yzbbds520/awesome-literature-review

A Claude Code skill is used to generate a well-structured literature review in four parts: Research Background and Objectives → Literature Review and Analysis → Current Research Status Summary and Review → References.

The design focuses all manual intervention before writing—first aligning titles, sources, output formats, and citation standards, then confirming the search criteria before searching literature, verifying the literature list by users, and then writing the entire first draft in one go without stopping to check each piece.

Acknowledgements

ClawsGO Science is a research AI Agent platform. The agent completes literature retrieval, code execution, data analysis, and drafting on a standalone cloud host, with no manual supervision required for the entire research cycle; Environment, commands, file changes, and literature retrieval are all recorded at the location where actions occurred and can be replayed and checked at any time; Web, mobile, and desktop share the same task status.

---

Thanks to F-API Hub for the funding support of the “Student Incentive Program.” F-API Hub is a global AI model aggregation API platform that integrates mainstream AI models into a single API, making it easy for developers to switch and compare calls as needed, saving the hassle of applying for and maintaining multiple sets of keys from each provider. If you need AI calls while learning, coding, or working on course projects, you can click the link above to learn more. As a student, you can also apply for exclusive funding slots.

Features

• Documents Must Be Authentic: Every article included in the main text must have verifiable sources. Users should paste them as they are; If AI is searching online, you must provide a source link and verify with users first, strictly preventing the creation of titles, authors, and years based on memory.
• No Conclusions Filled in Blanks: In the main text, what a certain document does or derives can only be based on the actual original text (user-provided abstract or WebFetch’s abstract page). If you only have the title and year, only the topic is written; it is forbidden to add method names, datasets, accuracy, or percentages not found in the original text. Each document should be labeled according to its level of evidence (Grade A includes abstract/full text, Grade B only has a bibliographical record), and how detailed you can write is determined by the level when writing.
• Determine the search criteria before searching: When using AI-connected search paths, before starting, use AskUserQuestion to clarify the document size, year span, source threshold (core journals/general journals/preprints), and Chinese-English ratio, without setting default values for users. Also keep search records (search terms, sites, dates) and write them into the “Research Background and Purpose” section.
• Preliminary alignment, one-time draft: Opening with four consecutive questions (title, source of reference, output format, citation standards). After searching the literature list and confirming, write the complete paper in four parts before submitting.
• Ready to use your existing materials: If you have a proposal report, draft draft, assignment book, or exported reference list, just submit it directly—after reading, first pull out the topic, direction, and reference list, only ask the missing items, no need to go through the Q&A again. Multiple materials will also be cross-checked and contradictions listed separately for your decision.
• In-depth commentary: The third part is not a repetition of the previous summary. It should state the divergent conclusions of different studies on the same issue (and where the differences come from), methodological common limitations, research gaps in specific subjects and methods, why these gaps still exist, and future directions corresponding to each gap. The current research status of the thesis should be connected to the author’s own research questions in the final paragraph.