← ClaudeAtlas

rjv-pr-descriptionslisted

Use when writing, generating, updating, or fixing a GitHub PR or GitLab MR description. Splits the body into what a tester can check in the UI alone and what needs code or environment setup; each part is written from the tester's view, with grounded steps, a screenshot or a command's output, and ticks for what was actually run. Enforces hard caps; on update preserves checked checkboxes + author content; posts the result.
rjvim/ai-skills · ★ 0 · AI & Automation · score 61
Install: claude install-skill rjvim/ai-skills
# PR Descriptions Two people read this: a reviewer deciding if the change is right, and a tester deciding what to check. Write for the tester and the reviewer is served too. In MFStack work the tester clicks through the product and does not read code. Treat length as a defect. An invented test step is worse than no test step: it sends a human to check something that does not exist. ## Shape The body always has these two sections, in this order, both present: 1. **Test in the UI**: what a tester can check by clicking through the running product, with no code, console, database or config change. 2. **Needs code or environment setup**: what can only be checked by changing config, an env var or a flag, seeding data, running tinker or a command, calling an API directly, or reading logs or the database. When a section has nothing, it holds one line: `None.` Never drop a heading. The tester must see at a glance which part they can do alone. Each section is a list of items. An item is one test sitting: one actor, one path, from where they start to something they can see or read. - In the UI section an item is a journey: `J1`, `J2`, ... - In the setup section an item is a check: `S1`, `S2`, ... Group by what a person would sit down and test as one sitting. Never by file, never by commit, never by layer. A check's first line says what happens today, its second what happens after. Then the setup, then the command or call and the output that proves it. ## The rule up