Agentic Coding

Based on the content provided and the user's specific interests in agentic coding, PRD workflows, and advanced Claude Code practices, here's the relevant summary: This repository documents comprehensive Claude Code workflows and best practices for agentic engineering, including the Command → Agent → Skill architectural pattern demonstrated in the /weather-orchestrator workflow. Key resources for advanced users include structured guidance on using PRD documents efficiently, building feature-specific subagents with progressive skill disclosure, establishing CLAUDE.md configurations (recommended under 200 lines), and leveraging .claude/rules/ for domain-specific instructions. The content covers practical tips for workflow optimization such as breaking PRDs into vertical slices for end-to-end feedback, using subagents for context management, implementing phase-wise gated plans with multiple test phases, and managing context degradation around 300-400k tokens. Advanced agentic patterns include test-time compute with separate context windows for parallel debugging, cross-model workflows for code review validation, and automation of recurring development tasks through commands and hooks. Multiple video resources from Claude Code creators like Boris Cherny and Matt Pocock are referenced for learning advanced "vibe coding" techniques.

After 30 days using Claude Code on a 40k-line Rails SaaS app, a developer found it most effective for boilerplate generation, test writing, and code documentation rather than architectural decisions or performance debugging. The tool excels at generating RSpec test cases and explaining unfamiliar code when given clear direction, but struggles with context window limitations on large codebases and occasionally provides confident but incorrect suggestions that require verification. Key workflow insight: treat Claude Code as a fast junior developer who needs clear direction and fact-checking rather than a replacement for deep understanding, yielding approximately 20-25% speed improvements on suitable tasks like test coverage expansion and clearing backlogged tickets.

The user seeks advanced guidance on Claude Code workflows and agentic coding best practices. This article directly addresses their core interests. The author describes a complete Claude Code workflow that begins with establishing project standards through CLAUDE.md (a configuration file documenting build commands, coding conventions, and architecture decisions) and organizing rules in .claude/rules/ with topic-specific markdown files assigned to code paths via YAML frontmatter. Before implementation, the workflow includes planning phases using skills like brainstorming and "Grill Me" to identify gaps, followed by creating a PRD and detailed implementation plan with vertical slices—semantically cohesive, independently reviewable units that prevent unwieldy pull requests. Quality checks run via Stop hooks with subagents, and code review uses the "requesting-code-review" skill to verify output against the plan. The author emphasizes that Claude Code's output quality depends directly on context provided, and that developers should focus on establishing proper documentation and structure rather than crafting perfect prompts. The workflow fundamentally shifts a developer's role toward product engineering, spending more time on upstream design decisions about what to build and why rather than pure implementation.