create-qa-list

Build a QA test-case list a non-developer QA tester can read and run without opening the code — as CSV, HTML, or both.

The two rules that govern everything

Every decision in this skill follows the two rules below. They pull in different directions (one is presentation, one is accuracy) and both must hold for every row.

Rule 1 — The reader is a non-developer. They run the product and watch what happens. They don't read code. So every case is written in behavior-and-scenario language — what the user does and what they see. Function names, file paths, table/column names, and code symbols never appear in the output. This rule drives the tone, the columns, and the verification steps.

Rule 2 — Every case must match the code. Don't make things up. A QA list that diverges from real behavior is worse than none: the tester tests behavior that doesn't exist, files bugs that aren't real, or marks a real bug as "pass". So every scenario and expected result must rest on something you actually verified (code, spec, this conversation). If you can't verify a behavior, do one of three things: read the code to confirm, ask the user, or drop the case. Don't write a guessed expected result as if it were fact. Flag any case resting on an unverified assumption as needs-confirmation instead of exporting it silently.

Together they are the skill's mission: translate verified code behavior into language a non-developer can act on — translate, but don't invent, and don't leak the code.

Input can be anything

The user might give you: specs and planning docs, a URL, a PR/diff, pasted requirements, a one-line request, or just "make a QA list for what we built." The current conversation is itself valid input — if you've been building or discussing a feature, that context is the input.

The first job is to figure out which feature/behavior is the target. Infer from everything available first, then confirm — don't assume.