Agentic Coding

This cheatsheet provides practical optimization techniques directly relevant to your workflow as a solo developer using Claude Code daily. For Claude Code workflow optimization, key tips include: using `/status` and `/compact` to manage context (watch context usage >70%), switching thinking modes with Alt+T to balance speed and cost, and leveraging Plan Mode (Shift+Tab × 2) for complex tasks before executing changes. The cheatsheet covers permission modes (Shift+Tab to cycle), essential commands like `/goal` for autonomous task completion and `/loop` for recurring checks, and context recovery strategies. For efficiency, use `/simplify` to detect over-engineering and `/batch` for large-scale refactors via parallel agents. For agentic coding and debugging, the resource details background agents (working while you code), agent teams for multi-agent coordination, and custom agents via `.claude/agents/`. The `/debug` command provides systematic troubleshooting. Advanced features include skill evals (capability uplift or encoded preferences), Tasks API for persistent multi-session projects with dependencies, and MCP servers like Serena for semantic search and memory. The cheatsheet emphasizes the "less scaffolding, more model" philosophy and provides prompting formulas: specify WHAT (deliverable), WHERE (file paths), HOW (constraints), and VERIFY (success criteria).

Claude Code workflow optimization includes installing pre-built agent skills and plugins from the marketplace — 425 plugins and 2,810 skills organized by category. For solo developers, the marketplace offers specialized skill packs like the cursor-pack (30 skills for AI code editing), skill-creator (for building production-grade agent skills), and devops-automation-pack (25 DevOps plugins). Key optimization approaches include using PRD documents with agent skills, auto-invoking skills based on context triggers rather than manual slash commands, and leveraging multi-agent orchestration plugins like geepers (51 specialized agents) or sprint (spec-driven autonomous development framework). Advanced debugging strategies are supported through plugins like code-cleanup (removes dead code and weak types across 11 quality dimensions), conversational-api-debugger (analyzes REST API failures with root cause analysis), and security-agent (code analysis subagent). For AI pair programming, the marketplace emphasizes skills that activate automatically when Claude detects relevant context — rather than explicit commands — allowing seamless workflow integration through SKILL.md instruction files with trigger phrases and allowed-tools scoping.

The repository contains comprehensive documentation of Claude Code's system prompts, agents, and tools—resources directly relevant to your interests in agentic coding and AI pair programming. Key highlights include detailed agent prompts for specialized coding tasks like /code-review (with multiple effort modes), /batch (for parallel codebase changes), and /security-review. The repository documents how Claude Code implements autonomous agents through subagents like Explore and Plan mode, along with memory consolidation and background job execution—capabilities useful for solo developers automating workflows. System prompts cover autonomous loop behavior for background execution, learning mode with human collaboration, and practical tools like TodoWrite for task tracking and workflow orchestration. The resource also references Piebald, a platform designed specifically for agentic AI development experiences, and provides access to tweakcc for customizing individual system prompt pieces. For optimization, the documentation shows how to use skills for specialized workflows (/loop for recurring tasks, /schedule for cloud-based job scheduling) and demonstrates best practices for prompt engineering through subagent delegation patterns and memory-based context management across sessions.