literature-review
Skill59 starsChanged 5 months agoApache-2.0View on GitHub
Builds a structured literature review — survey of prior work, state of the art, related-work landscape — where every claim cites a verified reference. Use when the user says "literature review", "lit review", "survey the papers on X", "what's the state of the art", "synthesize prior work", or "organize these papers by theme". Composes sibling skills, search via find-papers (key-free DBLP/Crossref/Semantic Scholar/arXiv), full text via fetch-paper (legal open-access only), and citation checking via verify-citations. Tracks a screening corpus, extracts claims with section-level anchors, organizes by theme, and gates the final document with a deterministic checker so no claim ships uncited, unverified, or resting on a fabricated reference.
- Reads credentials
92 other projects use a version of this.
Adopted by
Most recently taken up by
- BraveDistribution/claude-skillstheir own version, first seen Nov 16, 2026
- KPISTOLAS/claude-skillstheir own version, first seen Nov 16, 2026
- Ivan-propio/claude-skillstheir own version, first seen Nov 16, 2026
- tatashbar/Economic-Modeling-Studiotheir own version, first seen Nov 16, 2026
- lengrufeng2008/dsh-skill-librarytheir own version, first seen Nov 15, 2026
- PerryLink/dsh-research-skillstheir own version, first seen Nov 15, 2026
Used with
Projects with it also use pdf-explore, paper-narrative and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
JimLiuSkillpdf-explore
Use this skill when the user has attached a PDF, paper, report, or other document and the answer needs content from more than one place in it: summarize the methods or any other section, compare sections, find where a topic is discussed, read a value or label off a figure or chart, or find/list/extract every instance of something across the whole document (datasets, benchmarks, citations, figures, table rows, accession numbers — including appendices). Skip it only for a single lookup of 1–4 pages quoted in your very next response — `read_file(pages=[...])` attaches pages as images that are dropped from context after one turn, so multi-section answers end up re-reading the same ranges repeatedly. Parses the PDF once in the Python kernel: `pdf_pages` (pages as persistent text), `pdf_outline` (TOC), `pdf_scan` (rank pages by relevance), `pdf_map`/`pdf_extract` (per-page summary / structured fields via parallel haiku calls). For PDF creation/manipulation, use reportlab/pypdf directly.
Also in 9 projects that use this
JimLiuSkillpaper-narrative
Judge and reshape the STORY a paper's figures tell. Input is the work itself — manuscript (or abstract) + figure deck — no hand-written brief. `derive_paper_brief(abstract, captions)` extracts pitch/vision/per-figure-claims; a handling-editor reviewer on the full deck returns hook_verdict (would Fig 1 make me send this for review?), arc (hook→mechanism→evidence→application), figure_moves (panels in the wrong figure), missing_panels (concrete analyses to RUN), kill_list, and boldest_defensible_fig1. Hands per-figure claims to `figure-composer`. Load when writing or revising a paper.
Also in 5 projects that use this
Norman-burySkillbrainstorming-research
Use before writing any paper - clarifies paper type, discipline, topic, methodology, and chapter structure through natural dialogue
Also in 4 projects that use this
lingzhi227Skilllatex-formatting
Handle LaTeX formatting, templates, and styling for academic papers. Set up conference templates (ICML, ICLR, NeurIPS, AAAI, ACL), fix formatting issues, manage packages, and ensure venue-specific compliance. Use when the user needs to set up a paper template, fix LaTeX formatting, or prepare for submission.
Also in 4 projects that use this
lingzhi227Skillnovelty-assessment
Assess research idea novelty through systematic literature search. Multi-round search-evaluate loops with harsh critic persona. Binary novel/not-novel decision with justification. Use before committing to a research direction.
Also in 4 projects that use this
- +6 more from JimLiu
- +1 more from lingzhi227
Other versions92 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 92
- advaitpaliwal/feynman★ 10ksame name
- companion-inc/feynman★ 10ksame name
- getcompanion-ai/feynman★ 10ksame name
- Norman-bury/research-writing-skill★ 3.4ksame name
- Norman-bury/articlewriting-skill★ 3.4ksame name
- HughYau/AcademicForge★ 2.6ksame name
- K-Dense-AI/claude-scientific-writer★ 2.6ksame name
- xuzhougeng/wisp-science★ 1.1ksame name
- wu-yc/LabClaw★ 1.1ksame name
- LeonChaoX/qinyan-academic-skills★ 910same name
82 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository43 other files it gives its agents.
- docs/llms-full.txtllms.txt
- docs/llms.txtllms.txt
- add-venue-profileSkill
- anonymize-paperSkill
- assess-paperSkill
- benchmark-paperSkill
- check-originalitySkill
- draft-related-workSkill
- draft-surveySkill
- fetch-paperSkill
- find-papersSkill
- fit-page-limitSkill
- make-posterSkill
- make-slidesSkill
- match-styleSkill
- orchestrate-paperSkill
- paper-profileSkill
- parse-cfpSkill
- plan-submissionSkill
- polish-proseSkill
- polish-tables-figuresSkill
- preflight-checkSkill
- prepare-artifactsSkill
- prepare-camera-readySkill
- refactor-research-codeSkill
- refactor-structureSkill
- reflect-and-improveSkill
- reflect-paperSkill
- rehearse-qaSkill
- render-workspace-htmlSkill
- select-venueSkill
- simulate-reviewersSkill
- study-exemplarsSkill
- tailor-to-venueSkill
- test-research-codeSkill
- triage-reviewsSkill
- verify-citationsSkill
- verify-claimsSkill
- verify-resultsSkill
- work-with-overleafSkill
- write-abstractSkill
- write-rebuttalSkill
- write-talk-scriptSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
14 rules. 0 are common, 4 are distinctive, and 10 are very specific to this project.
0
Common
Most of the 72 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
10
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 36
Never reimplement these inline (no ad-hoc curl against scholarly APIs).
- Very specific: line 50
Write the agreed criteria into corpus.json under criteria.
- Distinctive: line 54
Follow references/methodology.md for query design, venue enumeration, and snowballing.
- Very specific: line 56
If the scope names venues, read venues/conferences/<id>.yml for the aliases block (DBLP key, S2 venue string, Crossref container title) — …
- Distinctive: line 98
Read the OA copy transiently — prefer arXiv HTML, else the PDF directly.
- Very specific: line 124
Run the verify-citations skill on references.bib, always with --json so there is a machine artifact to reconcile against: check_bibtex.py …
- Very specific: line 135
If you set a flag manually instead, corpus.py set KEY --verified yes now requires --source <provider> (the index whose canonical record …
- Distinctive: line 146
Never upgrade a PARTIAL-PASS to "verified", and never collapse "PARTIAL-PASS, 11 WARN, 1 skipped" into "verified, 0 errors".
- Very specific: line 151
Write review.md following the structure templates in references/review-structure.md.
- Very specific: line 153
Every factual claim about prior work cites a corpus key: [@key] or \cite{key}.
- Very specific: line 163
Run python3 scripts/forward_refs.py <ws>/review.md [plan.md] --corpus <ws>/corpus.json.
- Very specific: line 180
Run python3 scripts/check_review.py <ws>/review.md --bib <ws>/references.bib.
- Very specific: line 199
Never fabricate or embellish a citation.
- Distinctive: line 208
Never submit or post anything on the user's behalf.
“Common” is measured: kept by at least 6 in 10 of the 72 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Literature Review
Produce a themed, citation-grounded review document from a research question. This skill is the orchestrator: searching, fetching, and citation verification are delegated to sibling skills; this skill owns the workspace, the screening corpus, claim extraction, thematic synthesis, and the final coverage gate.
When to use
- "Do a literature review on X" / "survey recent work on X"
- "What's the state of the art in X since 2023?"
- "Read these N papers and synthesize them by theme"
- A standalone survey is wanted. For a paper's Related Work section,
do the corpus-building phases here, then hand off to
draft-related-work.
Inputs
- A research question or topic, ideally with year range and (optionally) target venues.
CONTACT_EMAILexported for the sibling skills' API politeness contract.- Optional: an existing list of papers/DOIs the user already has.
Sibling skills this skill delegates to
| Stage | Delegate to | What it provides |
| Search | find-papers | dblp_search.py, crossref_search.py, s2_search.py, arxiv_search.py (in that skill's script dir) — key-free, rate-limited, cached |
| Full text | fetch-paper | resolve_oa.py — one DOI/arXiv ID → legal OA copy, transient |
| Reference check | verify-citations | validates every BibTeX entry against Crossref/DBLP/S2, flags retractions |
Never reimplement these inline (no ad-hoc curl against scholarly APIs);
the sibling scripts carry the rate-limit, backoff, caching, and User-Agent
contract.
Process
Phase 1 — Scope
- Pin down with the user: research question, year range, inclusion and exclusion criteria (2–4 each), target size (10–15 papers is a solid default; >30 needs explicit user buy-in).
- Run
python3 scripts/init_review.py "TOPIC"— createslit-review/<slug>/withcorpus.json,themes.yml,notes/, and areview.mdskeleton. - Write the agreed criteria into
corpus.jsonundercriteria.
Phase 2 — Search (delegate to find-papers)
- Follow references/methodology.md for query design, venue enumeration, and snowballing.
- If the scope names venues, read
venues/conferences/<id>.ymlfor thealiasesblock (DBLP key, S2 venue string, Crossref container title) — this solves the venue-alias problem. Profiles are a starting point: if any profile fact becomes load-bearing for the review (e.g. which tracks exist), re-verify it against the livecfp_urlbefore relying on it. - Record every search hit batch:
python3 scripts/corpus.py --corpus <ws>/corpus.json import hits.json --source dblp --query "..."(acceptsfind-papers --jsonoutput; dedupes on DOI). Add single papers withcorpus.py add. - Snowball from multiple, topically-diverse seeds — one per theme, not a single convenient paper (which drifts the harvested set into one sub-area and misses the rest). Admit each neighbor only if you can name its role in the question's argument (method-we-extend / baseline / eval-task / foundational-lineage), and run the co-citation sanity check before screening. Full protocol in references/methodology.md.
- Treat a degraded search as not-yet-done, not done.
find-papers'resolve_papers.pystamps each fan-out COMPLETE or PARTIAL and flags any result confirmed by only a single index. If a theme's search comes back PARTIAL (a provider was rate-limited), or its core papers are single-index only, the citation-graph stage silently fell back — recall for that theme rests on luck. Re-issue the rate-limited leg after a cool-down (addS2_API_KEY) or substitute a provider before declaring the theme covered; record incorpus.jsonwhich papers were recoverable only via a fallback path so the gap is auditable.
Phase 3 — Screen
- For each pending paper, decide included/excluded against the written
criteria using title + venue + (fetched-on-demand) abstract. Use
s2_search.pyfrom find-papers for abstracts — never paste abstracts into any committed file. - Record every decision with a reason:
python3 scripts/corpus.py --corpus ... set KEY --screened excluded --reason "out of scope: no spatial component". - Show the user
corpus.py statsand the included list before reading full texts. Confirm the set.
Phase 4 — Fetch and extract (delegate to fetch-paper)
- One paper at a time, run fetch-paper's resolver (from the repo root):
FP=skills/fetch-paper/scripts; python3 "$FP/resolve_oa.py" <DOI-or-arXiv-ID>. Read the OA copy transiently — prefer arXiv HTML, else the PDF directly. If no legal OA copy exists (exit code 3), mark the paper and synthesize from verified metadata only, or ask the user for their library copy. Never use shadow libraries. - Take grounded notes into
notes/<key>.mdusing the claim-record format in references/claim-extraction.md: each claim gets an anchor (section/page), a strength label, and paraphrase only (verbatim quotes < 25 words, always quoted). Never store abstracts or paper text — notes are your words. - Mark progress:
corpus.py set KEY --fetched yes --extracted yes.
Phase 5 — Organize by theme
- Cluster the extracted claims into 3–6 themes; record them in
themes.ymland tag papers:corpus.py set KEY --theme <slug>. - Build the synthesis matrix (paper × theme × approach × evaluation × result) per references/review-structure.md.
Phase 6 — Verify citations (delegate to verify-citations)
- Generate skeleton BibTeX:
python3 scripts/corpus.py --corpus ... bibtex > <ws>/references.bib.corpus.pyis the single source of truth for cite keys —bibtexemits the.bibwith the corpus's exact keys. Never hand-edit a key in the.bib; change it in the corpus (corpus.py add --update) and regenerate, so the corpus, the.bib, and the review's[@key]s never diverge. - Run the
verify-citationsskill onreferences.bib, always with--jsonso there is a machine artifact to reconcile against:check_bibtex.py <ws>/references.bib --json <ws>/citecheck.json. Fix or drop anything unresolvable; flag retractions to the user. - Set verified flags from that artifact, not by hand:
python3 scripts/corpus.py --corpus <ws>/corpus.json verify-audit --report <ws>/citecheck.json. The audit marksverified:yesonly for keys the report confirmed (status VERIFIED), clears any flag the report did not confirm, recordsverified_viaprovenance + the date, and echoes the report's verdict line verbatim. It exits nonzero on a PARTIAL-PASS or FAIL — so an incomplete run cannot read as clean.- If you set a flag manually instead,
corpus.py set KEY --verified yesnow requires--source <provider>(the index whose canonical record confirmed the entry) and records it; a bare--verified yesis rejected. - "Verified" means an authoritative index actually returned the canonical
record for this entry this session — not merely that some API echoed an
id once, and never with
fetched:false. A key markedverified:yeswith no provenance is a broken gate;corpus.py statsnow WARNs when it sees one. - Mirror the verify-citations verdict line verbatim in the review's prose and any summary/README: the exact verdict (PASS / PARTIAL-PASS / FAIL) plus raw counts (N verified, M warnings by flag, K skipped checks). Never upgrade a PARTIAL-PASS to "verified", and never collapse "PARTIAL-PASS, 11 WARN, 1 skipped" into "verified, 0 errors".
- If you set a flag manually instead,
Phase 7 — Draft
- Write
review.mdfollowing the structure templates in references/review-structure.md. - Every factual claim about prior work cites a corpus key:
[@key]or\cite{key}. Claims must trace back to a note anchor — if there is no note, do not write the claim.
Phase 7b — Reconcile forward references
The draft (and any related-work plan) names methods, baselines, and backbones in prose as slots — "we extend X", "compared against Y", "built on the W backbone". Make sure none ship unbacked:
- Run
python3 scripts/forward_refs.py <ws>/review.md [plan.md] --corpus <ws>/corpus.json. It extracts every named method/baseline/ backbone, diffs against the verified corpus, and emits the unresolved ones as a retrieval worklist. - For each genuine missing work, loop it back through the full pipeline (find-papers → screen → fetch-paper → extract → verify-citations) before the corpus is declared complete. Dismiss only the entries that are not citable works (the user's own system, a dataset, a metric) — never invent a citation to clear the list.
- Re-run until the worklist holds no real references.
Phase 8 — Gate
- Assert the corpus and
references.bibkeys are in lock step:python3 scripts/corpus.py --corpus <ws>/corpus.json check-keys <ws>/references.bib. If it fails, regenerate the.bibfrom the corpus (corpus.py bibtex) — never reconcile by hand-editing keys. - Run
python3 scripts/check_review.py <ws>/review.md --bib <ws>/references.bib. It fails on unknown keys, unverified or excluded citations, any cited key missing from the.bib(key drift), placeholder markers, and verbatim quotes over 40 words; it warns on uncited included papers and uncited long paragraphs. - Fix and re-run until
RESULT: PASS. Deliver only a passing document, and tell the user the review passed the coverage gate.
Output
lit-review/<slug>/review.md— the themed review, every claim cited, gated bycheck_review.py.lit-review/<slug>/corpus.json— auditable trail: searches, screening decisions with reasons, verification status.lit-review/<slug>/references.bib— verified BibTeX.lit-review/<slug>/notes/— per-paper grounded notes (paraphrase only).
Guardrails
- Never fabricate or embellish a citation. A paper enters the review only via
a real search hit, and a citation ships only after
verify-citationspasses it. If evidence for a claim is missing, say so instead. - Copyright: paper text and abstracts are fetched on demand and processed transiently — never written into the workspace, the repo, or the review. Metadata (DOI, title, authors, BibTeX fields) is fine. Quotes < 25 words, marked and cited.
- Politeness: all network access goes through the sibling skills' scripts
(≤1 req/s per host,
CONTACT_EMAILUser-Agent, 429 backoff,.cache/). - Never submit or post anything on the user's behalf.
- Present screening decisions and the final theme structure to the user for approval — the user owns scope judgments.
References
- references/methodology.md — query design, venue enumeration patterns, multi-seed snowballing with role-based inclusion and a co-citation sanity check, stopping criteria, screening protocol, forward-reference reconciliation.
- references/claim-extraction.md — grounded note format, anchor and strength labels, copyright rules for notes.
- references/review-structure.md — review
document templates, synthesis matrix, citation syntax, handoffs to
draft-related-workandwrite-abstract.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

