Agentic Coding

Agentfiles is an AI skills manager for Obsidian that enables browsing, creating, and managing skills, commands, and agents across 17 coding tools including Claude Code, Cursor, Codex, and Windsurf. The plugin supports advanced workflows by centralizing skill management, offering inline editing with markdown preview, a skills marketplace for importing pre-built skills, and conversation history browsing for Claude Code sessions. It includes optional analytics via skillkit CLI to track context usage and agent health metrics, directly addressing code context management and workflow optimization for developers using multiple AI coding agents.

Boldare embeds Claude Code across the full software development lifecycle—from discovery and design through testing and deployment—treating it as part of a broader AI-augmented delivery system used daily by senior engineers rather than an isolated tool. They handle complex challenges like codebase analysis, subagent orchestration, legacy refactoring, and DevOps optimization, integrating Claude seamlessly into CI/CD pipelines with production stacks like Laravel, Python, React, and Nuxt, while using MCP (Model Context Protocol) to connect Claude to client systems and internal tools without custom integration code. When selecting a Claude Code partner, strong practices include using Claude across codebase exploration, planning and refactoring, testing and QA, and delivery pipelines—not as a standalone productivity hack. Red flags include treating "better prompts" as the primary strategy for handling complexity, focusing only on speed rather than long-term quality and maintainability, and lacking MCP integration or experience building MCP servers for production use. The difference between using AI and actually benefiting from it depends on engineering maturity: balancing speed with quality requires clear rules on where AI is trusted versus reviewed, testing built into the workflow, and AI workflows adapted to team size and existing processes rather than forcing adoption of new tools or platforms.

Graft is a tool that optimizes AI agent coding workflows by building persistent knowledge graphs of codebases. It reduces token usage by 42%, tool calls by 46%, and latency by 60% while improving correctness from 54% to 66% on SWE-bench tasks. The system creates linked markdown nodes documenting code structure, dependencies, and functionality—written once and reused across sessions—eliminating the repeated exploration overhead agents typically face. Graft integrates natively with Claude Code, Cursor, and other AI coding assistants through MCP servers and instruction files, automatically syncing graphs after edits. It supports 21 programming languages and works with monorepos, submodules, and multi-repo layouts. The structural parsing uses tree-sitter (deterministic, no model cost), while optional LLM summaries are cached and regenerated only when code changes, making graph maintenance cheap and always fresh.