Agentic Coding

Over a year of daily practice with Claude Code distilled into a comprehensive guide teaching advanced agentic workflows, not just configuration. The resource covers core concepts including trade-offs between agents vs skills vs commands, internal architecture through 48 Mermaid diagrams, and methodologies like TDD and SDD for maintaining code quality with AI. It includes 271 production templates, a unique threat intelligence database tracking 123 CVEs and 93 malicious skills, 473-question knowledge validation quiz, and security-first workflows for MCP vetting and production hardening. The guide provides structured learning paths for developers at different levels, comparative analysis of when to use which tool, and practical examples showing how to design custom agentic workflows rather than copy-pasting configurations. Complementary resources include an MCP server for querying the guide from Claude Code sessions, 11 whitepapers covering in-depth topics, and 57 reference cards with key concepts.

The content discusses how to properly configure Claude Code with AGENTS.md and CLAUDE.md files for efficient agentic coding workflows. The key technique is using a single-line import (@AGENTS.md) in CLAUDE.md to maintain shared instructions while allowing agent-specific guidance on top, avoiding the "stupid complexity tax" of maintaining duplicate files. The guide explains why Claude Code doesn't read AGENTS.md automatically (a deliberate design choice for deterministic behavior), clarifies that each coding agent (Claude Code, Cursor, Copilot, Gemini) has different capabilities and permission models requiring layered configuration, and provides practical solutions including automation scripts to ensure both files exist across monorepos. This approach prioritizes one source of truth for repo-committed instructions that version with code and reach all developers, rather than syncing custom configurations across individual setups.

The article discusses how agentic coding shifts programming effort from mechanical tasks toward higher-level design and reasoning, but cautions that delegating too much implementation can obscure the design decisions behind the code. When agents handle entire subsystems without human oversight, developers risk losing understanding of the underlying design space—they may have working code without grasping why certain architectural choices were made. The author emphasizes that implementation creates a "forcing function" that exposes gaps in understanding, which agents can bypass. To maintain comprehension, developers should pace implementation in small vertical slices they can absorb and understand, rather than letting agents produce large systems at once. This approach treats reading and questioning as part of the development process, allowing developers to gradually build mental models of the design space. The key insight is that understanding large systems remains expensive even with AI assistance, making selective delegation and deliberate pacing more valuable than maximizing agent output.

The website content provides practical guidance on using Claude Code for refactoring workflows, including skills, agents, and prompts for common tasks like extracting modules, migrating frameworks, and modernizing code patterns. It covers advanced techniques such as using the refactor skill for code splitting, the migration-specialist agent for framework upgrades, and the monorepo-manager skill for identifying and consolidating duplicated utilities. These resources align with optimizing agentic coding workflows through structured agents and reusable skills that maintain code quality while reducing technical debt.