Agentic Coding

This repository documents advanced Claude Code practices directly aligned with your interests. The core workflow pattern converges on Research → Plan → Execute → Review → Ship, with specialized sections on using PRD documents, AGENTS.md files, and Skills for efficient agentic engineering. For context management, the guide emphasizes keeping sessions under 300-400k tokens before context rot occurs, maintaining context below 40% for optimal performance, and using subagents to isolate intermediate work and reduce main context pollution. Key best practices include starting with plan mode and minimal specs, breaking PRDs into vertical slices for early feedback, using phase-wise gated plans with multiple tests, and leveraging separate Claude instances for code review via test time compute. The repository includes 83 tips covering prompting strategies like "fix" without micromanagement, session management techniques such as /compact with hints and /rewind for failed attempts, and debugging approaches using agentic search (glob + grep) over RAG. Advanced patterns include orchestration workflows combining Commands → Agents → Skills, cross-model flows with other AI providers via routers and MCPs, and skill collections (140k+ stars in Anthropic's official skills repo) organized with gotchas sections and progressive disclosure folders. Featured videos from Boris Cherny and others cover scaling from vibe coding to agentic engineering and workflow automation.

The website content from easy-vibe is relevant to your interests. The Advanced Developers section includes dedicated Claude Code materials covering MCP integration, Skills packaging, Agent Teams coordination, long-running task management, and workflow best practices for refactoring and code review. The site also features a "Claude Code workflow best practices" guide and documentation on spec coding as an evolution from vibe coding. Additionally, there's a "Claude Agent SDK complete guide" for building custom agent workflows. The content specifically addresses engineering-grade development practices and provides cross-platform project tutorials, which aligns with your focus on advanced AI agent usage, PRD workflows, and best practices for tools like Claude Code and Cursor.

When debugging AI-generated code, the approach has fundamentally shifted from using traditional debuggers to communicating errors in plain English with agents, which find and fix bugs faster than manual debugging. The engineering job itself has transformed, with writing code by hand becoming the least valuable skill; instead, product sense, the ability to clearly express intent to agents, and reviewing agent output now define strong engineering. Spec-driven development using structured markdown files has become central to agentic workflows, requiring tools designed around steering agents and reviewing their output rather than writing code manually. Cross-repository context is essential for agents working on features spanning multiple repos, making a dedicated spec repository that documents connections between codebases a promising approach. Skills are becoming reusable building blocks like libraries, but tooling for discovering, cataloging, sharing, and testing them remains immature, with emerging solutions like Port's skills catalog and Microsoft's Agent Package Manager beginning to address this gap.

Based on the user's interests in advanced Claude Code workflows, PRD best practices, agentic debugging, and context management, here's the relevant content from this resource: The curated list covers 80+ CLI coding agents and orchestration tools. For Claude Code specifically, relevant projects include Claw Code (192k stars), a Python/Rust rewrite of Claude Code architecture, and Claurst (9.7k stars), a Rust rewrite that incorporated discoveries from Claude Code's source including KAIROS persistent assistant and buddy system features. For PRD and workflow management, the list includes Plandex (15.4k stars), a plan-first CLI agent supporting 2M token context for structured multi-file development, and projects like Roo Code CLI offering architect/code/debug/orchestrator modes with skills and checkpoints. Regarding debugging and context management, agentic tools like agent-lsp provide LSP integration with speculative execution for preview edits and measured 5-34x token savings, while projects like Nex offer persistent memory and knowledge graph integration across agent sessions. For advanced vibe coding, tools like CodeMachine-CLI and VibePod are explicitly vibe-coding oriented, and orchestration platforms like claude-flow, Bernstein, and Aeon enable autonomous loops with multi-agent coordination, quality scoring, and self-healing capabilities.