Literature Review Skills

A practical five-skill workflow for planning, structuring, and sharpening literature reviews. It helps you refine a question, break the review into modules, compare methods, handle contradictions, and locate a defensible research gap. It does not write the review for you; you still make the academic decisions, verify the literature, and write the final argument.

中文说明

Who This Is For

Graduate students preparing a thesis proposal, course paper, or dissertation literature review.
Researchers who need to turn scattered readings into a clear review structure.
Supervisors who want a repeatable workflow for guiding students through review design.
Claude Code and Codex users who want reusable skills for academic thinking tasks.

The Five-Skill Workflow

Step	Skill	Use it to
1	`refine-literature-question`	Turn a broad topic into a focused review question that guides search and screening.
2	`decompose-review-modules`	Break the question into 3-5 review modules with clear boundaries.
3	`compare-research-methods`	Compare method routes and explain what each method can and cannot answer.
4	`map-literature-contradictions`	Turn conflicting findings into structured synthesis instead of vague disagreement.
5	`locate-review-research-gap`	Convert the synthesis into a concrete, defensible research gap or future direction.

Install This Skill Set

You do not need to copy commands for each individual skill. This repository includes a root SKILL.md as the entry point, plus the five working skills under skills/.

Method 1: One-Line Install (Recommended)

If your environment supports the Agent Skills installer, run:

npx skills add https://github.com/xingtaxueshu/literature-review-skills

This is the cleanest option when you want the installer to place the skill set in the right local skills directory.

Method 2: Ask Your Agent To Install It

Ask Claude Code, Codex, Cursor, or any compatible coding agent:

Please find and install this skill set: https://github.com/xingtaxueshu/literature-review-skills

The agent should clone the repository, place it in the appropriate skills directory, and make the root SKILL.md available as the entry point.

Method 3: Manual Clone

For Claude Code users:

git clone https://github.com/xingtaxueshu/literature-review-skills.git ~/.claude/skills/literature-review-skills

For Codex users:

git clone https://github.com/xingtaxueshu/literature-review-skills.git ~/.agents/skills/literature-review-skills

After installation, ask your agent to use the skill set:

Use literature-review-skills to help me turn my broad topic into a literature review question, modules, method comparison, contradiction map, and research gap.

You can also call an individual skill directly when you already know the stage you are working on:

Use refine-literature-question on this topic: rural e-commerce participation.

Suggested Workflow

Sample prompt:

I am writing a literature review on how rural e-commerce affects household income and local development. Help me turn this broad topic into a review question, modules, method comparison, contradiction map, and research gap.

Recommended sequence:

Use refine-literature-question to convert the topic into a focused question, such as how rural e-commerce changes household income through market access, labor allocation, or platform participation.
Use decompose-review-modules to create review sections, such as adoption conditions, income mechanisms, platform roles, regional differences, and policy environment.
Use compare-research-methods to understand what surveys, panel data, interviews, case studies, and mixed methods can and cannot answer.
Use map-literature-contradictions to compare studies that report positive, weak, uneven, or negative effects and identify why results differ.
Use locate-review-research-gap to state what the literature explains, where it gets stuck, and what a next study could examine.

Responsible Use

These skills are a thinking workflow, not a ghostwriter. Use them to clarify questions, organize evidence, test logic, and improve research design. Do not use them to fabricate sources, outsource academic judgment, or submit generated prose without checking it against real literature and your own argument.

Since v0.2.0 the skills follow references/source-integrity.md: only sources you provided or that have been verified are used, each is tagged (用户提供 / 已核验 / 未核验), and anything not found is written as "未查到" rather than filled in. Every output ends with a to-verify list and questions for the author, and generated paragraphs are labelled as drafts for you to rewrite and check (output contract). These are process rules for the agent, not a guarantee; check every citation yourself.

Limitations And Unverified Items

This is an early version. Please read the following before relying on it.

Not evaluated. There is no test set, no comparison against a baseline prompt, and no measurement of whether the skills improve a review. Any benefit described here is the design intent, not a result.
Usage feedback has not been collected systematically. These skills were already used by a good number of outside authors before v0.2.0 (mostly reached through our WeChat Official Account posts; the repo has 72 stars). That use was voluntary: we have not systematically collected feedback and have not run a controlled evaluation. The v0.2.0 changes have not yet been validated on real manuscripts.
The example has no real sources. examples/ is a method demonstration; its claims are illustrative, not drawn from specific papers, and must not be cited.
Install methods are not individually tested. We have not verified the npx skills add command, agent-assisted install, or manual clone in each environment (Claude Code, Codex, Cursor). The agents/openai.yaml files are also untested. So that a skill still works if an installer copies only one skill folder, each skill folder bundles its own copy of references/, and the core rules are also repeated in each skill's Guardrails.
Source verification depends on the agent's tools. Without network or search access, the agent cannot verify anything and will leave sources as 用户提供 or 未核验. Crossref and OpenAlex coverage of Chinese-language and book literature is incomplete.
No Chinese-literature workflow. CNKI, Wanfang, and VIP checks are listed as manual steps for you to do; there is no dedicated process for CSSCI-style or other Chinese-language search, screening, or citation formats.
Method templates lean empirical. The method and question templates are written mostly for survey, regression, interview, and case designs. Argument-driven reviews (conceptual history, textual interpretation, intellectual history) have no dedicated templates yet.
The v0.2.0 output contract is untested. Models may still skip sections or mislabel sources; spot-check the output.

Contact

Questions, examples, or collaboration ideas are welcome.

WeChat: xingtacc
Email: [email protected]

Scan the QR code to add the WeChat account:

Follow the WeChat Official Account:

WeChat Official Account QR code for 星塔学术

Contributing

Contributions are welcome if they make the workflow clearer, more practical, or easier to use. Good contributions include sharper skill instructions, better examples, field-specific templates, and corrections to installation notes. Keep examples realistic and avoid inflated claims.

skills/*/references/ are copies of the root references/, which is the authoritative version. When you change a file in the root references/, copy it into every skill folder and run bash scripts/check_references_sync.sh (it exits non-zero if any copy differs).

License

This project is released under the MIT License. See LICENSE for the full text.