Corrected edition of October 22, 2026
The free self-assessment tool is published on npm as vibeqm-check: npm did not accept the name earlier copies use, vqm-check (the introduction, chapters 3, 4, 18 and 19, and About the Author). Run it from the root of your repository with npx vibeqm-check. Once installed, it also answers to vqm-check.
Chapter 4. Where You Stand: The Honest First Assessment
- "With the tool". It said: "The companion repository, linked from vibeqm.com, has the current install command". It now gives the command: "Run it from the root of your repository with
npx vibeqm-check(it needs Node.js and git)".
Corrected edition of October 21, 2026
This edition corrects the two agent prompts below. If your copy is earlier (bought before October 21, 2026), use the new wording when you paste them.
The same edition also rewrites chapter 2 and parts of the introduction as an account of how software quality was handled before AI, and renames three headings ("Quality Before AI, and What AI Breaks"; "How quality was handled before AI"; "Why this model"). These are not errors, so they are not listed one by one.
Chapter 1. Vibe Coding Works Until It Doesn't
- "What to do on Monday", the prompt. It said: "List every hit with its file and line number, never the value of a credential". It now says: "Never open, print or quote a credential or an environment variable that may hold one: report only the file name and line." The old wording still let the agent open a credential.
Chapter 14. Agent Permissions and Safety
- "What to do on Monday", the self-audit prompt, step (2). It asked the agent to try a force-push and skipping the commit checks, and said only "Don't work around a refusal." It now also says: "If the system does not refuse one, don't repeat it: undo what it did, delete the scratch branch and report it as a gap." The old prompt did not say what to do when an attempt went through.
Corrected edition of October 19, 2026
This edition corrects the errors below. If your copy is earlier (bought before October 19, 2026), here is what changed. Page numbers differ between editions, so each item gives the chapter and section.
The same edition also calls the model by its short name VibeQM (earlier copies say "VQM"). It numbers the chapters and groups them into four parts. These are not errors, so they are not listed one by one.
Chapter 1. Vibe Coding Works Until It Doesn't
- "What the evidence says", opening. It said: "All of them come from 2024 and 2025". It now says: "Most of them come from 2024 and 2025". Not every study cited in the section is from those two years.
Chapter 7. Level 3: Standardized
- "4. Peer review by risk". It said: "Two practical points:". It now says: "Four practical points:". Four points follow.
- "The Level 3 gate", after the table. It said: "The twelve items above fill the first thirteen rows (item 6 has two)". It now says: "The twelve items above fill the first fourteen rows (items 4 and 6 have two each)". Item 4 also takes two rows.
Chapter 8. Level 4: Measured
- "DORA's four delivery metrics". It said: "AI tends to push deployment frequency and lead time up." It now says: "AI tends to raise deployment frequency and shorten lead time." Faster delivery shortens lead time; the old wording said the opposite.
Chapter 18. Assessment by AI
- "What you have now". It said: "An assessor's ten rules". It now says: "An assessor's twelve rules". The chapter gives twelve rules.