max / alloy
| 1 | #!/bin/sh |
| 2 | # alloy-secret-copy — put something on the clipboard WITHOUT recording it. |
| 3 | # |
| 4 | # The write half of the private clipboard. Alloy ships a password manager |
| 5 | # (gopass) and a clipboard history (cliphist, two `wl-paste --watch` watchers |
| 6 | # in the sway config), and until 2026-08-17 the second recorded the output of |
| 7 | # the first: `gopass show -c` put a password on the clipboard, the watcher |
| 8 | # stored it in cliphist's database on disk, and $mod+Shift+v read it back |
| 9 | # later. gopass clearing the clipboard after its timeout does not help, because |
| 10 | # the copy has already been archived by then. GO alloy problem e7a9e38c. |
| 11 | # |
| 12 | # WHY A SENTINEL RATHER THAN A MIME TYPE. The ecosystem convention is an extra |
| 13 | # offer type, `x-kde-passwordManagerHint: secret`, which KeePassXC emits and |
| 14 | # KDE's Klipper honours. Two things stop it working here: the cliphist in this |
| 15 | # image carries no such string, so there is nothing to honour it, and wl-copy |
| 16 | # offers one type at a time, so a password offered under the hint alone could |
| 17 | # not be pasted into anything that asks for text/plain. Honouring the hint is |
| 18 | # still the right long-term fix and belongs in a clipboard daemon Alloy owns; |
| 19 | # this is the version that works with the tools already installed. |
| 20 | # |
| 21 | # WHY NOT STOP THE WATCHER. Killing `wl-paste --watch` for the duration and |
| 22 | # restarting it afterwards was the other candidate. It fails open: if the |
| 23 | # process that killed it dies before it restarts, clipboard history is dead |
| 24 | # until the next login and nothing says so. The sentinel fails the other way. |
| 25 | # |
| 26 | # THE FAILURE DIRECTION IS THE POINT. A stale sentinel pauses clipboard |
| 27 | # history, which is a feature quietly not working. A missing sentinel writes a |
| 28 | # password to disk. Those are not the same size of mistake, so the mechanism is |
| 29 | # built to fail toward the first: the sentinel is created BEFORE the clipboard |
| 30 | # is written, and a backstop removes it even if nothing ever calls |
| 31 | # alloy-secret-clear. |
| 32 | # |
| 33 | # Reads the secret on stdin so it never appears in argv, which is world |
| 34 | # readable through /proc. |
| 35 | |
| 36 | |
| 37 | |
| 38 | sentinel="/alloy-secret-clipboard" |
| 39 | |
| 40 | # The backstop, in seconds. Longer than gopass's own clipboard timeout (45s by |
| 41 | # default) so a normal copy-then-clear cycle is covered by the clear command, |
| 42 | # and short enough that a crash costs a minute of history rather than a |
| 43 | # session. A second copy while one is live re-arms it rather than stacking. |
| 44 | BACKSTOP=90 |
| 45 | |
| 46 | # Before the write, not after: the watcher fires on the clipboard changing, so |
| 47 | # a sentinel created afterwards would race the thing it exists to prevent. |
| 48 | |
| 49 | |
| 50 | # Detached, and it survives this script exiting, which is the whole job: gopass |
| 51 | # calls this and returns immediately. |
| 52 | ( ; )& |
| 53 | |
| 54 | |
| 55 |