Paperpile

Paperpile

Manage your Paperpile library and resolve citations to PDFs via the paperpile CLI.

CLI-first — this is the whole point of the skill. Every library operation (add, search, fetch, label, edit, trash, auth) goes through the paperpile CLI, which is pure HTTP. Never drive the Paperpile web app (app.paperpile.com) via browser automation to add or manage references — it is slower, the React forms reject synthetic input, and the CLI already does it. If the CLI fails, it is almost always stale auth — refresh cookies (below), do not reach for the browser.

Prerequisites

• paperpile binary at ~/.local/bin/paperpile (Bun-compiled from ~/projects/paperpile-cli)
• Valid auth cookies. If paperpile auth fails, refresh them automatically (no manual Cookie-Editor export needed): ${CLAUDE_SKILL_DIR}/scripts/refresh-auth-from-dia.sh pulls the cookies from the logged-in browser over CDP (:9222 — Dia on macOS, Chromium on Linux), imports them, and verifies.
• For PDF resolution: Chrome running with CDP on port 9250 (dedicated instance at ~/.config/chrome-cdp)

Library Management

Key behaviors:

• add auto-detects input type:
  • DOI (10.xxx) -- lookup metadata via Guru, create entry
  • Web URL (https://...) -- create entry with url:[] field (no --force needed)
  • Remote PDF URL (https://.../*.pdf) -- download PDF, copy to Google Drive, attach to entry
  • Local PDF (/path/to/*.pdf) -- copy to Google Drive sync folder, create entry with PDF attached
  • All modes accept --title, --author, --year, --pubtype for metadata
  • --force only needed for DOIs when Guru metadata is unavailable
• edit updates metadata fields on existing items via sync API. Dry-run by default -- pass --confirm to apply.
• fetch resolves a bibkey to PDF via Paperpile API + Google Drive download.
• search scores against a local index cache. Run paperpile index first.
• trash and label delete are dry-run by default -- pass --confirm to execute.

Find and Add

One command: citation string → paper in Paperpile with PDF.

paperpile find-and-add "<citation>" [--doi DOI] [--ssrn ID] [--title T] [--author A] [--year Y] [--journal J] [--volume V] [--page P] [--json] [--no-pdf]

Examples

# Law review article (HeinOnline)
paperpile find-and-add "Robertson, Passive in Name Only, 36 Yale J. on Reg. 795 (2019)"

# SSRN working paper
paperpile find-and-add --ssrn 5093097

# Paper with known DOI
paperpile find-and-add --doi 10.1016/j.jfineco.2024.01.001

# Multi-author law review
paperpile find-and-add "Montagnes, Peskowitz and Sridharan, How Well Do Voting Choice Policies Represent Public and Investor Preferences, SSRN 5093097 (2024)"

What it does

1. Parse citation string → extract author, title, journal, volume, page, year, DOI, SSRN ID
2. Discover metadata: Guru title search → CrossRef → OpenAlex (law reviews skip CrossRef — too many false positives)
3. Dedup against cached library index
4. Add to Paperpile via POST /api/library
5. Get PDF (automatic, no manual steps):
   • Guru pdf_url preprint/publisher download (pure HTTP)
   • HeinOnline via UVA EZproxy CDP (law reviews with volume/page/known journal)
   • EZproxy/OpenAthens/SSRN via native CDP
   • Google filetype:pdf search via CDP (final fallback)
6. Copy PDF to Paperpile Google Drive sync folder (~/Google Drive/.../Paperpile/All Papers/<Letter>/)

Supported law review journals (HeinOnline)

UCLA L. Rev., Yale J. on Reg., Yale L.J., Harv. L. Rev., Stan. L. Rev., Colum. L. Rev., Mich. L. Rev., Va. L. Rev., U. Pa. L. Rev., N.Y.U. L. Rev., Chi. L. Rev., Geo. L.J., Duke L.J., Cornell L. Rev., Nw. U. L. Rev., Tex. L. Rev., B.U. L. Rev., and more (see heinonline.ts in paperpile-cli).

Key design decisions

• Law review citations with volume+page+known journal handle skip CrossRef entirely (unreliable for HeinOnline-only journals) and construct metadata from parsed citation fields
• Paperpile's PDF crawler runs in the browser extension, not server-side -- neither POST /api/library nor POST /api/sync triggers it. PDF resolution is always active (CDP or HTTP).
• Guru title search works for finance/econ journals but returns 0 for law reviews (HeinOnline-only)
• Shibboleth cookies (shibidp.its.virginia.edu) are now snapshotted -- EZproxy re-auth is transparent across Dia restarts

Edit Metadata

Update title, author, year, and other fields on existing library items.

paperpile edit <bibkey-or-_id> --title "..." --author "..." --year 2024 [--confirm]

Supported flags: --title, --author, --year, --journal, --volume, --issue, --pages, --abstract, --url, --doi, --publisher, --language, --pubtype, --citekey

Examples

# Edit title and year (dry-run)
paperpile edit Smith2024-ab --title "New Title" --year 2025

# Apply the edit
paperpile edit Smith2024-ab --title "New Title" --year 2025 --confirm

# Institutional author (& treated as single author)
paperpile edit abc123 --author "Davis Polk & Wardwell" --confirm

# Multiple personal authors (comma-separated "First Last" pairs)
paperpile edit abc123 --author "Jack Pitcher, Emily Glazer" --confirm

# Set a DOI (e.g. after a PDF-first add mangled the metadata)
paperpile edit abc123 --doi "10.1111/jofi.12422" --confirm

# SSRN working paper: real DOI is 10.2139/ssrn.<abstract_id> + the abstract URL
paperpile edit abc123 --doi "10.2139/ssrn.666962" \
  --url "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=666962" --confirm

Fixing PDF-first metadata damage: paperpile add <local.pdf> lets Paperpile's server-side PDF-text extractor populate metadata, which often drops co-authors, blanks the year, or duplicates the title. Prefer DOI-first adds. To repair an existing entry, set the correct DOI and repopulate authoritatively from Paperpile's own Guru resolver rather than hand-typing every field. SSRN papers do have real DOIs (10.2139/ssrn.<abstract_id>), but they resolve to preprint metadata — safe for genuine working papers, but don't attach them to a version that was later published (it regresses the journal/volume/page citation).

Key behaviors:

• Dry-run by default -- shows what would change without applying. Pass --confirm to mutate.
• Resolves bibkeys to pub _id via the local index (same as trash).
• Uses POST /api/sync?v=3 with action: "update" to patch fields.
• Author strings with & are treated as institutional (single author). Comma-separated "First Last" pairs are split into multiple personal authors.

Utilities

warmup.sh auto-clicks the NetBadge cert login via CDP. Runs every 25 min via launchd (com.paperpile.warmup.plist).

WRDS SOCKS Tunnel (preferred for PDF acquisition)

If the /fetch-paper skill is available (lives in dotfiles, not this plugin), prefer it for downloading paywalled PDFs. It uses a WRDS SOCKS tunnel to present a Penn IP — which is registered with publishers for IP-based access — and bypasses EZproxy entirely.

Check availability:

which wrds-tunnel && which fetch-paper-browser && echo "fetch-paper available" || echo "fetch-paper not available — falling back to EZproxy/CDP"

If available: use /fetch-paper for PDF acquisition, then attach the result:

• entry already in the library (a DOI add that got no PDF) → paperpile attach <_id> <local.pdf> --confirm. Do NOT add the PDF — that creates a SECOND entry and lets the server's PDF-text extractor write the metadata (see the damage note above).
• nothing in the library yet → paperpile add <local.pdf>. If not available: use the EZproxy/CDP pipeline in find-and-add as before.

See ${CLAUDE_SKILL_DIR}/references/institutional_access.md for technical details.

Integration

fetch-paper (dotfiles) → download PDF via WRDS SOCKS
Paperpile (this skill) → cite-check (upload PDFs to Gemini)
                       → nlm (upload to NotebookLM)

Auth & Data

• Cookies: ~/.claude-work/skills/paperpile/cookies/<domain>.json
• Cache: ~/.claude-work/skills/paperpile/cache/paperpile-index.json
• Paperpile All Papers: ~/Library/CloudStorage/GoogleDrive-eddyhu@gmail.com/My Drive/resources/Paperpile/All Papers/
• Cookies expire (~30 days for Paperpile, ~8-12h for Shibboleth hard expiry). To refresh, run ${CLAUDE_SKILL_DIR}/scripts/refresh-auth-from-dia.sh — it extracts the live cookies from the logged-in browser via CDP (:9222 — Dia on macOS, Chromium on Linux) and paperpile auth imports them (no Cookie-Editor export). Requires being logged into Paperpile in that browser; if not, log in there first.

Red Flags