Agentic Coding

The guide provides practical steps for creating and managing Claude Code skills to optimize workflows. Key sections cover: opening the skills section in Claude, creating new skills by describing them to Claude for automatic generation, and running skills through the chat interface. The content details official skills across categories like creative design, document processing, technical development, and enterprise communication, with specific examples such as frontend design skills using React 18 and TypeScript, PDF/Excel processing capabilities, and MCP server construction. Additionally, it lists 20 community-created skills from GitHub organized by function—including development tools like the debug-skill for breakpoints and variable inspection, data analysis skills for crypto and ecommerce, scientific research tools with 125+ bioinformatics skills, and security-focused skills covering OWASP Top 10 2025 and agentic AI security. The guide also explains converting documents into reusable skills using Kimi, allowing SOPs and internal guidelines to become structured, actionable instructions that work alongside built-in skills for improved task accuracy and workflow efficiency.

Claude Code workflow optimization with persistent memory using Obsidian Zettelkasten and Graphify knowledge graphs. The setup eliminates session amnesia by storing project decisions, architecture, and progress in a centralized vault accessible to Claude Code through custom skills and CLAUDE.md instructions. Graphify transforms your codebase into a queryable AST-based knowledge graph that persists across sessions, reducing token consumption by querying structured code relationships instead of re-reading files. The complete workflow integrates chat import pipelines to index Claude conversations, /resume and /save commands for session continuity, and auto-save hooks that capture session facts when you exit. A Git hook automatically rebuilds the graph on commits, and the system supports multiple projects within a single vault with cross-project knowledge linking through Zettelkasten wikilinks.

In the world of software development, the .md file serves as a strategic blueprint for projects, functioning as the AI's "long-term memory" in Claude and as the architectural map in Cursor. A high-impact .md file should be structured with a clear North Star (vision and goals), Context (guardrails and audience), Roadmap (feature checklists), Glossary (project-specific terminology), and Evolution Log (change history). The PRD (Product Requirements Document) is the most critical .md file type, acting as a contract between vision and execution by defining what's being built, why, and what success looks like, which removes guesswork and drastically reduces the need for rewrites. To maximize effectiveness, use the .md file as a central hub by linking to related artifacts like JIRA tickets, Confluence pages, and Figma files, creating a contextual corpus that connects vision to execution. The most effective .md files employ storytelling—framing projects as narratives with logical flow—which helps AI understand intent rather than just following commands, leading to smarter suggestions. Treating the .md file as a living document that evolves with the project ensures the source of truth remains current and clear as the project scales.