What this is
Tool focus: Anthropic Claude Code
Workshops on Claude Code (Anthropic CLI agent) for UK engineering teams: workflow design, hooks, MCP servers, sub-agents and review discipline.
Claude Code workflows that survive a real week.
Claude Code is a terminal-resident coding agent from Anthropic. It is fast, opinionated, and rewards teams that build careful workflows around it.
Hooks intercept every tool call. Sub-agents run in the background.
Workflow first, prompt second
The workshop starts with what the team actually does week by week — feature work, refactors, bug triage, doc updates — and only then maps Claude Code onto each. Prompt patterns are the consequence of the workflow, not the starting point.
Hooks and slash commands
Hooks are used to enforce gates the team cares about: pre-commit checks, format on save, custom validators. Slash commands are wired up for the small set of recurring tasks where the prompt would otherwise be re-typed.
Sub-agents for boring work
Long-running, low-stakes work — research, repo audits, dependency scans, test triage — is delegated to sub-agents with their own prompts and budgets. The main session stays focused on the change actually under review.
MCP servers where they earn their keep
A short list of MCP servers is wired up where the team genuinely needs them: a private docs lookup, an internal API tool, a query bridge. Anything else is left out. MCP is a force multiplier, not a checklist.
Memory and context discipline
CLAUDE.md and per-feature context files are written deliberately, not accreted. The training covers what belongs in memory, what belongs in the prompt, and what belongs nowhere — a common failure mode when teams adopt a CLI agent for the first time.
How an engagement runs
Day 0
Read-only review
Repo, CI, and current developer workflow are mapped. We agree the first hooks, sub-agents, and MCP servers worth wiring up.
Day 1
Workflow design
Half day on workflow design, hooks, slash commands, sub-agents, MCP server short list, and the written review patterns.
Day 2
On the real repo
Half day inside the team’s real codebase. Real Claude Code sessions on the agreed tasks. Hooks fire, sub-agents run, diffs land in review.
Weeks 2 to 4
Rollout window
Light coaching, an evaluation suite carried by the team, written rules of engagement, and a check-in to confirm the workflow holds in real weeks.
Common questions
How is this different from training people on ChatGPT?
Claude Code lives in the terminal and reads the repository. The training is about workflow design and review discipline around an agent that has access to the codebase, not about prompt phrasing for a chat box.
Do you build custom MCP servers?
Yes, where the team has a genuine internal tool worth surfacing. MCP server design has its own page; on the workshop we focus on wiring up servers that already exist or that the team can build in a day.
Will this work in a regulated environment?
It is built for one. Read-only access first, named review gates, hooks that enforce policy, and the M365 and identity layer underneath stays part of the engagement, not an afterthought.