Repository navigation
Shorten the Touch ID prompt - #78
Conversation
enable-fingerprint no longer prints the Secure Enclave explanation (the shared anti-replay state, the FileVault and DFU worst case, the Neo factory-records note) or asks "Turn on Touch ID? [y/N]". Where Touch ID reads available, or ready after being turned off earlier that boot, it execs touchid-enable straight away, and a failure there still stops the setup. The already-on path, the driver options write, the not-ready, failed and unsupported messages and the fall-through that lets other readers enrol are unchanged. The Neo and earlier-key-store checks existed only to tailor that text, so they go with it. With nothing to agree to, /etc/omarchy-mac/touchid-on is described as the record that Touch ID is on rather than the owner's agreement, in comments, the README and tests. touchid-enable and touchid-status read it from OMARCHY_MAC_TOUCHID_RECORD (was OMARCHY_MAC_TOUCHID_AGREED), and touchid-enable's error for it now reads "its record could not be written". The record's path and contents, the driver options and the units behave as before. touchid-consent-test.sh becomes touchid-setup-test.sh. Its entrypoint cases run with standard input closed and check that an available Mac goes to touchid-enable asking nothing, that a failure there stops the setup, and that a Touch ID already on is not turned on again. No manual page describes the prompt, so the manual is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
[Medium impact] The current changes appear safe to merge; no actionable defect was found. SummaryThe current code keeps Touch ID consent, despite the PR title and description. It shortens the warning and moves the detailed explanation into the Mac manual.
Reviews (2) · Last reviewed commit: "Keep a short Touch ID prompt" · Reviewed by Greptile |
scottjones
left a comment
There was a problem hiding this comment.
We read through this against main and it looks right to us, so we're approving.
The only thing that left enable-fingerprint is the prompt and the text around it. Everything that decides whether Touch ID gets turned on is still there:
- The already-on path (
omarchy-mac/entrypoints/enable-fingerprint:13-48) is unchanged, including the options write under the lock and the restart message. failed,not-ready,unproven,blockedandunsupported:*still let setup go on with other readers (:59-67), andnoneor an empty status still exits 0 (:56-58).execmeans atouchid-enablefailure is still the entrypoint's exit status, so Omarchy's setup still stops (:54).- Board gating doesn't depend on the entrypoint.
lib/touchid-enable:89still refuses anythingtouchid-eligibilitydoesn't callavailable, so a stalereadystatus can't get past it.
The rename is complete. Nothing in this repo, omarchy or omarchy-mac-installer still reads OMARCHY_MAC_TOUCHID_AGREED, and nothing reads the record's contents, only whether it exists. So an install that already has /etc/omarchy-mac/touchid-on behaves exactly as before.
The new tests still cover the safety paths. With stdin closed, an available Mac has to call the stub and print nothing. A failing stub has to stop setup, and every not-ready, unsupported and no-sensor case has to call nothing. So if someone widened the case, dropped the failure, or brought back a read, a test would fail.
Three non-blocking things:
- The risk text is gone from the repo entirely. The anti-replay state shared with macOS, the FileVault on/off worst case with the DFU restore, and the Neo factory-records note now aren't in the README, the manual or the installer docs. We'd move a short version into the Touch ID part of
mac-manual/content/04-security.mdso it's still written down somewhere an owner can find it. #69 is already adding a Touch ID section there. omarchy-mac/versionis still 0.2.3, which omarchy-pkgs already ships. CONTRIBUTING.md:64 asks for a bump in the PR that changes behaviour. Main already has unbumped changes since that pin, so this may be handled at release. Just flagging it.- Over in omarchy,
bin/omarchy-setup-security-fingerprint:188-189asks for the password again because the owner "may have read past sudo's timeout before answering". With no prompt that second ask can probably go, along with the comment and docs lines you already listed.
Heads up for #69 and #76. Both touch the same README paragraph, so whichever lands second will need a rebase. #69 also edits touchid-consent-test.sh, the prompt text this removes, and the $agreed line in touchid-enable. And its SDDM record assumes the prompt told the owner about the login-screen change, including that a passwordless account can't log in there anymore. Once this lands, that disclosure needs a new home. #76 calls the record "package-owned consent" (preserve-existing-consent). Nothing breaks, since it only checks whether the path exists, but the wording will drift from this.
maralcbr
left a comment
There was a problem hiding this comment.
Thanks @DjDeveloperr. The code is clean: every gate stays, touchid-enable:89 still refuses anything that isn't available, a failure still stops setup, the already-on path is untouched, and the rename is complete. I'm not approving yet, because this reverses a deliberate safety decision and the PR doesn't show why that's now safe.
The policy change. #12 made Touch ID "turn on only after the Mac's owner agrees" on purpose. The prompt was the only place an owner learned that Linux and macOS share the enclave's anti-replay state in the iBoot System Container. A bug that writes it wrongly can leave a FileVault macOS unbootable, which takes a DFU restore that erases the Mac. On the Neo, setup also reads device-private factory records. Choosing "Set up fingerprint" shows the owner wants a fingerprint, not that they accept that risk on the macOS side. The diff doesn't touch the risk itself: the driver still writes.
"Touch ID is stable now" may well be true, but it needs to be shown before the warning goes. What would settle it:
- the driver/firmware versions this rests on, and what (if anything) now guards the shared state;
- field data: how many Macs and which boards, how long, how many enclave writes, and any incidents. Include macOS-side checks, FileVault on and off, after Linux use;
- interruption testing: power loss or a forced restart during an enclave write.
Without that, I'd keep a short version: two or three lines naming the macOS/FileVault worst case, then [y/N]. That drops most of the friction and keeps informed consent. Removing it entirely needs an explicit maintainer decision.
Also needed, whichever way that goes
- omacom/omarchy's
docs/lifecycle-dispatch.mddefinesenable-fingerprintas the hook that "explains … asks … turns the reader on only on yes". That's a contract, not just a comment, so it changes in the same breath as this, along with the secondsudoask inomarchy-setup-security-fingerprint:188-189that Scott mentioned. - If the prompt goes, the risk text needs a home an owner will find, such as the Touch ID section #69 adds to
mac-manual/content/04-security.md. - #69 relies on this prompt to say that an account with no password can't log in at SDDM once Touch ID is on. That disclosure must not disappear when both land.
omarchy-mac/versionneeds a bump for the behaviour change (CONTRIBUTING.md:64).
Minor: touchid-status:3 still says the status exists "so Omarchy can ask the owner".
|
Maintainer decision: keep a short consent prompt rather than dropping it. Please trim the current prompt to two or three lines (Touch ID writes to the Secure Enclave shared with macOS; a driver bug can leave FileVault macOS unbootable and need a DFU restore that erases the Mac) followed by With that, the docs asks shrink to:
The omarchy lifecycle-dispatch contract then stays as written. Happy to approve once that's in. |
Maintainer decision on this pull request: the owner still agrees before Touch ID turns on. enable-fingerprint now says in three lines that Touch ID is experimental and can be unstable, that it shares the Secure Enclave with macOS so macOS should be backed up first, and where the details are, then asks "Turn on Touch ID? [y/N]", default no. With the prompt kept, the rest of the earlier commit goes: the record keeps its agreement wording and OMARCHY_MAC_TOUCHID_AGREED, the units and comments read as they did, and touchid-consent-test.sh keeps its name. The full account moves to the manual's Encryption and passwords page: the shared anti-replay state, the FileVault on and off cases and the DFU restore, the key store and the Neo's sensor key. The test now checks the prompt's wording, that it stays at three lines and 80 columns, and everything it checked about the answer. omarchy-mac goes to 0.2.4. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
@maralcbr Done in 4e83a32, and sorry for the slow reply. The prompt stays at three lines plus Four things differ from your list:
The updated test fails against the old prompt. On an Ubuntu host 18 of 20 |
Summary
enable-fingerprintstill asks before Touch ID turns on, but the long explanation of the Secure Enclave driver is now three lines:The answer works as it did. Yes runs
touchid-enable. No, an empty answer or a closed terminal leaves Touch ID off. The Neo and earlier-key-store checks only chose which lines to print, so they go.The text that left the prompt is now on the manual's Encryption and passwords page, under "Turning on Touch ID". That covers the shared anti-replay state, FileVault on and off, the DFU restore, the key store, the Neo's sensor key and what turning Touch ID off leaves behind.
omarchy-mac/README.mddescribes the new prompt, andomarchy-mac/versiongoes to 0.2.4.The first commit here removed the prompt and the second restores it in short form, per the maintainer decision below. Against
main, only five files change. The branch name is left over from the first commit.Evidence
Run on an aarch64 Ubuntu host next to unchanged
2d86163, one script at a time:omarchy-mac: 18 of 20 test scripts pass, with 109 checks passing, and both trees match script for script.touchid-consent-test.shandtouchid-activate-test.shstop atsystemd-analyze verify --root, which that host's systemd doesn't support. Every check before that passes, including all the prompt cases.mac-manual/build.pybuilds all 14 pages.The integration suite wasn't run locally. CI is the authoritative run.
Elsewhere
omacom/omarchy's
docs/lifecycle-dispatch.mdstays as written, becauseenable-fingerprintstill explains, asks and turns the reader on only on yes.#69 adds its SDDM lines to the old prompt, so that hunk conflicts with this one. Its note about accounts with no password still has to reach the owner when it rebases.
🤖 Generated with Claude Code