by everyone. It is how a practitioner leaves something behind after a session.
Advanced
-
The adoption problem: one enthusiast in an organisation is not the organisation using AI.
Uplift that stays with one person leaves when they do. Shared context, agreed limits and written methods are what make it stick.
How Anthropic's own staff use Claude Code ↗Skills for enterprise, governance at scale ↗
Links checked 5 September 2026.
Missing something? Propose an entry through GitHub.