Claude Code Workshop for Teams: Agenda, Labs and Outcomes
Short answer: A Claude Code workshop for teams should teach the features that change daily work, namely CLAUDE.md context, plan mode, permission settings, hooks, custom commands and subagents, through labs on the team's own repository. A one-day format with six labs and a shared settings file is enough to get a team of up to 12 engineers productive and consistent.
On this page
What a Claude Code workshop should cover
Claude Code is Anthropic's agentic coding tool that runs in the terminal, in IDE extensions and in the browser. It reads the repository, edits files, runs shell commands and works through tasks in a loop. A team workshop is not about learning commands; most engineers figure those out in an hour. It is about agreeing how the whole team configures and uses the agent so results are predictable.
The topics that matter most for teams are shared project memory in CLAUDE.md, checked-in settings that define which commands the agent may run without asking, plan mode for larger changes, hooks that enforce formatting or tests, reusable slash commands for recurring workflows, and subagents for parallel or isolated work. Everything else can be learned on the job.
Agenda for a one-day team workshop
This agenda assumes 6 to 12 participants, a working local build and a few prepared backlog tasks. Each block mixes a short demo with hands-on time.
- Setup and permissions (45 min): install, log in with the team plan, review permission modes, check in a shared settings file with an allowlist for safe commands such as the test runner.
- CLAUDE.md that actually helps (60 min): build commands, architecture map, coding conventions and known traps. Test it by asking the agent to explain the codebase back.
- Lab block one (90 min): bug fixes and small refactors with a failing test first.
- Plan mode and task slicing (45 min): turning a ticket into a plan the team can review before any code is written.
- Lab block two (90 min): a vertical feature slice in pairs, with one person driving and one reviewing.
- Automation (45 min): hooks, custom slash commands, headless runs in CI.
- Conventions and retro (45 min): write down what the team will and will not delegate.
Six labs that work on real repositories
Labs should be drawn from the team's own backlog. The pattern below gives a range of difficulty and teaches a distinct skill in each lab.
- Explain and map: ask Claude Code to document an unfamiliar module and compare the result with what the module owner knows.
- Test-first bug fix: write a failing test from a bug report, then let the agent fix it until the test passes.
- Mechanical refactor: rename a concept across the codebase or replace a deprecated API, verified by the existing test suite.
- Feature slice with plan mode: review the plan as a team, adjust it, then execute.
- Custom slash command: package a recurring workflow, for example creating a migration plus tests, into a reusable command.
- Hook guardrail: add a hook that runs the linter or blocks edits to protected paths, then watch it trigger.
Need a hand? neooutput helps with exactly this. Book a call for free and get a clear plan in 15 to 30 minutes.
Team configuration worth checking in
The most durable output of a Claude Code workshop is configuration in the repository, because it keeps working after the trainer leaves. Commit a project-level CLAUDE.md, a shared settings file with permission rules, and a folder of custom commands. Keep personal preferences in user-level files so they do not leak into the team setup.
Treat these files like code: review changes in pull requests, keep CLAUDE.md short and factual, and remove instructions that the agent ignores or that no longer apply. A bloated context file costs tokens on every request and dilutes the instructions that matter.
Common mistakes teams make with Claude Code
The first mistake is granting broad permissions on day one to avoid prompts. Start with an allowlist of read-only and test commands and widen it only when the team understands what the agent does. The second is handing over vague tickets and blaming the tool for vague results. The third is skipping review because the tests passed: agents can write tests that confirm their own wrong assumptions.
A fourth mistake is running one giant session for a large feature. Long sessions drift as context fills up. Splitting work into tasks of an hour or less, each with a clear done criterion, produces cleaner diffs and easier reviews.
Outcomes to expect and how neooutput runs it
By the end of the day, each participant should have merged or opened at least one pull request created with Claude Code, the repository should contain the shared configuration, and the team should have a written list of task types it delegates. After 30 days, check whether those conventions survived and update them in a short follow-up session.
neooutput delivers Claude Code workshops remote or on-site, in English or German, with AI Code Engineers who use the tool in production projects every week. Labs are prepared from your backlog in a short scoping call beforehand. For apps that were built with Claude Code and now need a reliable path to the App Store and Google Play, apprat.io offers a Claude Code-to-App module with human review of every change. A free 15-minute call with neooutput clarifies scope, format and preparation.
Frequently asked questions
Is Claude Code only for terminal users?
No. Claude Code started as a command-line tool, but it is also available through IDE extensions such as VS Code and JetBrains and in a web interface. The concepts taught in a workshop, including CLAUDE.md, permissions, plan mode and hooks, apply across these surfaces, so mixed teams can train together.
What should go into CLAUDE.md?
Keep it short and practical: how to build, test and lint, a brief architecture overview, naming and style conventions, and known pitfalls such as generated folders that must not be edited. Avoid long essays. Ask the agent to summarize the project after reading it to check whether the file actually conveys what matters.
How many people can join one Claude Code workshop?
For hands-on labs on a real codebase, 6 to 12 participants per trainer works well. Larger groups can attend a demo-heavy kickoff, but labs need someone available to unblock setup issues and review results. Pairing participants keeps groups manageable and doubles as review practice.
Can Claude Code run in CI pipelines?
Yes. Claude Code supports a non-interactive mode that can be called from scripts and CI, for example to triage issues or draft fixes. Teams should restrict permissions tightly in that mode, store credentials in the CI secret store, and require human review before anything it produces is merged.
