curate-peptide-papers

name

curate-peptide-papers

description

Curate the awesome-peptide paper list. Use when Codex needs to pull latest notes from the vendor/paper-read submodule, scan paper-read for peptide-related papers, enrich paper metadata through Crossref or arXiv, find, verify, add, classify, deduplicate, or update peptide deep-learning papers in data/paper.csv or data/paper-read.csv; manage DOI, code, dataset, blog, quality, pinned, abstract, or tag metadata; regenerate README.md from the CSV sources; or validate paper-list consistency for this repository.

Curate Peptide Papers

Use this skill for content curation in awesome-peptide. The manually curated paper data lives in data/paper.csv; paper-read sourced entries live in minimal form in data/paper-read.csv; README.md is generated from both CSV sources plus DATABASE.md and the Liquid templates. vendor/paper-read is the local submodule source for discovering newly reviewed paper notes.

Quick Start

1. Inspect the worktree before changing content:

git status --short

2. Read the relevant references:

• references/repo-content-map.md for source files and generation flow.
• references/paper-curation.md for CSV field rules and curation policy.
• references/paper-read-source.md when scanning the paper-read submodule.

3. Edit data/paper.csv or data/paper-read.csv, then validate and regenerate:

pip install -e .
awe-check
awe-pep
git diff -- data/paper.csv README.md

If the command names are unavailable, run the modules directly after installing dependencies:

python -m awepep.check
python -m awepep.main

Run the tag audit after changing tags:

python .codex/skills/curate-peptide-papers/scripts/audit_tags.py

Paper-Read Workflow

Use this workflow when the user asks to pull latest papers from paper-read or scan for peptide papers:

git submodule update --init --recursive
git submodule update --remote vendor/paper-read
python .codex/skills/curate-peptide-papers/scripts/scan_paper_read_candidates.py --limit 200 --enrich-crossref

Then automatically curate high-confidence candidates into data/paper-read.csv:

1. Review the scanner Markdown and JSON reports.
2. Skip candidates already present by DOI.
3. Verify DOI, code/data links, and peptide/deep-learning relevance from the note plus primary sources.
4. Choose sec and subsec from the current README taxonomy; bibliographic metadata must not be used as a substitute for classification.
5. Add clear candidates to data/paper-read.csv with doi,title,source,sec,subsec,code,dataset,quality,pined,tags.
6. Let README generation fetch authors, venue, publish date, DOI link, and abstract from Crossref or the arXiv fallback.
7. Run awe-check and the no-write generation smoke test.
8. Summarize added and skipped candidates.

Do not edit files inside vendor/paper-read; it is an external source submodule.

Curation Workflow

For adding or updating papers:

1. Verify bibliographic metadata from primary sources. Prefer DOI landing pages, arXiv/bioRxiv, journal pages, PubMed, Crossref, and official code or dataset repositories. Browse for recent papers or any metadata that could have changed.
2. Check for an existing entry by DOI and title before adding a row.
3. Classify with the exact current section/subsection values from awepep/config.py.
4. Keep publications in Markdown DOI-link form so awepep.check can extract the DOI.
5. Default optional fields to blank. Use quality=high only for clearly important work. Use pined=true only for selected important papers and keep the count below max_pined = 30.
6. Prefer short, slash-separated canonical tags from awepep.config.tag_groups; add new tags to the vocabulary before using them.
7. Regenerate README.md only after CSV edits, then inspect the diff.

For data/paper-read.csv, keep only the minimal source row. Required decision fields are DOI and the agent-selected sec/subsec; optional title, source, code, dataset, quality, pined, and tags help display or audit the row. Use conservative official source links only.

CSV Row Drafts

Use the helper to produce a correctly ordered row without mutating data/paper.csv:

python .codex/skills/curate-peptide-papers/scripts/draft_paper_row.py \
  --title "Example peptide design paper" \
  --sec "Peptide Design & Generation" \
  --subsec "Structure-Based Design" \
  --authors "A. Author, B. Author" \
  --venue JCIM \
  --doi 10.1021/example \
  --publish-date 2026-6-16 \
  --code-label GitHub \
  --code-url https://github.com/example/project \
  --tags "Diffusion/Cyclic"

Review the printed row before appending it to data/paper.csv.

Guardrails

• Do not hand-edit generated paper sections in README.md; edit data/paper.csv and regenerate.
• Do not add papers without a DOI unless the user explicitly accepts that awepep.check will need to change.
• Do not put generated authors, venue, date, publication link, or abstract into data/paper-read.csv; those are generated live.
• Do not invent new section or subsection labels during paper-add tasks.
• Do not invent one-off tags. Unknown tags are warnings in awe-check, but they should be resolved before release.
• If dependencies are missing, install the package in editable mode before validating.
• If README.md generation fails for a paper-read DOI, verify the DOI with Crossref, arXiv, or the DOI landing page before changing classification or source rows.

References

• Read references/paper-curation.md before editing data/paper.csv.
• Read references/repo-content-map.md when locating source files or explaining the generation flow.
• Read references/paper-read-source.md before using the paper-read scanner or submodule.