Skip to content

test: two source checks follow the Chrome track's code (claims 27, skriv-ankommer 6g) - #13

Merged
GLouv merged 2 commits into
mainfrom
fix/0.4.1-proever
Nov 7, 2026
Merged

GLouv merged 2 commits into
mainfrom
fix/0.4.1-proever

Conversation

@GLouv

@GLouv GLouv commented Nov 7, 2026

Copy link
Copy Markdown
Member

The full suite with the real dialog (CMCP_DIALOGS=1 ./test/run-all.sh, the dialog receipt CI does not run) was red on main after #12, on two checks that read the helper's source too strictly. No product code changes; the helper's source tree is unchanged.

  • claims 27 read vaultNavne in Accessibility.swift (search words that recognise a Danish password-manager page by its title, never shown to anyone) as Danish text a person reads. Only that list is exempt, closing at its first ]; anything after ], and any Danish string before or after the list, is still red, and a list that never closes is red.
  • skriv-ankommer 6g required the secure-field guard in set-value on one line; the Chrome track split it over two lines. Both conditions are still required.

Evidence: the two checks run alone in a harness - red on main exactly as in the dialog run, green with this PR; a Danish sentence after the list, a comment / .map / ; static let after its ], and the guard without isSecure or without the macOS signal each turn them red. Reviewed (R25). The full claims.mjs and skriv-ankommer.mjs run in the next dialog receipt.

🤖 Generated with Claude Code

Gustav Louv and others added 2 commits November 7, 2026 01:16
…riv-ankommer 6g)

The dialog receipt on main 9f66e2e (the full run-all.sh, which CI does not run) found two checks
that read the helper's source too strictly:
- claims 27 read vaultNavne in Accessibility.swift - search words that recognise a Danish
  password-manager page by its title, never shown to anyone - as Danish text a person reads.
  Only that list is exempt, from its declaration to its «]»; a Danish string before or after it
  is still red.
- skriv-ankommer 6g required the secure-field guard in set-value on one line; the Chrome track
  split it over two (and put the vault check in front). Both conditions are still required.
Calibrated: the old checks are red on main, the new ones green; a Danish sentence after the list,
and the guard without isSecure or without the macOS signal, each turn them red.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Opus R25 (measured): the exemption ended only on a line that ends in «]», so a list closed by
«]  // comment», «].map {…}», «] + x» or «]; static let …» left the rest of Accessibility.swift
unread and the check green. It now closes at the first «]»; if anything follows it, the line is
read like any other (and turns red on the list's own words), and a list that never closes is red.
Measured with the harness: base green; A1, A3, A4, A6 red. (An unterminated list cannot compile,
so a list that closes at a later «]» is not a reachable state.)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@GLouv
GLouv merged commit c59ab91 into main Nov 7, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant