Turn a rough research direction or literature gap into a structured, falsifiable research question ready for a proposal.
Get an AI mentor or grant-panel critique of your early research idea before investing weeks of work.
Check whether your question has rival hypotheses and a concrete two-week pilot you could actually start.
Use in Chinese or English to sharpen questions in fields like ecology, biomedicine, or AI for science.
Install as a skill in Claude Code or OpenAI Codex CLI. No external API keys or databases required. Works in English and Chinese out of the box.
good-question is a skill for AI coding agents (tested with Claude Code and OpenAI Codex CLI) that helps researchers sharpen a vague idea into a well-formed research question. It is most useful when you have a rough direction, a literature gap, or an early proposal sketch but are not sure whether the question is actually worth pursuing. The skill is bilingual, with full support for Chinese and English. The tool starts by diagnosing where you are in the research process: do you have only a broad interest, a spotted gap, an early idea, a proposal draft, or a stalled project? It then operates in one of four modes (mentor, reviewer, collaborator, or grant-panel) depending on the context. If the question depends on current knowledge in a field, the skill first builds a brief from public-source information and labels claims as either evidence-backed or inferred. For specialized fields like ecology, remote sensing, AI for science, social science, and biomedicine, it loads a lightweight domain adapter with relevant norms and common pitfalls. The main output is a structured card called a Good Question Card. It contains a working title, the research question itself, why it matters, which default assumption it challenges, at least two competing hypotheses, a discriminating observation or experiment that could tell the hypotheses apart, what result would falsify the question, a concrete two-week pilot someone could actually start, the strongest reviewer objection, and the best next action. The point of the card is not to produce polished wording but to give the researcher enough structure to decide whether the question deserves real investment. For a question to pass the project's internal standard, it must satisfy seven checks: it matters to someone, it is specific enough for evidence to touch it, it has rival explanations, it can be falsified, a pilot is feasible, a negative result still teaches something useful, and any field-context claims trace back to public sources. The project is licensed under MIT and includes field guides, evaluation cases, and a release checklist.
← pillarpaheat on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.