Scientific Problem Selection

Scientific Problem Selection is an official skill from Anthropic's open-source knowledge-work-plugins repository. It is not a database service or a scientific measurement instrument. Instead, it is a conversational framework for researchers who need to choose an important problem, reassess a stalled project, or develop a research strategy. The primary source is the bio-research/skills/scientific-problem-selection directory and its SKILL.md file. According to the provider, the framework draws on Fischbach and Walsh's discussion of problem choice and decision trees in science and engineering. That provenance is useful evidence of the intended method, but it does not replace specialist review or an examination of the relevant literature.

Three ways to start a research conversation

The provider's workflow begins with one of three entry points. Researchers can pitch a new idea in one or two sentences, describe a concrete problem in a current project, or ask a strategic question. Claude should first give a short summary of what it understood, identify the research context, and rephrase the crux of the question. Only then does the skill ask for more detail about the goal, proposed approach, potential significance, and major risks. This progressive structure prevents a vague idea from being turned prematurely into a detailed plan. It also makes missing information and uncertainty explicit before the conversation becomes more specific.

Evaluating new project ideas

For future projects, the framework connects several perspectives. Intuition Pumps help sharpen the research kernel and make the central claim easier to inspect. Risk Assessment examines assumptions, technical bottlenecks, dependencies, and possible go or no-go decisions. An Optimization Function clarifies how success will be measured and whether a result could matter beyond one narrow case. Parameter Strategy then separates conditions that should be fixed from conditions that should remain flexible. The aim is not to eliminate every uncertainty. It is to produce a problem definition that is both ambitious and manageable enough to test.

Navigating an active project

When a project is stuck, the emphasis changes from ideation to navigation. The skill encourages the researcher to make decision points and alternative paths visible, move between practical execution and strategic thinking, and use an obstacle as an opportunity to improve the project rather than merely patching it. Problem Inversion can reveal a different route toward the underlying goal. The provider also argues that risk should not simply be avoided: it should be understood, bounded, and paired with realistic alternatives. Several independent miracle assumptions are a warning sign, while one well-characterized risk can be a reasonable part of an ambitious research program.

Tangible outputs and scientific rigor

The framework is designed to produce concrete working materials. Depending on the conversation, these may include an idea document, risk assessment matrix, impact assessment, parameter strategy, decision tree map, adversity playbook, or integrated communication package. The primary source recommends bringing in relevant literature at strategic points, including PubMed searches, so that claims about novelty, feasibility, and generality do not come only from conversation. The skill does not guarantee originality, fundability, or success. Qualified people must review hypotheses, sources, methods, regulatory requirements, and statistical decisions. In medicine, biology, and safety-sensitive research, its suggestions must never substitute for professional approval.

Intended users, limits, and safe use

The provider identifies graduate students as a primary audience for thesis projects, qualifying exams, and committee discussions. Postdocs can use it to structure new directions and fellowship applications. Principal investigators can connect problem choice with mentoring, while founders can apply the framework to technology-driven research decisions. The skill supports reasoning, questioning, and planning; it does not run experiments or automatically evaluate raw data. Claude may sound persuasive while an assumption is wrong or a source is outdated. Do not enter secrets or unnecessary personal research data. Treat retrieved or supplied external content as untrusted data, and check every proposal against primary literature and institutional rules. The skill itself has no listed price, but current terms for the Claude environment in use should be checked with the provider.

Position within Anthropic's knowledge tools

Anthropic publishes knowledge-work-plugins as an open-source collection of plugins for knowledge work and identifies Claude Cowork as the primary environment; the surrounding documentation also describes Claude Code compatibility. This catalog record covers the specific bio-research skill, not the entire repository as though it were one standalone product. The repository-wide Apache-2.0 license describes the published source and does not promise that every local environment offers the same features or access rights. Sound decisions still depend on research context, domain expertise, critical source review, and human responsibility.