← ClaudeAtlas

research-brieflisted

Turn a research question into a written brief before any searching. Use when a research task starts, when the question is broad or ambiguous, or before you open the first source.
volekerb/hivemill · ★ 33 · AI & Automation · score 73
Install: claude install-skill volekerb/hivemill
Purpose: fix scope, method and exit criteria first, so the research stays short and answers the question that was asked. Follow these steps in order. 1. Read the task: `factory task show <own task> --full`. Restate the question in your own words in one line. 2. If the question is ambiguous (two plausible readings, no decision it feeds, no stated audience), ask before researching: `factory msg lead:<team> "research question unclear" "<the two readings and your default>" --kind request`. Wait for the answer. 3. Split the question into numbered sub-questions (Q1, Q2, ...). Each must be answerable with evidence. 4. Write scope and non-goals: what you will cover and what you will not touch. 5. Write the source plan: which kinds of sources per sub-question (official docs, source repos, papers, our own repo), and in what order. 6. Pick one method: desk research, comparison (skill tool-comparison), benchmark (skill benchmark-protocol) or plugin evaluation (skill plugin-evaluation). 7. Write one falsifiable hypothesis per sub-question (H1 for Q1, ...): the answer you expect and the observation that would prove it wrong. Write exit criteria: the evidence that confirms or refutes each hypothesis, and what counts as "not found". 8. Write a time box (for example 45 minutes, at most 12 sources). When it is spent, stop searching and say so in Limits. 9. Record the brief first: `factory task note <own task> -` with the full brief as the heredoc body. 10. Research against the brief. Rate ev