Defuddle-extracted article content is injected raw into the PDF HTML
and rendered by a real browser (browserless). As defense-in-depth
against a compromised or malicious source page, remove script/style/
iframe/object/embed elements and on* event-handler / javascript: href
attributes from the parsed DOM before serializing it into the template.
The generated script's setup comment hardcoded school.second-brain.ru
even though baseUrl is already interpolated elsewhere in the template,
so the comment would lie on any other host. Also add a regression test
that the generated script parses as valid JavaScript, to catch escaping
mistakes in the template literal.
Connects to browserless over CDP (playwright-core), extracts article
content with Defuddle/JSDOM, renders it through buildCleanHtml, and
prints a PDF on a second page. Adds an in-browser request filter as a
second line of SSRF defense against redirects to private/localhost
hosts, on top of assertPublicUrl's DNS check.
Integration test is gated by RUN_PDF_INTEGRATION=1 (describe.runIf) so
it is skipped in a normal npm run test and only runs against a real
browserless instance.
admins