[GLouv]

What this is

One release that combines three tracks that touch the same files, merged in this order:

1. 0.4.1: honest background consent text, LICENSE in the package, patched dependencies #11 (proeve/0.4.1): honest background consent text, LICENSE in the npm package, patched dependencies.
2. 0.4.1: the web goes through the Chrome you already use #10 (feat/browser-overlevering): instructions for handing web work to the Chrome the person already uses.
3. The Chrome track: the server reads Chrome pages through the accessibility layer, blacks out password and payment fields
   before anything leaves the machine, refuses to type into a password manager's own page, and asks before any click that sends.

0.4.0 was never published to npm (npm has 0.1.0), so 0.4.1 is the first npm release since 0.1.0. See CHANGELOG.

Merge fixes made here

• 0.4.1: the web goes through the Chrome you already use #10 said "This server does not see inside pages" in the same instructions where the Chrome track says it does. Rewritten.
• One 0.4.1 section in CHANGELOG (no duplicate heading), with 0.4.0 marked as never published.
• Send verdict: an ancestor's name can only add danger, never remove it; button names in the question are shown on one line.
• scripts/release.sh puts only the helper and status icon that the receipt proved into the npm package (sha1 / tree sum
  required). The working tree is a snapshot after step 0 and must be unchanged before the tag and before npm, and the
  script packs the release itself, measures the helper and icon inside the tarball, and publishes exactly that tarball.
  An unknown argument stops (—tjek with a long dash was a full release). The GitHub release asset and the .mcpb
  are built by CI from the tag, not the proven binary.
• A name only counts as a name when it has a letter or a digit, so a space or an invisible mark cannot switch off the
  «we don't know what this is» rule in the danger gate; in the send gate, recipient and message must be visible, and an
  invisible mark between or after words in a browser window's title is read as a space, so it cannot remove the gate there.
  (A mark inside a name, such as «Wh»+U+00AD+«atsApp», is not caught - the same as in 0.4.0.)
• Test 15b/15c: no longer racing the shutdown signal under load; a missing SIGTERM-before-SIGKILL now fails the test.
• Test ikon-lyd check 13 (red in this PR's first CI run): it now allows the standing screen loan's early return, on its own
  line and with exactly that condition, before the sound. It checks the condition, not the rest of that line (tightening it
  is listed for 0.4.2).

Evidence

• On 27d9fc0 (which only changes test/ikon-lyd.mjs over e9ca16a): the 56 tests of run-all.sh that do not take the screen or
  use Chrome: 55 green, and optagelse-udelukker failed once under heavy load and passed when run alone. ikon-lyd was red on
  e9ca16a (this PR's first CI run) and is green on 27d9fc0. The tests
  that take the screen or Chrome run in the dialog receipt. Page check green; sync-tal.py idempotent. New guards mutation-proven (each mutant turns exactly its scenario red).
• Receipt for helper 5a3784bd: six live Chrome tests passed twice, nine mutants all red, Chrome's screen-reader switch
  measured off afterwards (on cbfc2a9 6 Nov 16:51 CET; on 4ac2797 6 Nov 17:53 CET; on e9ca16a 6 Nov 22:51 CET; on 27d9fc0 6 Nov 23:35 CET)
• Review: two-seat panel R1-R18; R19-R24 on the final commits.

Known and accepted

O4, O5 (need a new helper), O7, O8, and a policy.js comment, as listed in review R18.

🤖 Generated with Claude Code