Agentic Coding

# Summary This is a repository for Agentic Awesome Skills (AAS), a local agent-first control system for managing and validating skill stacks for AI coding assistants like Claude Code and Cursor. The current release (V15.12.0) introduces AAS Core, which allows coding agents to inspect projects and select specific skills from a complete local catalog, with validation and immutable planning before any changes are applied. The system preserves reproducibility through aas-stack.json manifests and selection evidence, keeping the chosen stack reviewable without uploading repositories. The repository includes 2,007+ skills across development, testing, security, infrastructure, and other domains, distributed through direct installers, specialized plugins for focused workflows, bundles for role-based recommendations, and CLI tools for validation and planning. Best practices emphasized include agent-owned selection without ranking or filtering by AAS Core, comprehensive capability coverage evaluation by agents, and approval-before-writes workflows. The system supports major coding agents including Claude Code, Cursor, Codex CLI, Gemini CLI, GitHub Copilot, and others, with optional advanced features like browser-local Workbench for stack review and choice between full-catalog or specialized plugin installations based on workflow needs.

# Summary Based on the content provided, here is the relevant summary for the user's interests: This article details a comprehensive Claude Code workflow optimization by founder Charlie Hills, informed by Claude's original creator Boris Cherny's recommendation to delete and rebuild your CLAUDE.md system prompt every 6 months. The key practice involves auditing your CLAUDE.md against Anthropic's current prompting guidance using specific structured prompts that fetch live best practices documentation, then iteratively removing obsolete instructions, contradictory rules, and role padding that newer models don't require. The author emphasizes keeping CLAUDE.md lean (under 200 lines per Anthropic's guidance) to preserve token efficiency, model accuracy, and attention weight, since bloated files create context conflicts and reduce adherence. The workflow also covers auditing redundant skills (195 skills down to 65 actually used), implementing hooks as automatic triggers for repeated requests, and organizing project-specific knowledge in separate files that get loaded on-demand rather than cluttering the main system prompt. The overall philosophy treats CLAUDE.md as a front desk receptionist pointing to organized knowledge files rather than a filing cabinet storing everything, with regular maintenance cycles keeping the setup efficient as models evolve.

The Agentic Engineering Academy is a 3.5-week live workshop starting August 11th that covers advanced agentic coding practices highly relevant to your interests. The curriculum explicitly addresses your stated priorities: it includes dedicated modules on efficient agent workflows, context management, agent memory and skills, "vibe coding" strategies versus methodical processes, and best practices for code review and quality evaluation. Week 3 focuses specifically on working with mature codebases—adding features to real, complex projects rather than building from scratch. The workshop covers practical strategies for managing long-running agents, using voice input, choosing models while controlling token costs, and distinguishing when it's safe to skip code review versus when full review is necessary. It uses OpenCode and teaches principles that transfer across Claude Code, Cursor, and other agents. The instructor emphasizes high-level strategies over tool-specific features, and the curriculum includes over 30 real-world projects you can work on during and after the workshop.