Training & Growth

Claude Code reads files to understand your business. If your files are scattered, duplicated, or poorly named, Claude Code wastes its context window finding information and may use outdated versions. A clean file structure is like a well-organised office: everyone, including AI, finds what they need faster.

The ideal structure separates concerns: a context folder for business knowledge, a skills folder for reusable instructions, a customers folder for client-specific data, and a projects folder for active work. Each folder has a clear purpose and nothing is duplicated. This module provides a complete template you can adopt immediately.

Incrementally. Start by creating the target structure alongside your existing files, then move files one folder at a time, updating any references as you go. Claude Code can help with the migration itself. Never attempt a big-bang reorganisation; it creates more problems than it solves.

Yes. When your CLAUDE.md references specific file paths, every team member needs those files in the same relative locations. Use version control to ensure consistency. If Ross's machine has files in different places than yours, Claude Code will produce different results for each person.

Progressive disclosure means giving Claude Code only the information it needs for the current task rather than loading everything at once. Your CLAUDE.md references detailed files that Claude Code reads on demand. This keeps the main instruction file manageable while making deep knowledge available when required.

Claude Code has a finite context window. A 2,000-line CLAUDE.md consumes valuable space that should be used for the actual task. Progressive disclosure keeps your CLAUDE.md concise and lets Claude Code pull in detailed specifications, brand guidelines, or client data only when the current task requires them.

CLAUDE.md should contain rules that apply to every task: brand colours, spelling conventions, security rules, and critical overrides. Anything that applies only to specific tasks, such as report templates, client details, or technical specifications, goes in separate files that CLAUDE.md references by path.

Yes. Claude Code can read any file in your project when asked or when it determines a file is relevant. But files referenced in CLAUDE.md are read automatically at the start of every conversation. The pattern is: must-know information in CLAUDE.md, should-know information in referenced files, and can-know information in the broader project.

Advanced skills include input validation, error handling, multi-step workflows, and output verification built into the skill itself. A basic skill says 'write a report.' An advanced skill specifies the data sources to read, the analysis to perform, the format to produce, the verification checks to run, and what to do if any step fails.

A skill that breaks complex tasks into sequential phases with checkpoints. For example, a client audit skill might have five steps: gather data, analyse performance, identify issues, draft recommendations, and format the report. Each step has defined inputs and outputs, and the skill specifies what must be verified before moving to the next step.

Use a three-tier testing approach: test with known good input to verify the happy path, test with edge-case input to verify error handling, and test with deliberately bad input to verify the skill rejects it gracefully. Document your test cases alongside the skill so anyone on your team can re-run them.

Refactor when the skill's core purpose is the same but the implementation needs improvement. Create a new skill when you need fundamentally different behaviour. If you find yourself adding conditional branches like 'if this type of client, do X, otherwise do Y,' that is usually a sign you need two separate skills.

Yes. Advanced skills can instruct Claude Code to call APIs, query databases, read spreadsheets, and interact with external services. The skill defines the integration pattern, authentication approach, error handling, and data transformation. This module covers secure integration patterns that keep credentials out of your skill files.

Claude Code does not retain information between conversations by default. Memory systems are files and structures you create so that Claude Code can learn from past work. A MEMORY.md file records key lessons, solved problems, and important decisions. Claude Code reads this at the start of each conversation, effectively remembering what it learned before.

Good notes are written for humans. Memory files are structured for Claude Code to parse and act on efficiently. They use consistent formats, categorise information by topic, and include specific technical details that Claude Code needs. The module teaches you a format that serves both human readers and Claude Code equally well.

Set a line limit and enforce it. When MEMORY.md approaches the limit, move detailed entries into topic-specific files and keep MEMORY.md as a concise index with pointers. The module teaches a pruning discipline: every month, review entries and archive anything that is no longer actively relevant.

Yes, and this is one of the most powerful patterns in the framework. You can instruct Claude Code to append key learnings to MEMORY.md at the end of significant tasks. Over time, the system genuinely learns from experience. The module covers how to structure this self-updating behaviour safely.

Each client project gets its own directory with its own CLAUDE.md, skills, and context files. You run separate Claude Code sessions for each project. Shared resources like brand guidelines and common skills live in a central location referenced by each project's CLAUDE.md. This module covers the architecture for managing ten or more projects cleanly.

Not if your file structure is correct. Each Claude Code conversation operates within a specific project directory and reads that project's CLAUDE.md. Cross-project contamination happens when people use a single flat directory for everything. Proper project isolation, which this module teaches, prevents it entirely.

Use shared skill files and a global CLAUDE.md for universal rules. Project-specific CLAUDE.md files inherit the global rules and add project-specific overrides. When you improve a skill, every project benefits immediately. This is the same pattern used by software teams managing shared libraries.

Open separate terminal windows, one per project. Each session loads its own context independently. There is no warm-up penalty for switching because Claude Code reads the CLAUDE.md fresh each conversation. The module covers practical workspace layouts for managing multiple concurrent projects efficiently.

A structured document that captures everything Claude Code needs to know about a specific aspect of your business: your services, pricing, team structure, client relationships, market positioning, or operational processes. It is your institutional knowledge written in a format that Claude Code can use to make informed decisions.

Share everything that is relevant to the tasks you want Claude Code to perform, excluding sensitive credentials. If Claude Code writes client emails, it needs to know your service offerings and tone. If it generates reports, it needs to know your analysis methodology. Incomplete context produces generic output.

Use clear headings, structured tables, and explicit rules rather than narrative prose. Claude Code parses structured content more reliably than flowing paragraphs. State facts directly: 'Primary market: United Kingdom. Currency: GBP. Date format: DD/MM/YYYY.' This module provides templates for every common context type.

Whenever the underlying facts change. Add a review schedule to your operations: monthly for client details, quarterly for service offerings, and immediately for any significant business change. Outdated context files produce outdated outputs. The module includes a maintenance checklist to keep your context current.

Yes, and it is excellent at it. You can have a conversation where you describe your business verbally and Claude Code structures the information into properly formatted context files. You then review and correct any inaccuracies. This is one of the most efficient ways to bootstrap your business knowledge base.

Claude Code can work with APIs, databases, spreadsheets, email systems, cloud storage, analytics platforms, advertising platforms, and most services that have a command-line interface or API. This module covers the most common business integrations: Google Workspace, CRM systems, analytics, and marketing platforms.

For basic integrations like reading Google Sheets or querying a database, no. Claude Code can help you write the connection scripts. For complex integrations involving OAuth flows, webhook endpoints, or custom APIs, you may need some technical support for the initial setup. This module teaches you to assess which integrations you can handle yourself.

Store all credentials in environment variables or dedicated credential files that are never committed to version control. Use service accounts with minimum necessary permissions. Never put API keys in your CLAUDE.md, skill files, or any file that might be shared. This module covers secure credential management patterns in detail.

Your integration breaks, and Claude Code will report the error. The advantage of having Claude Code manage your integrations is that it can often diagnose and fix API changes faster than doing it manually. Keep your integration patterns documented in specification files so Claude Code knows the expected behaviour and can identify what changed.