self-evolving/literature-template/literature-review

Conduct a systematic review on the user's requested topic. Use the repo-local setup script if paper or paper-search is missing:

.skills/literature-review/setup.sh

The setup installs the pinned agent-papers-cli version and runs non-secret smoke checks. Search and browse commands may need SERPER_API_KEY, S2_API_KEY, or JINA_API_KEY in the shell environment.

Scope

Define the review scope before searching:

• Topic boundaries and excluded areas
• Default year range, normally the last five years unless the topic needs seminal work
• Target communities, venues, organizations, repositories, datasets, or benchmarks
• Desired number of core research items
• Source mix: papers, blog posts, GitHub repositories, docs/releases, datasets, benchmarks, and optional tweets/threads

If the user did not provide scope constraints, make reasonable assumptions and state them briefly in the final report.

Search Plan

Use several query variants to improve coverage. Track queries and deduplicate by title, DOI/arXiv ID, URL, or repository owner/name.

paper-search semanticscholar papers "<main query>" --limit 20 --year <range>
paper-search semanticscholar papers "<synonym query>" --limit 20 --year <range>
paper-search google scholar "<topic>"
paper-search google web "<topic> blog OR release OR benchmark OR dataset OR github"
gh search repos "<topic>" --limit 20

Use PubMed for biomedical topics:

paper-search pubmed "<query>" --limit 20

Triage

For each candidate research item, capture:

• Type: paper, blogpost, github_repo, docs_release, dataset, benchmark, or tweet_thread
• Canonical URL and source
• Date or year
• Relevance: high, medium, low
• Reason for inclusion or exclusion
• Relationship to other items, such as code for a paper or benchmark used by several papers

Prefer primary sources. Treat social posts as discovery hints unless corroborated by papers, repos, docs, datasets, or benchmark pages.

Deep Review

For high-relevance papers:

paper outline <arxiv_id_or_pdf>
paper skim <arxiv_id_or_pdf> --lines 2
paper read <arxiv_id_or_pdf> introduction
paper read <arxiv_id_or_pdf> method
paper read <arxiv_id_or_pdf> results
paper read <arxiv_id_or_pdf> conclusion
paper-search semanticscholar citations <paper_id> --limit 20
paper-search semanticscholar references <paper_id> --limit 20

For non-paper primary sources:

paper-search browse <url>
gh repo view <owner/repo> --json nameWithOwner,description,url,stargazerCount,pushedAt,defaultBranchRef
gh release list --repo <owner/repo> --limit 5

Record problem, method or artifact, evidence, limitations, adoption or impact signals, and open questions.

Report

Organize the final review by theme, not by individual item. Include:

• Scope and search assumptions
• Research-item table with type-aware metadata
• Key themes and points of agreement
• Contradictions or weak evidence
• Gaps and future directions
• Reference list with URLs and paper IDs
• BibTeX entries for cited papers when useful

Use this table shape for the candidate set or final shortlist: