I told Claude Code the wrong bug. On purpose.

The exact clicks

1. A "terminal" is a plain window for typed commands — every Mac and Windows computer already has one, nothing extra to download for this part. Open it — Mac: click the magnifying glass icon top-right of the screen (Spotlight), type "Terminal", press Enter; Windows: click the Start menu (bottom-left), type "PowerShell", press Enter.
2. Mac/Linux: paste curl -fsSL https://claude.ai/install.sh | bash into that window and press Enter. Windows: paste irm https://claude.ai/install.ps1 | iex instead. This is the official installer from Anthropic's own site — safe to paste. Lines scroll by while it installs; when they stop and you get a new blank prompt, it's done.
3. Type claude --version and press Enter. If it prints a version number, the install worked — if you see "command not found", close the terminal window and open a fresh one (the first install needs a new window to be recognized).
4. Type claude and press Enter inside any project folder. The first time, a browser tab opens asking to log in with the same email and password used for claude.ai.
5. Already have a Claude Pro or Max subscription? Nothing else to buy — Claude Code is included at no extra cost. On the free plan, it asks to upgrade before it will run any real request.
6. Type a real request in plain English, describing what you think is wrong — for example "there's a bug in this file: it skips the first item in the loop, fix it" — and press Enter. It reads the actual file itself rather than only trusting the description, so if the real bug is somewhere else, it can find and fix that instead.
7. Optional: the same tool also works inside VS Code, Cursor, and JetBrains IDEs as a built-in extension, not just the terminal — same login, same account, no separate setup.

What changes

• Claude Code is included with a normal Claude Pro subscription ($20/month) — confirmed current via Anthropic's own pricing page before this episode was written, not assumed from an older episode.
• It reads the actual file before acting, rather than only trusting a plain-English bug description — a wrong or incomplete description of the symptom doesn't stop it from finding the real defect.
• The demo in this episode was run for real, not scripted: a genuinely broken script, a deliberately false bug description, and the tool's own unedited response as the evidence.

What it will not do

• This one test showed the tool looking past a wrong description on one small, clear-cut bug (an operator-precedence mistake) — it isn't a claim that every wrong description gets caught on every bug, especially subtler or more ambiguous ones.
• Free-plan accounts can't run this at all without upgrading first — the $20/month claim is specifically about Pro-and-up, not every Claude account.
• Needs a real project folder and an actual broken file to point it at — it doesn't diagnose a bug from a description alone with no code to read.