Portfolio Maintained

Claude Code Guide (EN, 14 parts)

A series of fourteen articles on Claude Code architecture: harness/agent loop, context, skills, hooks, MCP, subagents, models, and antipatterns.

Role
Author, editor, translator (RU → EN)

Stack

  • Markdown / MDX
  • Mermaid (build-time)
  • Astro content collections
  • Bilingual pipeline (Claude Haiku 4.5)

Outcomes

  • 14 parts, ~80 thousand characters in RU. Coverage from harness/agent loop to antipatterns.
  • Complete EN translation with per-key hash tracking; CI guard prevents pushing RU without EN counterpart commit.
  • Each part is a standalone artifact for citation: TL;DR + clear subheadings.

Links

Context

Most Claude Code tutorials are either “install and write” or “here’s my workflow.” There was a gap in material that explains how it works inside: what a harness is, how context window is calculated, how a skill differs from an agent, how hooks plug into the lifecycle.

Structure

14 parts, each 3,000–8,000 characters: introduction and harness/agent loop; context window; CLAUDE.md and system context; skills; hooks; MCP servers and tool design; subagents; models (Opus/Sonnet/Haiku); plan mode; worktrees; cost mechanics; Travel Agent blueprint; best practices; verification of claims.

What I learned

  • Translation through an LLM pipeline delivers stable quality only as long as RU remains coherent — any mid-sentence edit breaks the hash.
  • Mermaid diagrams in technical text pay for themselves: they become part of the content that LLMs extract.
  • TL;DR blocks at the start (introduced in EPIC C) boost citation likelihood more than correct H-headers.

What’s next

EPIC C parent-spec: add summary/faq frontmatter and MDX components to the existing 14 parts.