Instructions for AI coding agents and contributors. See also CLAUDE.md (read its warning about the real vault first).
Run exactly this and read the output:
scripts/verify.sh
Exit code 0 = pass. Anything else = not done. CI runs the same script (job verify) plus scripts/check-test-integrity.sh (job test-integrity).
Rules for AI agents and humans:
- Reproduce first. For a bug, write or run a failing test/command that shows it before editing; the fix is done only when that flips to green.
- Do not weaken tests. Never delete tests, add skip/ignore/xfail markers, remove assertions, or loosen expected values to get green.
test-integrityfails the PR if you do; only the owner may approve via the PR labeltest-change-approved. - Report only what was run. State the exact command and its real result. If something was not run or could not run, say so. Never claim success without the output.
- Do not edit
.github/,scripts/verify.sh, orscripts/check-test-integrity.shto make a failing check pass; those are owned by @grloper (CODEOWNERS).
Not checked by scripts/verify.sh: Windows-only behaviour, the PyInstaller/Inno release build, the real GUI. verify.sh sets NOMOREPWN_DATA to a temp dir so it never touches a real vault.