Vibe Code Textbook

Agentic coding, written down properly.

The tools that write code with you are changing monthly. This site keeps up by reading the documentation and the repos at a named commit, running its own scripts against public records such as release timestamps and benchmark results, timing Claude Code itself on ten small tasks, and writing down what was checked and what was not.

28 chapters in five parts. The newest was added on .

Six chapters where we measured something or ran a script ourselves: benchmark records, release timestamps, cost rules, instruction loading, transcript redaction and tool definitions.

Recently added

All 28 chapters by date

How it is made

Published under the owner’s direction. Every chapter says what was read, which of its scripts ran, and what did not run. The one chapter that ran an agent, time-to-green, records 60 headless Claude Code attempts from 30 October 2026 and publishes their redacted transcripts. Chapters with code link the complete listing and, where there is one, the dataset. How this site is made.

Contents

Part I

Harnesses

The agent runners themselves: what each one does, its configuration, its permission model, and where it breaks.

  1. 1What is a coding-agent harness? The loop, tools, gate, and context6 min
  2. 2Claude Code vs Codex vs Gemini CLI: what an agent may run7 min
  3. 3CLAUDE.md, AGENTS.md, and .cursor/rules: what goes in a context file6 min
  4. 4MCP server tutorial in Python: a stdio server from scratch6 min
  5. 5Claude Code and Codex headless: running a coding agent in CI safely7 min
  6. 6Coding agent security: MCP prompt injection, servers, and secrets21 min
  7. 7Coding agent memory across sessions: Claude Code, Codex, Gemini CLI12 min

Part II

Workflows

Repeatable ways of working with an agent: specs, tests-first, review loops, context files, and the habits that keep a session on track.

  1. 8Spec first: how do you write a task a coding agent will finish?5 min
  2. 9Tests first with an agent: why must the test fail before it passes?6 min
  3. 10Reviewing coding-agent diffs: what does the checklist catch first?10 min
  4. 11Git worktrees for coding agents: one checkout per session, why?6 min
  5. 12When should you stop a coding agent? Budgets, loops, and the signs6 min
  6. 13Prompt patterns for coding agents: plan mode, skills, saved prompts12 min
  7. 14Monorepo strategies for coding agents: what loads and what runs7 min
  8. 15AI code review bots compared: triggers, rules files, blocking, cost6 min
  9. 16Agent transcripts as documentation: export, redact, link the commit9 min

Part III

Repos Worth Reading

Open-source projects in the agentic-coding space, read at a specific commit, with what they get right.

  1. 17SWE-agent and the agent-computer interface idea, paper and repo7 min
  2. 18Aider explained: repo maps, edit formats, and why it still matters6 min
  3. 19OpenHands architecture explained, read at commit f7fb0c4b8 min
  4. 20ccusage explained: how an agent cost dashboard prices your logs8 min

Part IV

Tutorials

Build something end to end with an agent: the prompts to give it, the tests that decide the result, and a script you can run.

  1. 21A one-paragraph spec for a coding agent: building a small CSV CLI5 min
  2. 22A refactor plan for a coding agent: pin the behaviour with tests first7 min

Part V

Measurements

Numbers instead of vibes, from public data and our own runs, each with its dataset and how it was collected.

  1. 23SWE-bench explained: what a resolve rate measures and what it misses11 min
  2. 24What one coding task costs: tokens, dollars, minutes, three ways6 min
  3. 25Coding agent time-to-green: ten small tasks, 60 Claude Code runs15 min
  4. 26Flaky tests: what a green rerun tells an agent7 min
  5. 27How often Claude Code, Codex and Gemini CLI update, measured11 min
  6. 28SWE-bench Verified task by task: 11 models, one harness, measured12 min