paper-deep-readinglisted

# Paper deep reading One paper, read properly: what it does, whether its conclusions hold, and whether the user should act on it. The second half is the point — anything can summarise a paper, and a summary is not what someone asks for when they say "read this properly". `scripts/fetch.py` gets the PDF or says why it could not. `scripts/note.py` enforces a shape — every claim credited to the paper has to name where in the paper it lives. The reading, the judging and the writing are **not** scriptable, and no attempt is made to script them. ## Workflow **Invoke the scripts by path** — `python3 scripts/fetch.py`, from any working directory. They import each other by bare name, and Python puts the script's own directory on `sys.path`, so `scripts/` becomes the import root on its own. ### 1. Get the paper ```bash cd /path/to/skills/paper-deep-reading python3 scripts/fetch.py <reference> --out-dir /tmp/papers > /tmp/fetch.json ``` `<reference>` can be a local `.pdf` path, an arXiv id or URL, a DOI, a bioRxiv/medRxiv link, a PMCID or a PMID. Three routes are covered — arXiv, bioRxiv/medRxiv, and Europe PMC's open-access mirror of PubMed Central. A paywalled journal article is not one of them, and the script will say so instead of returning something that looks like a paper. Read `fetch.json` before anything else: - **`fulltext: "abstract-only"`** — ⚠️ stop and tell the user. Something can still be written, but it is a summary of an abstract and must be labelled that wa