Agentic Coding

Claude Code Guide covers advanced AI agent development, context engineering, and memory systems directly relevant to your workflow. Key sections include Agent Harness Engineering for structuring AI-assisted work with proper context and constraints, Memory Systems for cross-session knowledge retention, and Context Engineering for optimizing what information Claude receives. The guide provides reusable templates, CLAUDE.md project instruction formats, and workflow patterns for test-driven development, spec-first development, and plan-driven development. For debugging and optimization, it includes Agent Evaluation methods, Observability guides for tracing agent behavior, and detailed workflows for code review and exploration. The repository includes runnable examples, cheat sheets, and an MCP server integration allowing you to query the guide directly from Claude Code or Cursor.

Based on the website content from this Hacker News discussion, there is relevant information about agentic coding and Claude Code memory management, though it's primarily discussion rather than news about new developments. The content discusses Claude Code's recent support for AGENTS.md files as an alternative to CLAUDE.md, enabling provider-agnostic agent instruction files. Users share practical workflows for managing AI agent configurations across multiple tools using AGENTS.md and CLAUDE.md files, including techniques like symlinks and nested configuration approaches. The discussion covers memory management strategies such as keeping instruction files minimal to preserve context, using tiered AGENTS.md by model capability level, and tactical use of skills files to avoid context bloat. Best practices mentioned include using AGENTS.md for project-specific gotchas and workflows, having agents conduct research phases before implementation, breaking complex tasks into verifiable chunks, and maintaining separate instruction files for different work patterns. However, users note ongoing frustrations with fragmented tooling ecosystem, missing support for .agents/skills standardization, and inconsistent configuration handling across different Claude Code versions.

Claude Cowork extends the agentic architecture from Claude Code to knowledge work beyond coding, allowing you to describe multi-step tasks and let Claude execute them autonomously in the cloud. Key features for optimizing your workflow include three permission modes (Manual, Auto, and Skip) to control how Claude requests approval before taking actions, global and folder-level instructions to provide standing context across sessions, and scheduled tasks that run automatically without requiring your computer to be active. Claude maintains visibility into its planning and actions throughout execution, letting you steer when necessary or step away until tasks complete, with cloud-based sessions that persist across desktop, web, and mobile surfaces and integrate with your connected apps and browser through managed connectors.