bulaev.ai

Claude Code

CCCPaid

Agents · Build

Anthropic's agentic coding tool in the terminal, IDE and browser; skills, hooks, subagents and MCP.

Open claude.com

What it is

Claude Code is Anthropic's coding agent. You run it in a terminal inside a repository, describe what you want, and it reads files, edits, runs tests and commands, and iterates until the task is done; it also ships as VS Code and JetBrains extensions, a web version at claude.ai/code, desktop apps and a Slack integration. The site's install line is curl -fsSL https://claude.ai/install.sh | bash, and usage is included in the Pro and Max plans or billed per token through the API.

The mechanics that make it an operating system rather than a chat box: a CLAUDE.md file per project that the agent reads first; skills (SKILL.md folders it loads on demand); hooks that run shell commands on events; subagents that fan out in parallel with their own context; MCP servers for external tools; and routines for scheduled or event-driven runs. Non-code work, from writing a spec to producing a design canvas, is the same loop with different files.

Where it fits in the AI OS

Build, the primary agent. Everything else in this catalog is either a tool Claude Code calls (Playwright, Linear, Publora) or a place its output goes (Cloudflare, GitHub). In the Work OS view it is also the "employee" that gets a Linear issue and a branch.

Insider notes

members

How Serge's stack uses it

The membership project is built with it end to end and its rules are written for it. CLAUDE.md in the repo fixes the conventions: one Claude session equals one Linear issue, one task equals one branch, every decision goes into the decision log the same turn, and nothing merges to main without a dual review by an independent Claude agent and Codex CLI. The ideas pipeline is a skill (idea-clip) that takes a URL and produces a screenshot, a short clip and a mechanism report; the social media bundles are Claude Code skills; even the design canvas and brand assets are generated from it.

Setup in 5 minutes

  1. Install with the one-line script (or npm install -g @anthropic-ai/claude-code), run claude in a repo, sign in.
  2. Run /init to draft a CLAUDE.md; edit it to state build, test and commit conventions.
  3. Add one MCP server you use daily, for example claude mcp add linear.
  4. Create .claude/skills/<name>/SKILL.md for a task you repeat weekly.
  5. Try claude -p "summarize open TODOs" from a cron to see headless mode.

Alternatives

  • Codex CLI — OpenAI's terminal agent; used here as the second reviewer.
  • Cursor — IDE-first agent with strong tab completion.
  • Gemini CLI / Aider / OpenCode — open-source terminal agents that work with several model providers.