← ClaudeAtlas

office-hourslisted

Force the clarifying questions out into the open and wait for answers before writing any code.
rohithkandula19/Ronin · ★ 1 · AI & Automation · score 71
Install: claude install-skill rohithkandula19/Ronin
# office-hours — ask first, build second The failure this prevents: you guess what the user meant, build the wrong thing confidently, and the mistake is only caught after the work is done. Use it whenever a request has more than one reasonable reading. While this skill drives you may read and search to inform your questions, but you may **not** write, edit, or run commands that change anything. 1. **Read the request literally**, then read it adversarially: where could a well-meaning engineer build something the user did not want and still claim they followed instructions? Each such spot is an ambiguity. 2. **Ground the questions in the code.** Use `grep`, `glob`, and `read` to check what already exists before you ask — a question the repository already answers wastes the user's turn. Ask about what the code cannot tell you: intent, scope, priorities, the shape of "done." 3. **Enumerate the ambiguities** as a numbered list. For each, state the decision to be made and the options you can see, with your recommended default in **bold** and one line on why. A question with a default is answerable in five seconds; an open-ended one stalls. 4. **Cover the axes that bite most often:** scope (what is in and explicitly out), inputs and their edge cases, the expected output or interface, error and failure behaviour, performance or scale constraints, and which existing code this must match. 5. **Ask, then stop.** Post the numbered questions and wait. D