Agentic Coding

The Claude Code Power User Tips guide covers advanced workflow patterns for shipping code faster, with several practices directly relevant to agentic coding and developer productivity. Key techniques include running 3-5 parallel Claude sessions using worktrees for isolation, and using /batch to fan out large migrations across dozens of worktree agents that test independently and create PRs. For planning complex tasks, the guide recommends using plan mode (Shift+Tab) to refine architecture before implementation, with Opus as the recommended model for its superior reasoning and tool use despite being slower per token. The guide emphasizes verification as the single most impactful practice—giving Claude ways to check its own output through domain-specific tools like the Chrome extension for frontend work or test suites. For managing knowledge across sessions, the team uses CLAUDE.md files checked into git to capture mistakes and rules, combined with auto-memory that consolidates preferences between sessions. Custom skills in .claude/skills/ and subagents in .claude/agents/ automate repeated workflows, with examples including /batch for migrations, /loop for recurring local tasks, and specialized agents for code review, validation, and simplification. Additional productivity patterns include using /btw for sidebar questions during active work, /simplify for parallel code quality reviews, and prompting techniques like "Grill me on these changes" to force validation before shipping.

The .claude/ folder structure is critical for effective Claude Code work. A well-organized setup includes CLAUDE.md (project instructions under 200 lines), rules/ (modular guidelines for code style, testing, API conventions), commands/ (repeatable workflows as slash commands), skills/ (auto-triggered capabilities), agents/ (isolated subagent personas with custom tools), and settings.json (permission controls). The two-layer system—project-level .claude/ (shared with team) and personal ~/.claude/ (global across repos)—allows both consistency and individual flexibility. Production-ready Claude Code requires more than just prompts; it's infrastructure that prevents Claude from consuming excess context, repeating mistakes, and losing project conventions across sessions. Key practices include running /init on day one, organizing the .claude/ directory properly, writing clear skill descriptions for auto-triggering, using the memory hierarchy, and adding hooks for critical rules since hooks are enforced while memory is advisory.

Claude Code Skills are structured instruction sets that optimize AI behavior for specific tasks, moving beyond generic outputs to tailored results. Nick Babich is running a masterclass on November 30 covering Skill anatomy and structure (including SKILL.md frontmatter and body), differences between Skills, Plugins, and Mods, ready-to-use workflow automation examples for UI design reviews and PRD-to-user-flow conversions, Figma Skills integration, and multi-step agentic workflows that chain Skills together—such as automatically updating usability screeners when a PRD changes. The masterclass also covers three valuable community Skill repositories: Agent Skills, Humanizer, and Claude Code Security Review, with a 40% discount available.

Claude Cowork is a new knowledge work agent that automates multi-step tasks across local files and connected apps like Slack and Google Drive, allowing you to focus on strategic work while Claude handles deliverables with citations. The product guide covers core capabilities including local file access, subagents, and long-running scheduled tasks, plus a matrix for choosing between chat for conversational drafting, Claude Code for coding, and Claude Cowork for cross-app knowledge work. Seven common workflows are documented including research briefs, meeting prep, and recurring reports, with plugins available for specific knowledge work in marketing and product management.