Set up Cursor Cloud dev environment for paperscraper

AGENTS.md

@@ -0,0 +1,27 @@
+# AGENTS.md
+
+## Cursor Cloud specific instructions
+
+`paperscraper` is a single Python library (no server, database, or GUI). Development
+uses [`uv`](https://docs.astral.sh/uv/); dependencies are defined in `pyproject.toml`
+(dev extras in the `dev` dependency group) and locked in `uv.lock`. Standard
+build/lint/test/run commands live in `CONTRIBUTING.md` and `.github/workflows/test_tip.yml`.
+
+Notes for working in this environment:
+
+- `uv` installs to `~/.local/bin`. The startup update script installs it there and
+  runs `uv sync --group dev`. Interactive shells pick it up via `~/.bashrc`; if `uv`
+  is ever not found, use `~/.local/bin/uv` or run `source ~/.local/bin/env`.
+- Many tests and all "scrape" workflows hit live external scholarly APIs (arXiv,
+  PubMed, bioRxiv/medRxiv/chemRxiv, Semantic Scholar, Google Scholar). They are
+  therefore network-dependent, slow, and can be flaky/rate-limited. arXiv in
+  particular returns HTTP 429 if you fire multiple queries back-to-back; space out
+  arXiv calls (a few seconds) when running demos or tests. GitHub Actions is the
+  source of truth for release readiness.
+- Optional API keys/credentials (`SS_API_KEY`, publisher tokens, AWS, Kaggle) only
+  improve rate limits and PDF/dump fallbacks; the core library and most tests run
+  without them.
+- `paperscraper/server_dumps/*.jsonl` (downloaded preprint dumps), `dist/`, `build/`,
+  and `*.egg-info` are generated artifacts and are gitignored — do not commit them.
+- Run a single test module to avoid the slow full network suite, e.g.
+  `uv run pytest paperscraper/tests/test_dump.py`.

paperscraper/pdf/__init__.py

@@ -1,2 +1,7 @@
-from .pdf import load_api_keys, save_pdf, save_pdf_from_dump, debug_save_pdf, debug_save_pdf_from_dump  # noqa
-
+from .pdf import (  # noqa
+    debug_save_pdf,
+    debug_save_pdf_from_dump,
+    load_api_keys,
+    save_pdf,
+    save_pdf_from_dump,
+)