Generative AI

As you build with generative AI, the AI keeps getting better — but is it okay if you don't? Guild Plugin cares not just about the codebase and the agent organization, but also about the growth of the person using it. Here's a look at the third growth engine and the safety model that underpins self-growth.

Most Skills only improve the output. Guild Plugin reads the traces left behind during development to grow the organization that does the developing — through habits, knowledge, and rules. This post breaks down that growth engine.

What changes when, instead of handing everything to a single AI agent, you build a role-divided "organization" to develop instead? Here's a rundown of the leader and specialist roles, how they collaborate, and the development flow that ties it together.

AI coding tools accumulate output, but the tool itself never actually grows. This post walks through how I tried to solve that by borrowing an idea from how people work together — an "organization."

What problem did each of Prompt, Context, Harness, and Loop Engineering set out to solve? Starting from LLM as the baseline, this post traces how the limits of each approach gave rise to the next — following the thread of evolution from one stage to the next.

Sharing the design principles and implementation process behind skill-quality-plugin, which evaluates Claude Code skill files against 38 criteria. Covers deterministic problems caught by rules, semantic problems judged by models, and the sub-agent isolation pattern.

Learn how to separate work and personal Claude Code accounts on a single PC. Split your settings with CLAUDE_CONFIG_DIR, solve the macOS Keychain sharing issue with CLAUDE_CODE_OAUTH_TOKEN, and configure .zshrc so the account switches automatically based on your working directory.

Learn how to hide the Co-Authored-By trailer that Claude Code automatically adds to commits and the Generated with Claude Code line on PRs. Covers how to use the attribution setting instead of the deprecated includeCoAuthoredBy.

Analyzing the AI engineering techniques applied to the SDD (Spec-Driven Development) Plugin for Claude Code through 4 paradigms: Prompt, Context, Agentic, and Harness. See how each paradigm is implemented in the plugin with actual code examples.

The 4 phases of building the SDD (Spec-Driven Development) Plugin for Claude Code — foundation, multilingual support, Issue template improvements, and the resume command. Sharing the development process and lessons learned.

SDD (Spec-Driven Development) Plugin is a Claude Code plugin that helps you collaborate systematically with AI through a GitHub Issue-based 4-stage process (Analyze → Design → Implement → Test). This post covers its commands, GitHub integration, and usage.

The problems encountered when using AI coding tools like Claude Code in practice (coding without analysis, implementation without design, lost conversations), and the background behind creating the SDD (Spec-Driven Development) Plugin to solve them.