ykdojo/claude-code-tips

1. This repository is a documentation-and-tooling package centered on Claude Code workflows. The primary “product” is the long-form tip guide in README.md, with supporting articles under content/, reusable utilities under scripts/, and Claude Code automation packaged as skills/ plus a plugin manifest.

2. The content layer is mostly markdown and HTML. README.md is the main entry point and table of contents for the tips. Additional standalone posts live in content/*.md, and a few slide/demo pages live in content/*.html. These are presentation assets rather than application logic.

3. assets/ contains images and media referenced by the README and content pages. It supports the narrative and demo sections, but does not drive any runtime behavior.

4. The scripts/ directory is the operational core. It contains shell and Node utilities for status-line rendering, context management, conversation cloning, setup, previewing colors, and generating a table of contents. These scripts are meant to be run locally inside a user’s Claude Code environment.

5. Several scripts directly support Claude Code session management: context-bar.sh powers the custom status line, check-context.sh monitors token usage and can trigger half-cloning, and clone-conversation.sh / half-clone-conversation.sh copy conversation history into new sessions. This creates a workflow boundary between “active session” and “fresh session with preserved context.”

6. The skills/ directory packages reusable Claude Code behaviors as SKILL files. Each subfolder is a distinct workflow, such as GitHub Actions triage (gha), conversation cloning (clone, half-clone), handoff generation, Hacker News summarization, Reddit fetching, CLAUDE.md review, and version checking. These are invoked by Claude Code as skills or slash commands.

7. The repository also acts as a Claude Code plugin distribution. The .claude-plugin/ manifests define the plugin metadata and marketplace packaging, and the README describes installing the dx plugin from this repo. In architecture terms, the plugin is the distribution boundary that bundles selected skills into a single installable unit.

8. The workflows are designed around external services and tools that Claude Code can call: Git and GitHub CLI (gh) for repository and PR operations, Claude Code’s own slash commands and hooks, and optionally browser automation via MCP/Playwright. Those integrations are not implemented as a web backend; they are invoked from the local agent environment.

9. The repository’s data flow is mostly one-way from documentation to tooling: the README explains the workflows, the skills and scripts implement them, and the plugin manifest exposes them for installation. In use, a user reads the guide, installs the plugin or scripts, and then Claude Code executes local shell commands or skill actions on the user’s machine.

10. content/3d/ is a separate mini-subsystem for interactive visual examples. It includes JavaScript modules and a build.js script for assembling 3D demo artifacts such as bat and cube visuals. This is presentation/demo code, isolated from the Claude Code workflow tooling.

11. There is a clear boundary between “reference material” and “automation.” The markdown articles explain patterns and usage; the scripts and skills encode the repeatable parts. That separation is important for the diagram: the docs are the user-facing knowledge layer, while scripts/ and skills/ are the executable workflow layer.

12. Overall, the repository is best diagrammed as a knowledge hub with attached local automation: README/content at the center, media assets as support, scripts as operational utilities, skills as reusable Claude Code actions, and the plugin manifests as the distribution mechanism.