agentsclimarketplace
Sign in

Draft survey

Skill ShaishavMaisuria/research-paper-lifecycle-skills/skills/draft-survey

Turns a topic into a ranked reading list of the best papers plus a two-column, arXiv-ready literature-review/survey draft grounded in verified citations. Use when a researcher says "write a survey on X", "do a literature review of X", "give me a ranked reading list on X", "what should I read on X first", "draft a survey paper for arXiv", or "two-column related-work survey on X". Searches across DBLP/Crossref/Semantic Scholar/arXiv, ranks candidates by citation impact, venue strength, recency, and citation-graph centrality, clusters them into a taxonomy, verifies every reference, and writes an original-prose survey (intro, taxonomy, per-theme synthesis, open problems) where each claim cites a real, resolved paper. Never fabricates citations, never copies source text — original synthesis only; the author stays the author. Trigger words - survey, literature review, reading list, ranked papers, what to read, arxiv survey, two-column review.From its SKILL.md

Install
npx -y skills add ShaishavMaisuria/research-paper-lifecycle-skills --skill draft-survey

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 23 stars23 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
  • runs commandsInstructs the agent to run 2 commands, including `python3 scripts/rank_papers.py candidates.json` and 1 more.

SKILL.md

5.0 KB, 984 tokens by cl100k_base, as published. Nobody here has run it

Draft Survey

Two deliverables from one topic: a ranked reading list (what to read, best first, with a one-line why) and a two-column, arXiv-ready survey draft that synthesizes the area in original prose with verified citations. Composes find-papers, verify-citations, study-exemplars, and draft-related-work.

When to use vs. literature-review

  • literature-review builds the related-work for your own paper — scoped to your contribution.
  • draft-survey produces a standalone survey/review document (a reading list + a publishable 2-column draft) of a whole area. Use it to learn a field fast or to draft a survey paper for arXiv.

Inputs

  • The topic (e.g. "geospatial data conflation"), optionally a sub-scope and a target length (default 5–6 two-column pages).
  • Optional: a venue/format (default: generic two-column article, which compiles on arXiv).

Process

  1. Gather candidates broadly. Run find-papers across DBLP + Crossref + Semantic Scholar + arXiv for the topic and its synonyms; then run its citation-graph expansion so seminal anchors and direct lineages are not missed (a survey that omits the foundational papers is a weak survey).
  2. Rank them. Run python3 scripts/rank_papers.py candidates.json — a composite of normalized citation count, venue tier, recency, and citation-graph centrality (weights documented in references/ranking-criteria.md). Output the ranked reading list: rank, title, authors, year, venue, a citation/impact signal, and a one-line why read this (seminal / survey / SOTA / dataset / contrarian). Keep seminal and recent both represented.
  3. Verify every entry. Route the list through verify-citations so each has a real DOI/arXiv id; drop or flag anything unresolved. A survey with a fabricated reference is disqualifying.
  4. Build a taxonomy. Cluster the verified papers into 3–6 themes/sub-problems (the survey's section structure), each with its lineage (foundational → recent).
  5. Draft the survey, in original prose. Write a two-column .tex: abstract, introduction (scope + why a survey now), one section per theme (synthesize and contrast methods — never copy source sentences), a cross-cutting comparison (a table helps), open problems / future directions, conclusion, and \bibliography. Every claim cites a verified paper. Target the requested length.
  6. Make it arXiv-ready. Ensure it compiles (latexmk), uses a portable two-column class, and the .bib is clean. arXiv has no peer-review desk-reject, but it expects a compilable source and a real abstract; preflight-check can sanity-check length/structure.

Output

  • reading-list.md — the ranked, verified, annotated reading list.
  • survey.tex + refs.bib — the two-column, arXiv-ready draft (original prose, verified citations).
  • Both written to paper-workspace/research/ or a path the user names.

Guardrails

  • Real citations only. Every reference is resolved via verify-citations; never invent a paper, DOI, author, or year to fill a gap.
  • Original synthesis, not copying. Summarize and contrast in your own words; do not paste sentences from abstracts or papers. No paper content is bundled (fetch on demand, process transiently).
  • No completeness theater. State the search scope and that a survey is never exhaustive; flag themes where coverage is thin rather than padding.
  • Copilot, not pilot: the author reviews the reading list and the draft, and is the author of any submission.

Source verification

Citation impact and venue facts come from live scholarly APIs, not memory, and each cited paper carries a resolvable identifier the user can open.

Memory

Uses .paper-memory/ as described by paper-memory-convention.md: read prior reading lists/lessons for the topic at start; append the topic, the chosen taxonomy, and any coverage gaps at end (deduped).

What ships with it: 2 files

6.3 KB alongside SKILL.md, 1 of them executable

references/

scripts/

Gives 0 of the 12 instructions most docs writing skills give in 984 tokens

Counted across 1,951 of the 3,904 authors here whose files we hold, read 2026-09-06

  • Use third-person for skill descriptionsin 54 of 1951, across 35 files
  • Start descriptions with Use whenin 43 of 1951, across 29 files
  • Run baseline scenarios before writing any skillin 40 of 1951, across 26 files
  • Use active voicein 40 of 1951, across 36 files
  • Map file responsibilities before defining tasksin 36 of 1951, across 29 files
  • Use checkbox syntax for tracking stepsin 35 of 1951, across 27 files
  • Ask one question at a timein 35 of 1951
  • Offer execution options after saving the planin 33 of 1951, across 24 files
  • Include complete code in every stepin 33 of 1951, across 27 files
  • Design units with clear boundaries and interfacesin 31 of 1951, across 23 files
  • Announce the skill usage at the startin 30 of 1951
  • Verify agent compliance after adding the skillin 29 of 1951, across 17 files

Said here and by no other author read

  • gather papers using citation graph expansion
  • rank papers by citation impact and venue strength
  • verify every citation using a scholarly API
  • cluster verified papers into a thematic taxonomy
  • write original prose survey in two-column format
  • include abstract and introduction in survey draft

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.