Report privately, not in a public issue: security@dargel-solutions.de.

Please include the version, the operating system, and — if the finding involves a document — the smallest HTML that reproduces it. A first response goes out within 72 hours. If a fix is warranted, the advisory names the reporter unless the reporter asks otherwise.

There is no bounty programme.

This tool executes foreign HTML, including its scripts. That is not a side effect to be engineered away — it is the job. Whether a page works can only be decided after the renderer and the paginator have run, and both run the document's own code.

What the tool does about it, and each of these is checked in the test suite rather than merely intended:

• A fresh browser profile per run, created with mkdtemp and removed afterwards. A run that cannot verify its own cleanup fails rather than reporting success.
• The browser sandbox stays on. There is no --no-sandbox anywhere in this repository and no flag that turns it off.
• The network is blocked by default. Request interception is on, the default policy is offline, and only the tool's own loopback origin plus data:, blob: and about: are let through. --allow-network <origin> opens exactly one origin per use and nothing else.
• No browser-wide file-access switch. The document is served from a loopback origin instead. A test measures that a document loaded this way cannot read a neighbouring file, because removing the reason for a switch and leaving the switch in place is a mistake that was actually made here once.
• The measurement primitives are captured before any author script runs, so a document that replaces getBoundingClientRect or getComputedStyle after pagination cannot forge its own measurements. A hostile document is measured correctly or the run fails; it does not produce a quietly wrong report.
• The document's Content Security Policy remains authoritative. The measured Paged.js bundle crosses the browser-driver boundary only after authored loading. breaklint does not disable CSP browser-wide merely to install its own apparatus; its own CSP-bearing HTML report is the live regression for that boundary.
• No model, no network client, no telemetry. The only runtime dependency is an HTML parser. Nothing is uploaded, and nothing about your documents leaves the machine.

What this is not. None of the above is protection against a determined attack on the browser sandbox itself. For untrusted third-party HTML, run this in a container, the same way you would run any other tool that executes code you did not write.

The stable Paged.js package still declares @babel/polyfill, which in turn retains the obsolete core-js 2 line. The current dependency audit reports no known vulnerability, and breaklint's measured integration loads Paged.js's browser bundle rather than importing those polyfill modules as application code. The available 0.5.0 beta retains the same polyfill dependency, while a larger unmeasured upgrade would cross the paginator trust boundary and invalidate the current live, visual, consumer and supply-chain evidence.

This is a time-bounded dependency-hygiene risk acceptance, not a claim that obsolete transitive packages are desirable. Re-review is due by 2026-11-24, or earlier if Paged.js publishes a compatible stable release without the chain, an advisory reaches the installed runtime path, the bundle/import surface changes, or Node/browser support changes. Any replacement must pass the full live, visual, consumer, licence, advisory and release-integrity gates before the pin moves.

In scope: anything that lets a checked document escape the sandbox, read files outside the run, reach the network against the policy, survive past cleanup, or cause the tool to report a clean result for a document it did not actually measure — the last one included deliberately, because a checker that can be made to go green is a security problem in the build it is guarding.

Out of scope: findings that are wrong because a threshold is wrong. Every threshold here is uncalibrated and says so; a false positive is a bug, not a vulnerability.

The latest published version on npm receives fixes. The 0.2.x line is the supported line, and each published patch supersedes earlier 0.2.x packages. breaklint@0.2.0 opened that line on 2026-08-22 with npm provenance; the 0.1.x line receives no separate long-term-support branch. The npm package page remains the authority on which patch is actually published.