Agentic Coding

Agent Skills is a production-grade framework of 24 structured workflows designed to enforce senior engineering practices in AI coding agents like Claude Code and Cursor. The system maps to a six-phase development lifecycle (Define, Plan, Build, Verify, Review, Ship) through eight slash commands that activate relevant skills automatically—for example, /spec enforces spec-before-code discipline, /build auto-generates plans and implements tasks autonomously with test-driven verification, and /review applies five-axis code quality gates. Each skill encodes anti-rationalization tables that counter common shortcuts (like "I'll add tests later") and includes mandatory verification checkpoints, ensuring agents follow the same practices that senior engineers use for production code. The framework integrates natively with Claude Code (via marketplace), Cursor, Gemini CLI, Antigravity, Codex, Copilot, and 70+ other agents as plain Markdown. Key skills relevant to agentic coding patterns include context-engineering for managing agent memory and information flow, incremental-implementation for thin vertical slices, test-driven-development with enforced red-green-refactor cycles, and source-driven-development to ground framework decisions in official documentation. Installation is one command (npx skills add addyosmani/agent-skills), with optional granular selection, and the system includes specialist personas (code-reviewer, security-auditor, web-performance-auditor) for targeted reviews.

This repository provides 17 Agent Skills focused on context engineering and agentic system design, directly applicable to Claude Code and Cursor workflows. Key skills include context-fundamentals (managing language model context windows and attention mechanics), memory-systems (short-term, long-term, and graph-based architectures), multi-agent-patterns (orchestrator, peer-to-peer, and hierarchical designs), and long-horizon-prompting (writing task briefs for autonomous agents with success predicates and audit gates). The repository emphasizes progressive disclosure for efficient context use, platform-agnostic principles transferable across Claude Code and Cursor, and includes practical examples like the digital-brain-skill (a personal operating system with 6 modules demonstrating context optimization and memory management) and llm-as-judge-skills (production-ready evaluation tools). Installation is available as a Claude Code plugin or Cursor Open Plugin, with individual skills loadable on-demand based on task context. The collection addresses workflow optimization through context compression strategies, tool-design patterns, filesystem-based context discovery, and self-improvement loops where the agent harness itself becomes the optimization target—all designed for solo entrepreneurs and SaaS builders working with AI coding assistants daily.

planning-with-files is a persistent file-based planning skill for AI coding agents that implements the Manus-style context engineering pattern Meta acquired for $2 billion. It maintains three core files—task_plan.md, findings.md, and progress.md—on disk to survive context loss, /clear commands, and crashes, with lifecycle hooks that re-inject plans before tool use and verify completion before stopping. The latest v3.0.0 release introduces autonomous and gated modes for long-running agentic runs with an opt-in completion gate, along with a structured run ledger; existing setups remain unchanged. The skill installs across 60+ agents via the SKILL.md standard and is available in five languages, supporting Claude Code, Cursor, GitHub Copilot, Codex, Gemini CLI, Mastra Code, Kiro, and many others. Recent versions added Windows compatibility fixes, parallel plan isolation via slug-mode directories, session recovery after /clear, and native Codex PreCompact hook support, achieving 96.7% pass rate on file-pattern fidelity benchmarks.

Oh-My-ClaudeCode (OMC) is a multi-agent orchestration framework for Claude Code that enables efficient agentic coding workflows without requiring learning curve. Key features directly relevant to your interests include: Team Mode as the canonical orchestration surface with staged pipelines (plan → PRD → execute → verify → fix), specialized agents for architecture/design/testing, automatic skill extraction from debugging sessions into reusable .omc/skills/ files that auto-inject when relevant, and natural language interfaces like /autopilot, /ralph (persistence mode with verify/fix loops), and /deep-interview for requirements clarification. For memory management, OMC supports multi-repo workspaces with shared state directories, persistent session artifacts in .omc/, and intelligent model routing (Haiku for simple tasks, Opus for complex reasoning) to optimize token usage. The framework includes CLI workers for cross-validation (Codex, Gemini, Antigravity, Cursor) via `omc team N:provider` commands, HUD statusline for real-time observability, and analytics tracking to understand token usage across sessions—all designed for solo developers and small SaaS founders working with Claude Code daily.