wake_alarm._alarm no longer exists: the PC's waking role was deleted in
kuhyx/wake-alarm, and the phone owns the alarm now. This removes ALARM_MODULE,
_run_alarm() and the --with-alarm flag, leaving one leg -- the workout lock.
Landing this in the same change as the deletion is the point. An orchestrator
left pointing at a module that no longer exists is exactly the failure this
repo has already had once: after screen-locker was extracted on 2026-05-28 the
reference here went stale and the workout lock died with ModuleNotFoundError on
every run for a month, while the alarm half kept working and hid it.
morning-routine.service loses --with-alarm and its comments now name what
actually starts it: wake-alarm-trigger.timer, at the phone's synced alarm time,
not the deleted systemd-sleep hook. Environment=DISPLAY=:0 stays -- the comment
credited it to the alarm's Tk window, but workout-locker.service needs it too.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PcdngP688mW6rJye3LRkSN
The forward-chain fix from 334c0cd9 only ever reached the live ruleset by
hand; /etc/nftables.conf kept the bare default-drop chain, so the next
reboot cut every bridged container off again and the Signal bot went deaf
a second time. Nothing compared generated against persisted.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The GTK tray icon was the only status surface, and it had been commented out
of ~/.config/i3/config since 2026-08-21. With nothing in the bar, the locker
miscounted the week for a fortnight and the machine never said so.
Shows "<count>/<minimum> · <gaming budget>", read from the local status API
(screen-locker-web.service on 127.0.0.1:8770); clicking opens the full UI.
Three things it deliberately will not do quietly:
- a dead status server renders "status server down" in red, or the last known
value marked (stale) in orange -- never a blank block. That server is what
the gaming budget reads, so its absence costs two hours;
- a disarmed or unscheduled timer overrides the compliance colour entirely
and shows "DISARMED" in red. A locker that cannot fire is the worse
problem and the one that hides;
- clicking only opens the UI. There is no bar action that can clear a lock.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fcwu78wk17FWMHbbfbyicc
`write_nftables_ruleset` emitted a forward chain with `policy drop` and no
rules at all. That chain runs at the same netfilter hook as Docker's own
DOCKER-USER/DOCKER-FORWARD chains, so it vetoed them: every bridged
container lost outbound networking while the host itself stayed online, so
nothing looked broken.
It cost the Signal bot three silent hours today. signal-cli lost its
connection to Signal at 15:32:50 and looped on reconnect for 2.5h, while
the bot kept answering /healthz with 200 -- `/v1/groups` still served from
local state, so every surface said healthy.
`flush ruleset` also deletes Docker's own tables, and dockerd only rebuilds
them on its next network event, so restore_docker_rules() now restarts
docker right after a successful apply.
Verified in vmbox (Arch guest, docker installed, alpine probe):
BASELINE=reachable -> OLD=blocked -> NEW=reachable, sshd active throughout.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KbjfWdAyHbCubac7wJKJZW
A new app at ~/restaurant-rater rates restaurant dishes, so it needs a
family icon. shield-cutlery is diet-guard's and the accent is shared
fleet-wide, so the new glyph has to be distinguished by silhouette alone: a
filled star above an open bowl -- a dish, and a verdict on it.
The star is filled rather than stroked for the same reason as the nib and
the anvil: at 72px weight its five interior notches close up well above
MIN_NEGATIVE_SPACE and it renders as a pentagon at 48dp.
The bowl went through two passes. The first used shallow walls stopping at
y=736 and read unmistakably as a smiling mouth with the star for a nose --
two curves with a shape above them is a face unless the vessel is clearly
deeper than it is wide at the base. The committed version drops to y=752
between narrower walls, which reads as a bowl.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012YDbT3ip2yeT18W3KCZdLq
Two failure modes compose cannot see, both of which report a healthy
container while it serves 404s:
1. The Caddyfile is a bind mount, so its contents are not in the config
hash compose compares. Editing it changes nothing observable, compose
says Running, and the new directives never load.
2. A bind mount resolves once, at start. Replacing dist/ rather than
emptying it leaves the container on the old unlinked inode: /srv reads
empty and every request 404s.
Hit both at once on a real deploy today -- the container was Up 6 hours,
Caddy logs clean, /srv empty, curl 404. Reproduced by rm -rf dist (inode
84695030 -> 84695031) and fixed by the patched function in one run.
wait_for_website also stops using curl -sf, which collapses every non-2xx
into the same "did not become reachable". It now reports the status it saw,
because unreachable and answering-404 are different bugs and the message
named only the first.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A7BaCSV8FEtcNgVDVsRAJj
The personal-website repo now prerenders each post to
dist/blog/<slug>/index.html. Caddy try_files does not resolve a directory
to its index on its own, so with the single-term directive every post URL
fell through to the SPA shell: right content once React booted, wrong
<title>, and no link preview. Verified against caddy:2.8 in Docker both
ways before and after.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A7BaCSV8FEtcNgVDVsRAJj
The guard prints `CI_GREEN_SKIP=1 git commit ...` and stops there, but it runs
at two stages: at commit, and again inside the pre-push CI mirror's
clean-worktree `pre-commit` run, which sees the same still-red baseline. So the
documented escape gets you past the commit and into an identical wall at push,
with no hint and nothing in the output connecting the two.
Hit today landing the pre-commit cache-key fix: the commit went through, the
push did not. The variable is read from the environment and `ci_mirror.sh` does
not scrub it, so `CI_GREEN_SKIP=1 git push` carries through to the nested run --
which is now what the message says.
Verified both paths: with the variable set the script warns and exits 0, and
against today's green main it exits 0 silently. shellcheck clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SfnJmxxsZ534AKk16rC2Gr
`requires-python` and the CI pin already said 3.14, but ruff, mypy and pylint
were still grading `meta/` against 3.10 -- three tools judging an interpreter
nothing runs any more, and ruff withholding the pyupgrade fixes that only
become safe at the newer target. `pin_ci_python.py` rewrites these three knobs.
What the stale target was hiding: `datetime.timezone.utc` where `datetime.UTC`
is now correct (UP017), and ruff's stable formatter applying PEP 758 --
`except (A, B):` becomes `except A, B:`. That makes the source 3.14-only at the
syntax level, which is what `requires-python = ">=3.14"` already promised; kuhy
chose it deliberately over holding ruff a release back (2026-08-29).
Also drops a comment in `wsg_grabber/models.py` that the bump makes false: it
explained why the file used ``str, Enum`` instead of ``StrEnum`` under mypy at
3.10, sitting directly above a ``class Verdict(StrEnum)`` that had already
stopped obeying it.
ruff, ruff-format, mypy --strict and pylint all clean at the new target.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SfnJmxxsZ534AKk16rC2Gr
Counting transcript records reports 1.00 tool calls per message no matter
what the model did: the harness writes one JSONL record per tool_use block,
all sharing message.id. A hand-rolled per-record count read the real 1.9%
batching rate as 0.0%, which nearly shipped as a headline finding.
Group by message.id instead, and report it, so the number comes from a
tested code path with the window handled by find_transcripts rather than an
ad-hoc timestamp. Three snapshot fields give next week's run a baseline.
Axes was at the 12-attribute cap, so the counters live in their own Batching
dataclass with the two ratios as properties.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NiSSDu4gKbp9GGxTTtXvYW
track-bars on the family charcoal/accent: rising bars with a plotted trend
point, which reads as odds movement.
linux:false because betting-sim's desktop target is a Chrome-wrapped web build
rather than the GTK embedder, so the hicolor set would go unread.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RDSQS61Vgb6wptLPT2Utbf
A Flutter/Material TextField publishes its label as `hint` and leaves text
and content-desc empty, so every field on such a form arrived as an
anonymous EditText. The workout app's manual-workout form is nine of them:
the only way to fill it was by screen position, which is precisely the
mis-tap-writes-the-wrong-field failure this tool exists to prevent.
`hint` ranks below text and content-desc but above resource-id. Material
clears the hint once content arrives, so a filled field still answers to its
real text and a caller can tell an empty field from a completed one.
The app-side alternative does not work and is documented as a trap: wrapping
the field in Semantics(label:) emits a second labelled node at the same rect,
and MergeSemantics collapses them by dropping the hint from Android's tree,
which leaves the form MORE anonymous than before. Widget tests cannot catch
that -- they assert on Flutter's SemanticsNode while this tool reads
Android's AccessibilityNodeInfo, so nine of them passed green on the build
whose on-device labels had just gone empty.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AWq9MRemX9ZWvNmnhNWUP2
`Pre-commit checks` has been red on main since 2026-08-24: the md-naming
and file-length hooks are thin shims that exec the real gate out of the
utils repo, and a runner has no ~/utils, so both died with "shared gate
not found at /home/runner/utils/...". The dedicated md-naming.yml
already checks utils out into .utils/; this does the same and points
UTILS_ROOT at it, which is the override the shims already support.
CI_GREEN_SKIP is set because this commit IS that fix.
Also registers punchme (github.com/kuhyx/punchme) in the shared icon
family on the family accent, Android-only so linux=False. Its glyph is a
time card with a clipped corner, a column of four punched holes and
three ruled lines; the first draft put two holes above two wide slots
and read unmistakably as a face at every size.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
focus_owner (com.kuhy.focus_owner, the Device Owner enforcer) was the one
Flutter app never registered with the shared generator. It shipped a
hand-written vector adaptive icon added purely to unblock its build — the app
had no icon resource at all and every release AND debug build failed on AAPT
"resource mipmap/ic_launcher not found". That placeholder used #E8EAED, so it
was off-family as well as unregistered.
None of the existing 13 glyphs fits a device-policy enforcer, so this adds
padlock-closed: shackle stroked (inheriting the group's round caps), body
filled in the accent, keyhole punched back out in the field colour. A fully
stroked padlock loses its keyhole at 48dp and a fully filled one loses the
shackle's opening; splitting them keeps both.
Geometry was computed against style.py's constraints rather than eyeballed,
per the lesson recorded on _NOTE_ASCENDER: body 292..732 x 520..792, shackle
centreline r=132 about (512,520) so its outer top is 352 (inside the 232 safe
edge) and its legs land at 380/644, well within the body. The keyhole leaves
60px of body above it and 76px below the stem, both clear of
MIN_NEGATIVE_SPACE=36 — checked by rendering at 48px, where the hole survives.
The repo path points into ~/phone-focus-mode/focus-owner: focus-owner was
absorbed as a subtree on 2026-08-24 and both the standalone clone and its
GitHub repo are gone.
app_icons stays at 100% branch coverage; 44 tests pass.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017KU9czAaz7wAkjyJ5uEB3t
Renames this repo's markdown onto README.md / CLAUDE*.md / DOCS*.md /
TODO*.md so a future session can tell a live task from a finished
record without reading the file to find out.
TODO files carry the literal "REMOVE ME AFTER FINISH" and are deleted
when the work lands. Enforced by the md-naming pre-commit hook and CI
workflow, both delegating to the shared gate in kuhyx/utils so this
repo's idea of a valid name cannot drift from every other repo's.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A47aX5DJwmA3QFn3pkEWi7
The "Depth-cap progress" table still advertised 294 violations and a 207-file
"step D worklist", which stopped being true when the campaign completed. The
gate now reports zero:
$ bash meta/scripts/check_directory_depth.sh --all
check_directory_depth: all checked paths are within 2 directories (exit 0)
A stale worklist is worse than no worklist: it sends the next session hunting
for 207 files that are not there. The table is hand-written --
report_live_state.sh generates the path inventory above it but never emitted
this section -- so regenerating the doc would not have fixed it.
Rewritten to state the guarantee (the directory-depth-cap hook enforces it on
every commit) and to point at the command rather than at a number, with the
original backlog kept as history.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
The previous commit made hosts-blocker's install.sh register the guard-lib
instances itself, which is what finally gives a fresh machine real hosts
enforcement. It also created a class of machine that --rollback refused to
touch: the state dir it demanded, /var/lib/guard-lib-migration, is written by
the one-shot migration, and a machine that never had a legacy layer has no
such state. So --rollback exited 1 while the instances it exists to remove
were live -- every freshly installed machine had no escape hatch at all.
Rollback now proceeds whenever a guard-lib instance is registered, says
plainly that there is no legacy layer to restore, and still refuses when
there is genuinely nothing to roll back.
The second half is worse and was only visible by running it. The existing
re-assertion of chattr +i works by starting hosts-guard.service and
hosts-bind-mount.service, neither of which exists on a fresh machine, so
every systemctl cat skipped and the rollback finished with /etc/hosts holding
no immutable flag and no bind mount -- less protected than before it ran.
It now re-asserts immutability directly when no legacy unit did it.
Verified in a pristine vmbox guest whose guards came from
install_core_system.sh --all: dry-run and real rollback both exit 0, the
three instances are uninstalled, shutdown-schedule (a different feature) is
untouched, and /etc/hosts keeps ----i----------------- and all 178410 lines.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Two of the seven modules invoked python_pkg/ paths that stopped existing when
screen_locker and steam_backlog_enforcer were extracted into their own repos,
so the first thing the installer did on a fresh machine was fail. Nothing
installed guard-lib either, which meant the CORE "Midnight shutdown timer"
module always died with "guardctl not found on PATH".
Every module now resolves through install_module_from_repo, and
require_extracted_repo clones a missing sibling instead of only printing
instructions -- otherwise "clone it yourself" is not actionable on the bare
machine this script exists to set up. The clone drops privileges to the
invoking human via sudo -u, because an installer that re-execs under sudo
would otherwise leave a root-owned checkout in the user's home; a sandbox
cannot catch that, since the guest is root throughout. A clone that fails is
still a loud, fatal module failure, never a silent skip.
guard-lib is addressed as the "utils" repo: it is a subdirectory of
github.com/kuhyx/utils, so asking for "guard-lib" would resolve to
~/guard-lib and derive a clone URL that does not exist. It installs first,
before the modules that need guardctl.
Separately, the security hardening suite's screen_locker section hardcoded
$HOME/testsAndMisc/python_pkg/screen_locker/screen_lock.py. That path went
away with the extraction, so all five checks silently SKIPPED -- reporting a
clean run for tests that never executed, which is the exact failure mode
lib/extracted_repos.sh was written to prevent. It resolves the repo properly
now ($HOME is also wrong under sudo), and the assertions are rewritten against
the current implementation: the four symbols they grepped for no longer exist
anywhere, having been replaced by SPORT_CHOICES, SPORT_TABLE_TENNIS,
_validate_table_tennis and MANUAL_WORKOUT_REFLECTION_MIN_CHARS.
On this host that takes the suite from 18 passed / 1 failed / 5 skipped to
23 passed / 0 skipped. The task file's "18 passed / 6 skipped out of 17"
target was stale -- the suite is 24 checks now -- so it could not be matched
literally.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Two post-push audits, both worth keeping.
vendor/common.sh sources three sibling libs unconditionally and only common.sh
had been vendored, so every entry point printed three "No such file or
directory" errors. Every suite passed anyway -- the functions in use are
defined above that block and a failed sibling source is non-fatal -- which is
exactly why it needed its own test. Fixed in digital-wellbeing c5eb028.
A grep across /etc, /usr/local and /usr/bin, broader than report_live_state.sh
(which reads only /usr/local/{bin,lib}, units and symlinks), found three
references. None from this work: pw_guard.sh's target still exists, and
dwm-session carries two pre-existing single_use/ paths. dwm is dormant and was
deliberately abandoned, so that one is reported rather than changed.
Also records that digital_wellbeing/data/ was safe to delete: never tracked on
any historical path, and no script in the unit references data/ at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
The last unit over the cap is now github.com/kuhyx/digital-wellbeing, cloned
at ~/digital-wellbeing, and `--warn` is gone from the directory-depth-cap
hook. `check_directory_depth.sh --all` exits 0 with zero violations, down from
46 at the start of the session. That was the point of the campaign.
History was filtered across all three paths the unit has lived at (63
commits); the current path alone holds 2.
The six suites that test this code lived in linux_configuration/tests/ rather
than beside it, so they were filtered separately (24 commits) and merged in as
a second root. A path filter of the unit alone would have left them here,
testing code that had left the repo -- and their names would still have been
in shell-tests.yml's hardcoded arrays, failing CI on missing files.
linux_configuration/lib/common.sh is vendored into vendor/ rather than trimmed
to the four functions actually used: set_actual_user_vars calls two of the
others, so a trim breaks silently. It is in vendor/ and not lib/ because
test_music_parallelism.sh copies the unit's own lib/ into a scratch worktree
while stubbing the shared lib; in one directory the real common.sh overwrote
the stub, and the suite silently measured the wrong library.
Running the callers caught what shellcheck and grep both missed:
check_and_enable_services.sh died with "DW_REPO: unbound variable" because the
definition sat below its first use. It now runs clean and reports ok for every
extracted-unit check.
Two long-standing live defects fixed as a consequence. The pacman-wrapper
drift manifest embeds ABSOLUTE paths and had named deleted scripts/ paths
since the flatten, so the check reported a permanent false "stale or
tampered"; it now verifies, with unchanged hashes proving the extracted bytes
are identical. And the last two stale live references are gone: 8 -> 0.
The two pre-commit excludes for setup_midnight_shutdown.sh are dropped rather
than moved. lsattr shows nothing in the unit is chattr +i any more, and the
fixer hooks leave the file alone when allowed to run.
Removing 80 shell files also shrank the jscpd denominator enough that the
virtualenv/tqdm clones inside the GENERATED .ci-mirror-venv/ tipped the repo
to 2.09% and blocked this commit, which adds no duplication. That directory is
the same vendored third-party shell the ignore list already excludes as
.venv/, just under another name; jscpd walks the working tree rather than the
index, so being untracked did not spare it. Excluded by name: 1.74%.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
161 tracked files, 3.2M. All 158 episodic entries have retrieval_count 0 --
the store was written to and never once read from. Last write 2026-05-28, the
tool is uninstalled, and nothing references it: the only code hit for "hippo"
anywhere in the tree is the blocked domain "wordhippo" in
leechblock_defaults.json.
No .gitignore entry was needed after all. The handoff anticipated one for
hippo.db, which was untracked only because *.db is ignored -- but with the
whole directory gone there is nothing left to ignore.
Tracked files: 1346 -> 1175.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Five live mechanisms were silently pointing at paths that no longer exist.
None of them announced it; each degraded to a log line or a missing file.
- hosts-file-monitor.service is ACTIVE and watches /etc/hosts for tampering,
but its embedded installer path was the pre-flatten scripts/ one. It logged
"Hosts install script not found" 37 times: it detected tampering it could
not repair.
- dns-blocklist-refresh.service (enabled, next run 00:01) named a deleted
single_use/features/ path, as did media-organizer.service under utils/.
- setup_periodic_system.sh read all TEN of its templates from
periodic_background/system-maintenance/, which has held nothing but
untracked .pyc and .coverage residue since system-maintenance was
extracted. Every install of a service, timer or script was reading a file
that was not there.
- /etc/sudoers.d/hosts-install-no-passwd granted NOPASSWD on the old hosts
installer path.
Running the fixed dns unit then caught a bug the static checks missed: under
systemd there is no SUDO_USER and `id -un` is root, so the new resolver
returned /root/hosts-blocker and the service died with "Feed generator not
found". extracted_repo_home() now handles the bare-root case by falling back
to the owner of the lib file itself. Verified across all four invocation
modes: as the user, via sudo, as root with SUDO_USER unset, and with an
explicit *_DIR override.
The service now reports "Feed refreshed and dnsmasq reloaded (189038 lines)".
/etc/hosts is untouched throughout -- still 176888 lines, still chattr +i,
still bind-mounted by its guard.
Live artifacts were patched via validated temp files, never in-place: the
sudoers drop-in was checked with `visudo -cf` before install, and every
original is backed up under /var/tmp/*.pre-{extract,fix}.
Stale live references: 8 -> 2. The 2 that remain name digital_wellbeing
paths and can only be fixed when that unit is extracted.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
linux_configuration/periodic_background/hosts/ is now
github.com/kuhyx/hosts-blocker, cloned at ~/hosts-blocker. Clears 5 of the 46
directory-depth violations; the remaining 41 are all digital_wellbeing, so
--warn stays on the depth hook until that unit moves too.
History was filtered across ALL FOUR paths the unit has lived at (hosts/,
linux_configuration/hosts/, linux_configuration/scripts/periodic_background/
hosts/, and the current one). Filtering on the current path alone would have
preserved 1 commit out of 51 -- the flatten moved it recently, so a naive
single --path silently truncates the history to the flatten commit and still
looks fine in `git log | head`.
No source line changed: all 15 files are byte-identical to the copies removed
here. install.sh already resolved its lib/ via BASH_SOURCE.
The six callers now resolve the checkout through a new shared helper,
linux_configuration/lib/extracted_repos.sh, which reads the invoking user's
home from SUDO_USER's passwd entry rather than $HOME -- several of these
scripts re-exec under sudo, where $HOME is /root. A missing checkout fails
loudly with a clone instruction; it is never a silent skip, which is the
failure mode that left screen-locker's self-repair dead for five months after
the 2026-05-15 reorg.
The live blocker is untouched: /etc/hosts keeps its chattr +i, and the guards
load their plugins from /usr/local/share/guard-lib-plugins/ (byte-identical by
md5), not from a repo path.
Verified: 25 tests pass from ~/hosts-blocker; shellcheck clean on the helper
and all six callers; sudo resolution prints /home/kuhy/hosts-blocker;
test_setup_dns_blocker.sh passes; report_live_state.sh --check is byte-identical
before and after, so no new stale references.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Shell tests went red on my last push. test_network_diagnostics.bats sourced
its scripts through $REPO_DIR/scripts/diagnostics, a level the flatten
removed, so setup() failed and all 30 tests errored.
Latent rather than new: the previous run was green only because ci-mirror
aborted before the bats stage. Fixing the concurrency OOM let the gate reach
this suite for the first time since the flatten.
Grepped the tree for the removed scripts/ and single_use/ levels; this was
the last one in tracked code.
Also rewrites the two handoff docs around the units that remain.
CI_GREEN_SKIP=1 because this commit is the fix for the red baseline.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
The push kept failing with "1798 passed" immediately followed by "Killed".
The suite was never the problem: it peaks at ~0.29 GiB against its own 2G
cap. ci_mirror.sh runs pre-commit --all-files and the pytest gate
concurrently, and the pre-commit tree peaks at ~1.75 GiB (mypy and pylint
repo-wide, invisible to /usr/bin/time because it measures only the parent).
The pair wants ~2 GiB at once; the pytest scope has MemorySwapMax=0, so on
a loaded machine it is the one that dies.
Concurrency is now conditional on MemAvailable. Above 4 GiB the fast path
is unchanged. The probe lives in ci_mirror_mem.sh, following the existing
ci_mirror_range/ci_mirror_shell split, which also keeps ci_mirror.sh under
the line cap.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
The pre-push gate was failing on a layout that is actually correct. The
script sources ../../lib/common.sh, which is right for where it sits after
the flattening; the test's guard still hardcoded ../lib/common.sh from
before the move, so it reported drift against itself.
The prefix is now read out of the script under test, so the two cannot
desynchronise the next time the tree moves. Verified non-vacuous: rewriting
the script to ../../../lib/common.sh in a throwaway worktree still fails
the guard, and it now names the real prefix.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
A gate that blocks a commit with no tooling to clear it just moves the
problem to whoever hits it. --prune stages the removal of the oldest
artifacts, keeping the newest 20 per directory plus the template, and
leaves the commit to the operator.
Running it caught a bug that reading it would not: sorting on %cs (date
only) ties every artifact committed the same day, so sort fell back to the
path and the prune deleted alphabetically - it picked one of today's
newest files instead of the oldest. Sorts on %ct with -rn now.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
The prune commit's evidence noted that "a recurring prune or a hard cap
would be the durable fix if this recurs" - which is advice, not a fix. The
tree would have regrown at one artifact per commit and been noticed months
later, exactly as it was this time.
check_repo_size.sh now caps evidence/ and contracts/ at 60 files each: the
headroom the prune bought, against 21 today. main accumulates failures
rather than returning on the first, so tripping both the source ceiling and
the artifact cap reports both.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
These accumulate roughly one per commit and were the largest single driver
of tracked-file growth: 695 artifacts, 487 of them from 2026-08 alone.
Kept the 20 most recent of each plus both templates. A count cap rather
than a time window, because the growth is bursty - a 7-day window would
still have kept 393 files and bounded nothing. Selection is by git commit
date, not by filename: the corpus mixes -2026-05, -20260604 and undated
slugs, so a filename filter would have mis-sorted ~48 files.
Safe to delete: nothing reads the historical corpus. Every reference in
meta/scripts is either template.json or validation of files staged in the
current commit, and the three hooks still pass. sessions/ is append-only
and untouched.
Tracked files 1993 -> 1340.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
The depth cap had 104 violations. 57 are structural: whole projects with
their own layout (C/, dwm/, artgate/), a tool split into modules to satisfy
the line cap, prose trees, and directories whose name an external format
dictates (systemd, hooks, workflows, patches, fixtures). Those are exempted
the same way lib/ and tests/ already were.
The other 47 all sit under periodic_background/{hosts,digital_wellbeing},
which can only come down by extracting those units - blocked on a manual
guard repoint. So the hook runs --warn: the number stays visible on every
commit without blocking unrelated work. Enforcement is one flag away.
Also guards the argument order: `--all --warn` used to ignore --warn and
exit 1, which as a hook entry would have blocked every commit.
check_repo_size.sh is the companion gate - depth stops one subtree growing
a taxonomy, this stops the repo absorbing everything. It counts tracked
files (git ls-files, never find: the tree holds ~30k untracked extraction
leftovers) minus vendored and hook-generated trees: 1049 against a 1200
ceiling, printed every run so the threshold is auditable.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
check_file_length.sh required a dot in the filename, so every
extensionless executable was silently uncapped. is_source_file() now
reads the first two bytes and accepts a shebang, which catches real
scripts while still skipping data files like i3/config.
That immediately surfaced boot_recovery/boot-repair at 756 lines --
three times the largest capped shell file in the repo, uncapped purely
because its name has no dot in it. Split along its own section markers
into lib/{discovery,repair_fs,repair_kernel,modes}.sh, leaving the entry
script at 218.
The report block stays inline on purpose. It is top-level code with exit
calls rather than function definitions, and sourcing it changed control
flow: 30 of 42 tests failed. That is recorded in the lib headers so the
next reader does not retry it.
No vmbox pass for this one, deliberately. boot-repair is UEFI/ESP tooling
and the guest is BIOS/GRUB, so a sandbox green would be actively
misleading -- the vmbox skill says so itself. The evidence is the fixture
suite, which drives the real script against synthetic root trees: 42
passed before, 42 passed after.
verify_shell_split.sh reports 31 of 32 functions byte-identical. The one
difference is clean_shadow_files, where quoting $ESP_MOUNTPOINT fixed a
genuine SC2086 that only became visible once the function left the
monolith -- shellcheck could no longer see the readonly declaration and
stopped assuming the expansion was safe.
lib/tests/run_all.sh delegates to that same suite rather than standing up
a second, weaker harness: these libraries need the entry script's strict
mode and globals, so testing them in isolation would test a fiction.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
i3 config was filed under periodic_background, which it never was. It is
now linux_configuration/i3/ (config + installer) and
linux_configuration/i3blocks/ (the status scripts), both at depth 2.
misc/testsAndMisc-bash/ was a directory named after the repo it already
lived in, so its contents moved up to misc/. Its stray .gitignore came
along, keeping the media rules working from the new location; its
.vscode/tasks.json did not, because it invoked Bash/transcribe.sh, a path
this repo dropped long ago.
Most of this commit is the references the previous flatten missed. Its
regex skipped anything with a variable prefix, which left 16 files --
including check_and_enable_services.sh and install_core_system.sh, the
orchestrators -- still naming $CONFIG_DIR/scripts/periodic_background/...
It had also rewritten three i3 paths into scripts/i3blocks/, a directory
that has never existed; the i3blocks suites caught that by failing.
13 more symlinks broke on the i3 move. Repointed, and cpu_monitor/volume/
disk were run again to confirm the bar still reports.
Depth 139 -> 104. Roughly 47 of the remainder sit in digital_wellbeing
and hosts, still deferred. The rest encode real structure -- a C project,
Python package directories, artgate's routes/styles, systemd unit dirs --
so flattening them further would destroy information rather than
taxonomy.
Verified: 1798 tests at 100%, both i3blocks suites pass, the shell
coverage gate is clean, and the relocated media rules still ignore what
they should while test_fw.srt stays tracked.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Two levels of pure taxonomy carrying no information, and a 2-directory
cap leaves no room for them. Categories now sit directly under
linux_configuration/: features, fixes, utils, misc, fresh-install,
installers, diagnostics, gaming, boot_recovery, periodic_background, lib,
meta, tests. A script's own lib/ and tests/ are exempt from the cap, so
those stay put.
Depth violations 248 -> 139; roughly 40 of the remainder are inside
digital_wellbeing and hosts, deferred until the guard-lib repoint.
The moves are the easy half. What breaks is every path that counted
levels, and three classes of it had to be found by resolving paths rather
than reading diffs:
- 52 relative `source` lines for lib/common.sh and lib/android.sh. The
correct prefix differs per file: 40 were ../../ and 12 were ../../../,
so a blanket replace would have fixed one set and broken the other.
- REPO_ROOT in 7 fixes/lib/tests harnesses and setup_syncyomi_guard.sh,
each walking up four levels from a directory now two levels shallower.
This caused a real regression -- the fixes suite went red with
"log_warn: command not found" -- caught by running it and comparing
against the pre-move baseline worktree.
- conftest.py's sys.path entries, which took the pytest gate down with
14 collection errors, and the `# shellcheck source=` directives, which
took shellcheck down with SC1091.
16 symlinks in $HOME broke, including the entire i3blocks bar. All
repointed, and cpu_monitor/memory/disk/time/volume were run to confirm
they emit real output rather than merely resolving.
Verified: 1798 tests pass at 100% coverage, 8 of 9 lib/tests suites pass
(the ninth needs the coverage jail, identically at baseline), pre-commit
is clean, and the two untracked secret configs are still ignored after
their gitignore paths changed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
github.com/kuhyx/phone-focus-mode
The two halves go as one repo: focus_policy parses phone_focus_mode's
config.sh into the typed policy that focus-owner consumes, and nothing
else references it. 77 commits of history, plus the two Python suites
from linux_configuration/tests/ and the six mutation fixtures from
meta/scripts/fixtures/, all of which named phone_focus_mode paths.
Running the suites -- not reading them -- turned up a real defect the
extraction would otherwise have inherited. config.sh sources the
untracked config_secrets.sh, and ctl_libs_harness.sh never pinned
HOME_LAT/HOME_LON, so the distance assertions were checking against
whatever real coordinates that file happened to hold. The suite passed on
this machine and could not run anywhere else, and a real home location
was load-bearing in a pass/fail condition. The harness now pins a pair
4300 m south of the coordinate its own case seeds, and config.sh sources
the secrets file only when it exists. 136/136 pass with no secrets file
present.
Before making the repo public I scanned the whole history for the known
home coordinates and for any committed config_secrets.sh: zero hits, and
the file was never tracked. The new repo also gained a .gitignore, since
none came across in the filter-repo pass.
test_magisk_service.sh hangs. It hangs identically in the pre-extraction
baseline worktree and appears in no CI list here, so the new repo's CI
skips it with that reason rather than pretending it passes.
check_python_location.sh loses its phone_focus_mode exception now that
the directory is gone; every tracked .py still passes it. The .gitignore
rules for config_secrets.sh and the keystore STAY -- the working copy is
still on this machine and still holds a real location.
Depth 262 -> 248.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
test_usage_monitoring_installer_efficiency.sh left with
system-maintenance, but shell-tests.yml keeps TWO hand-maintained lists
and I had only cleaned the Arch-specific one. The side-effect-free list
still named it, so ci-mirror rejected every push since that extraction.
Cross-checked the whole workflow this time: every test_*.sh it names now
exists on disk.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
github.com/kuhyx/android-guardian, cloned to ~/android-guardian
periodic_background/utils/ held two unrelated things: the Magisk module
and heavy_job_lock.sh. Only the module was a unit, so the directory is
split rather than moved.
heavy_job_lock.sh goes to digital_wellbeing/, NOT to ~/utils. Measured
rather than assumed: of its consumers only install_pacman_wrapper.sh
reads the repo copy, while pacman_wrapper.sh, makepkg_capped.sh and
meta/scripts/run_with_heavy_lock.sh all read the installed
/usr/local/bin copy. A library belongs with whoever installs it, and
exporting it would have added a cross-repo edge for one caller.
Fixed a bug that predates the extraction: update_android_hosts.sh set
GUARDIAN_MODULE_DIR to "$SCRIPT_DIR/android_guardian", a sibling that
never existed at that path -- packaging could not have worked from this
checkout at all. It now reads ${ANDROID_GUARDIAN_DIR:-~/android-guardian}
and require_guardian_module() gates build_module(), the single funnel
every packaging path goes through, so a missing checkout names the clone
command instead of failing inside a cp four helpers deep.
Verified both directions: with no checkout the guard fires with
instructions; with one, the module builds -- 182331 blocked domains, a
1.1 MB zip. The subsequent adb push failed on device-side storage, which
this commit does not touch.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
github.com/kuhyx/system-maintenance, cloned to ~/system-maintenance
First of the four live subsystems, and the one with no cross-unit source
edges, so it establishes the pattern for the harder three. Passed in
vmbox first -- the installer ran from an extracted-layout copy in the
guest and detected its distro without any monorepo path -- then on the
host.
Its 21 Python and shell test files lived in linux_configuration/tests/,
reaching bin/ through a shared conftest. Twenty moved with the code and
got their own conftest; the shared one stays, since it serves several
script directories that are not going anywhere.
test_shutdown_timer_monitor.sh stays here. It is a pair test: it checks
that digital_wellbeing's installer template and the monitor it installs
still agree, so it belongs to whichever half owns the installer. It now
reads the monitor from ~/system-maintenance and SKIPs when that checkout
is absent.
An intermediate version of that test fell back to /usr/local/bin, which
HUNG -- the installed copy is stale and differs from source, so the test
was comparing an installer template against a binary it never generated.
Falling back to it would have been worse than skipping, so the fallback
is gone and the comment says why.
Also removed test_hosts_file_monitor.sh from shell-tests.yml's explicit
list: it left with the subsystem, and CI would have failed on the missing
file rather than quietly skipping it.
Verified: 136 tests pass in the new repo, 202 still pass here, the pair
test passes against the checkout and skips without it, and both the shell
coverage gate and shellcheck are clean. Depth 294 -> 267.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Found while surveying system-maintenance: /usr/local/bin holds copies
that hardcode repo paths, and no section of the inventory looked there.
The installed copy is what actually runs, so grepping the repo for its
own paths never finds them.
Three were invisible until now:
periodic-system-maintenance.sh 3 paths into digital_wellbeing + hosts
hosts-file-monitor.sh 1 path into hosts
wsg-grabber PYTHONPATH into the repo root
The first runs hourly, which means the two subsystem extractions coming
next have to reinstall it -- editing the repo copy alone would leave the
live one pointing at a directory that no longer exists.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
github.com/kuhyx/mtk-root
Unlike the app extractions, mtk_root could not travel alone: its four
libraries and three of its four test suites sat in scripts/lib/, one
directory up from the stages that source them. Taking only the directory
would have orphaned eight files here and left the new repo unable to
source anything, so all twelve moved in one filter-repo pass. The
mtk_root/ + lib/ sibling layout is preserved, so every
`source "$SCRIPT_DIR/../lib/..."` resolves with no call site edited.
Two things the new repo needed that the monorepo provided implicitly:
lib/tests/run_all.sh -- the copy here is shared with the non-mtk suite
and stays. Without a new one, shell-tests.yml's
glob over */lib/tests/run_all.sh would have
stopped covering these suites on a green build.
.shellcheckrc -- external-sources + source-path=SCRIPTDIR, which
is what lets shellcheck run with no severity
filter and still pass on the ../lib/ sources.
mtk_harness.sh stays here despite the name: test_common_datetime.sh
sources it. Removing it broke that suite outright, which running the
suite caught immediately. Its header now describes what it actually
serves; the misleading name is kept on purpose, since renaming would
churn its one consumer for nothing.
Verified: 84 tests pass in the new repo with shellcheck clean, 55 still
pass here, and check_shell_coverage.sh --all exits 0 with the four
mtk_root allowlist entries gone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Three self-contained apps move to their own public repos with full
history (8, 14 and 30 commits, each matching the monorepo count):
github.com/kuhyx/reverse-survivors
github.com/kuhyx/kcd2-dice-solver (cloned to ~/kcd2-dice-solver)
github.com/kuhyx/focus-owner
Each gained a CI workflow. Only reverse_survivors had one here; the other
two were never gated by anything, which is part of why they no longer
belong in a repo whose CI is shared.
What stayed: setup_kcd2_dice_solver.sh, which deploys the solver rather
than being part of it. It now builds from ~/kcd2-dice-solver, and the
post-commit hook it generates drops its `git diff -- kcd2_dice_solver/`
filter -- in a standalone repo every commit is solver code, so the old
filter could never match again.
focus_owner's two references to python_pkg.focus_policy now name the
phone-focus-mode repo, where that generator is headed.
Running the live kcd2 unit -- rather than reading it -- found three real
breakages that no amount of grepping would have: pnpm ignoring the
package.json "pnpm" field, pnpm refusing to purge node_modules without a
TTY, and the Docker container still bind-mounting the deleted monorepo
path. All three are fixed; the unit exits 0 and
https://dice.kuhy.duckdns.org/ serves the freshly built bundle.
Depth violations 417 -> 303.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
The coupled case the previous archive commit deferred. billsplit is not
just a directory: a coverage package exists only to gate it, app_icons
holds a registry entry pointing into it, a CI workflow builds it, and
.binary-allowlist / .gitignore carry a paired set of rules for its
launcher icons. Removing the directory alone would have failed the 100%
coverage gate and left three configs naming a missing path.
So all of it goes in one commit:
billsplit/ 73 files
python_pkg/billsplit_coverage/ 5 files, gated billsplit only
.github/workflows/billsplit-ci.yml
the app_icons AppIcon(key="billsplit") entry
the .binary-allowlist patterns and their .gitignore ! overrides
the .prettierignore fixture line
billsplit/ and billsplit_coverage/ were extracted in a single filter-repo
pass (8 commits, matching the monorepo) and pushed to
github.com/kuhyx/testsAndMisc-archive, where the coverage package joins
the archive's existing python_pkg/ tree. Presence was confirmed through
the GitHub API before anything was deleted here.
.binary-allowlist is now empty of patterns -- billsplit's icons were its
only entries. The policy header stays and records that no exception is in
force, and check_no_binaries.sh still exits 0 on a comment-only file.
Verified: 2095 tests pass at 100% coverage, down from 2114 by exactly the
19 billsplit_coverage tests. Depth violations 451 -> 417.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Two self-contained apps, neither referenced by any other code, systemd
unit or symlink -- which is why they go first in the restructure.
Full history preserved via git filter-repo and merged into
github.com/kuhyx/testsAndMisc-archive at their original repo-relative
paths, matching how C/, CPP/, horatio/ and the rest already sit there.
Archived commit counts match the monorepo exactly: bucket_catch 10,
poker-stakes 2.
poker-stakes-ci.yml goes with the app it builds, so no workflow is left
pointing at a missing path. bucket_catch had no workflow.
Verified before deleting: file counts match (64 and 65), diff -r against
the working tree is clean, and the GitHub API lists both units in the
archive. Depth violations drop 556 -> 451.
billsplit is deliberately NOT part of this commit: it is coupled to
python_pkg/billsplit_coverage and an app_icons registry entry that the
100% coverage gate requires be deleted in the same commit as the app.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Two scripts and a baseline record, ahead of the archive/extract/flatten
campaign. Nothing moves in this commit.
report_live_state.sh enumerates everything on this machine that names an
absolute path inside the repo -- 6 systemd units, 2 /etc/guard-lib plugin
targets and 19 symlinks. Moving a subtree silently breaks these, so each
later step repoints its own references and re-runs this with --check.
check_directory_depth.sh is the 2-directory cap, modelled on
check_file_length.sh. It is deliberately NOT wired into pre-commit yet:
the tree has 556 violations, so enforcing it now would block every
unrelated commit -- the same reason file-length-cap landed after its
campaign rather than before.
Run as a dry run it sizes the work: 349 of those 556 disappear with the
planned removals, leaving 207 to flatten, ~130 of them under
scripts/single_use.
Exemptions are prefix-based (third_party/, docs/superpowers/,
.github/skills/) plus the lib/ and tests/ path segments, matched as whole
segments so libvirt/ and lib.sh are not caught. Capping lib/ and tests/
would put this gate in conflict with the 250-line cap and the
shell-coverage ratchet, which require exactly that layout.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
The run that added the hook-environment cache could not benefit from it:
its key is a hash of .pre-commit-config.yaml, which that same commit
changed, so it logged "Cache not found" and then saved. Six hook
environments were still built from scratch -- including pylint's, which
pulls a dozen extra packages -- and that is most of the 126s hook step.
Without a dispatch there is no way to re-run the job on an unchanged tree
to see the cached number, so the saving could only be asserted, not
measured. Now it can be measured.
Confirmed in the same log that the previous commit landed:
"pytest with coverage enforcement....Skipped".
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Per-step timings showed the real cost was not the install I had just
cached:
Pre-commit checks Run pre-commit hooks 152s
Install dependencies 55s
The log explains it -- "pytest with coverage enforcement....Passed"
inside the Pre-commit checks job. `pre-commit run --all-files` fires the
pytest-coverage hook, which runs every python_pkg suite with
--cov-fail-under=100: precisely what the Python tests workflow had
already done on the same commit. Two full 2174-test runs per push.
SKIP=pytest-coverage in that job only. It is the one hook skipped, and it
is skipped for being DUPLICATED rather than for being slow -- noted in
the workflow so it gets dropped if Python tests ever goes away.
Also checked and deliberately left alone:
session-autopsy-hook-mutations costs ~50s locally, but self-skips when
~/.claude/hooks is absent, which is always true on a runner. It was never
a CI cost.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
python-tests.yml runs pytest and nothing else, but installed the full
requirements.txt -- 79 packages, of which ~62 are linters, formatters and
type checkers that job never invokes. Installing them was the workflow's
slowest step: 49s, against 35s actually running the tests.
meta/requirements-test.txt holds the 17 that matter. Derived by static
import analysis over python_pkg/ and linux_configuration/tests/, then
VERIFIED rather than assumed -- a clean Python 3.11 venv with only this
list runs the whole suite:
2174 passed, TOTAL 100.00%, 7138 stmts / 1748 branches, 0 missing
identical to the full-requirements run. Install of the subset: 6.2s.
The pytest-* entries are in the list even though nothing imports them:
randomly reorders tests, timeout enforces a limit, sugar changes
reporting and xdist provides -n. Dropping one changes how the suite runs.
Drift is gated, not left to discipline: a new requirements-test-sync hook
reuses check_requirements_sync.py's existing --requirements flag, so a
new third-party import fails at commit time rather than as a bare
ModuleNotFoundError in CI. The new file is also in both path filters.
pre-commit.yml keeps the full list -- it needs every formatter.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Measured per-step timings, rather than guessing where the time went:
Pre-commit checks Install dependencies 63s (next slowest: 16s)
Python tests Install dependencies 49s (pytest itself: 35s)
Both workflows re-downloaded ~80 packages from PyPI on every run. The
Node workflows already cache npm; the Python ones were the outliers.
* cache: pip on actions/setup-python in both, keyed on
meta/requirements.txt -- the REAL file, not the root symlink, since
the key is hashed from the file at the given path and hashing a
symlink is not hashing its target. (That same symlink already cost
us a silently-skipped workflow earlier today.)
* ~/.cache/pre-commit via actions/cache@v4, keyed on
.pre-commit-config.yaml. pre-commit builds a virtualenv per hook repo
on first use, which costs more than the pip install itself, and it
only needs rebuilding when a hook's repo or rev changes.
The first run after this lands is a cache miss and will be no faster;
the saving starts from the second.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
The previous commit fixed the dependency resolution failure that had kept
Python tests red since ~2026-08-19 -- and then did not run it, because
the workflow's path filter lists "requirements.txt" while the root file
is a SYMLINK into meta/. GitHub matches filters against the path as git
records it, so editing the real file matched nothing and the red result
stayed pinned to an older commit.
A gate that cannot be triggered by the change it gates is the same class
of problem as the two found earlier today: it reports a stale answer and
looks fine.
Both spellings are now listed, in the push and pull_request filters.
The other root symlinks -- pyproject.toml, run.sh, lint_python.sh,
.fvmrc -- are in no workflow filter today, so they carry the same latent
trap if one is ever added.
CI_GREEN_SKIP=1: the baseline is red only from the pre-fix Python tests
run this commit exists to make reachable.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Python tests has been red since ~2026-08-19, failing in "Install
dependencies" with pip's resolution-too-deep. That is an install failure,
not a test failure -- which means no Python test has actually run in CI
for days, on a repo whose Python bar is 100% branch coverage.
The CI log names the cause: the resolver backtracks through Brotli
1.0.6 -> 1.0.4 -> 1.0.1 and asgiref 3.8.0 -> 3.7.1 -> 3.6.0, all dragged
in by mitmproxy>=10.0, then gives up. Nothing imports it: `git grep
mitmproxy` finds it only in a pacman whitelist and a docs file, and
check_requirements_sync.py passes without it.
Reproduced and fixed locally rather than guessed, on Python 3.11.11 with
pip 26.2.1 -- the versions CI uses:
pip install --dry-run -r <with mitmproxy> exit 1, resolution-too-deep
pip install --dry-run -r <without mitmproxy> exit 0
Left standing: 78 of the 80 requirements are unpinned >= ranges, so this
can recur the next time a package with a wide dependency graph is added.
CI_GREEN_SKIP=1: this commit IS the fix for the red baseline.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
check_shell_coverage.sh has existed for a while and was never in
.pre-commit-config.yaml, so nothing ran it: a new shell library with no
suite beside it could land freely. Wiring it was unaffordable while the
gate measured a jail-run 100% bar (--all took over 600s). After 578c4a6c
dropped that, --all is 0.31s and the per-file mode this hook uses is
milliseconds, so it goes in.
Also fixes a bug 578c4a6c introduced: the failure path still interpolated
${COVERAGE_BAR}, which that commit deleted, so every rejection printed
"COVERAGE_BAR: unbound variable" and advised a 100% line-coverage bar
that no longer exists. The message now describes what is enforced.
Three libs blocked the wiring -- meta/scripts/lib/shell_coverage_{bar,
case_runner,jail_setup}.sh, the code that BUILDS the jail, and so the one
corner of the repo the jail cannot measure itself. They are pure
file-writers, so they get a suite that calls them against a temp $JAIL
and asserts on what was written: 48 assertions, nothing mounted, no
unshare, no kcov.
Several assertions check that an emitted file contains an UNEXPANDED
${BASH_SOURCE} or $1 -- the whole point of the quoted heredocs the jail
relies on. Written as single-quoted literals that is SC2016 on every
line, so the needles compose the dollar from an exported constant
instead. No suppressions were added.
check_shell_coverage.sh --all: 3 BLOCK -> 0 BLOCK
hook, repo-wide: 0.16s
new suite: 48 assertions, green
Proved the ratchet still bites: an untested lib dropped into
mtk_root/lib is refused with "needs: .../tests/run_all.sh". Probe removed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
bca70b67 (2026-08-22) turned the shell ratchet from "a run_all.sh exists
beside this lib" into a measured 100% line-coverage bar, run through the
kcov+userns jail. It works -- dot_resolver_install.sh really does measure
39/39 -- but the price is one hand-written test file per lib: fixes/lib
took 34 test files for 22 libs, the jail runs serially at 37-90s a pass
(concurrent jails give false failures), and most libs need a production
seam added before they can be driven at all.
That put the remaining 84 exempt libs at several days of work, for a
repo of personal automation scripts. User's call today: not worth it.
python_pkg keeps its 100% branch bar, which is already met -- 2101 tests,
100.00%, zero missing statements or branches. The difference is the
instrument: pytest-cov is fast, reliable and parallel; the shell jail is
none of those.
Also tried and rejected, recorded in the file: a middle option where the
suite must MENTION the lib by name, closing the empty-suite hole the
100% bar was reaching for. It newly BLOCKED 68 libs sitting in
directories that already have suites -- 38 in utils/lib alone -- because
those passed the old presence check and so were never allowlisted, and
seed_allowlist refuses to grow by design. Tightening a ratchet with no
way to grandfather the existing tree turns it into a wall.
The jail is not deleted; it is still the tool for measuring a lib on
purpose, and ci_mirror.sh still runs jailed suites.
check_shell_coverage.sh --all: >600s -> 0.31s
BLOCK entries: 3, all pre-existing meta/scripts/lib
coverage-tooling libs
allowlist: unchanged at 84 (--seed correctly
refused to add the 3)
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
Endurain API keys are upload-only by design: the scope allow-list is
frozenset({"activities:upload"}), enforced in both validate_api_key_scopes
and the create schema, and the activities read route is mounted with
validate_access_token rather than validate_access_token_or_api_key. A key
therefore fails at authentication before any scope check, so the HTTP
dedupe check added in 325d5a8b could never succeed and every pass fell
back to the ledger alone. Verified against v0.19.0; upstream documents no
plan to widen the allow-list.
Read the start times out of postgres instead. The container publishes no
host port, so this goes through docker exec -- no new dependency and no
port opened just to read four rows. The HTTP path is still tried first so
the supported route resumes working on its own if the allow-list ever
grows.
None and [] are kept strictly apart: a dead container, timeout or non-zero
exit means "could not ask" and returns to the ledger, while an empty result
set is a real answer. Conflating them would re-upload everything on any
docker hiccup, which is the bug the dedupe check exists to prevent.
Proven with an empty ledger: the known-duplicate TCX was skipped and the
activity count held at 4, and the same file WAS imported with
ENDURAIN_NO_DB=1, so the check is what blocks it. The duplicate that
negative control created was removed afterwards.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GbYEnfGnf1988cdFdGebot
`git push` took ~130s because ci-mirror re-ran `pre-commit --all-files`
plus every lib/tests suite on every push, regardless of what changed.
The larger problem found while measuring it: ci_mirror.sh invoked
pytest_changed_packages.py with no arguments, and that script returns 0
on empty argv because it is normally a pre-commit hook fed staged paths.
The pre-push python stage has therefore been running ZERO tests and
reporting success. Behind it sat a genuinely failing test, fixed in
ff3ea99a. The stage is now passed the changed paths and actually runs.
Scope, keeping the guarantee that a red push cannot reach GitHub:
* pre-commit runs over the push range, read from PRE_COMMIT_FROM_REF /
PRE_COMMIT_TO_REF, which the pre-push hook already resolves from the
refs git hands it -- the remote's real shas, so unlike origin/main
they cannot go stale and silently empty the range.
* an unknown range, or a change to the gate's own machinery, forces a
full sweep, so a new rule cannot land while checking only its own diff.
* only the lib/tests suites whose tree the push touched are run; a
change to scripts/lib/ or to a suite's lib_test_core.sh fans out to
all of them.
* pre-commit and pytest run concurrently. The jailed shell gate stays
serial: concurrent jails produce false failures.
* an unchanged tree short-circuits on a recorded tree hash.
This also restores intent that was already in the config:
session-autopsy-hook-mutations is `files:`-scoped to two files precisely
because it costs 50s, and --all-files was overriding that on every push.
Measured, against a ~130s baseline:
docs-only push 7.6s
python push 25.3s (includes a 22s stage that never ran)
one shell suite 43.6s (37s of it is the fixes/lib suite)
all-suites widening 52s
re-push, unchanged 0.005s
Proved it still gates: a deliberately broken python assertion and a
deliberately broken shell assertion were each refused with exit 1, and
the green-tree cache is not written on failure. Both probes reverted.
Split into ci_mirror_range.sh and ci_mirror_shell.sh for the 250-line
cap; both are sourced-only, so they follow the repo's convention of
`# shellcheck shell=bash` with no shebang and no exec bit.
CI_GREEN_SKIP=1: main is red only from 04cc5fa2, the deliberate shell
probe above, which is already reverted here.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
test_passes_store_and_repo_root asserted
env["PYTHONPATH"].endswith("testsAndMisc"), which is a property of the
checkout's directory name rather than of the code. It fails in any
worktree, which is where ci-mirror runs its gates.
Assert against the value reindex.py actually computes,
Path(reindex.__file__).resolve().parents[2]. Exact equality is a
stronger check than the suffix match it replaces.
This was invisible until now because ci-mirror invoked
pytest_changed_packages.py with no arguments, and that script returns 0
on empty argv -- so the pre-push pytest stage never ran. The next commit
fixes the invocation.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PJrqKERAo5QUeEHWS5Asgp
RunnerUp can now upload to Endurain directly as well as through the WebDAV
inbox, and the two paths cannot see each other: the same run landed twice
(activities 14 and 15, identical start time, distances differing only by
tcx/gpx rounding). The ledger only knows what this importer handled, so it
could not catch it.
Add a third dedupe check that asks Endurain what it already holds, queried
once per run. Endurain exposes no query-by-start-time endpoint, so a page of
recent activities is fetched and compared client-side, with a +/-120s
tolerance: the file for activity 15 embeds 21:51:04Z while the server stored
21:51:05Z, so exact equality misses the very duplicate that motivated this.
A failed or unauthorised query returns None, never an empty list, so it can
never be read as "nothing matches" and upload on the strength of it.
Export filenames are parsed as LOCAL wall-clock time. RunnerUp_2026-08-14-
00-58-45_Running is 2026-08-13 22:58:45 UTC; reading it as UTC would move a
late-evening run onto the wrong day and defeat the comparison.
Also fix the adb fallback, which broke the moment a second device attached:
phone_attached() returned true if ANY device was in the 'device' state, then
every bare adb call failed with "more than one device/emulator" as a warning
on a path nobody watches. The serial is now resolved once and threaded into
every call as -s. Ambiguity is settled by configuration, never by guessing --
several devices and no ENDURAIN_ADB_SERIAL means the fallback is skipped.
The remote check stays inert until the API key is granted activities:read;
until then it falls back to the ledger, which is the previous behaviour.
85 tests, 100% branch coverage on all four modules, no suppressions. The
oversized test modules are split by concern to stay under the 250-line cap.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GbYEnfGnf1988cdFdGebot
The deletion did not survive the seam-restore commit, so the exemption
was still in place and the gate was passing on it rather than on
coverage. Measured 76/76 = 100.00% with the gate's own jail flags, in a
detached worktree -- the main tree aborts the jail right now because
another agent is editing python_pkg/endurain_import concurrently and the
tree fingerprint reads that as a write escaping the jail.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Three changes, none of which touch what a single test asserts:
thorium_repairs.sh's `sleep 4` becomes `sleep ${THORIUM_STARTUP_WAIT:-4}`
and the harness sets 0.2s. Three cases were each paying four real
seconds to watch a stubbed browser stay up; the stub sleeps 30s, so the
`kill -0` check is unchanged. That file: 17.0s -> 5.6s.
_t_hide was the rest of it. It walks PATH and creates a symlink per
executable -- ~13.5k here, 4.6s a build -- and the suite has ~16
distinct tool sets. Now the full farm is built ONCE and `cp -al`
hard-links it per set, with only the hidden names deleted from the copy.
The cache also moved out of $TEST_TMPDIR, which is per-file, into a
run-scoped dir run_all.sh creates and removes, so it survives between
files.
110s -> 98s (sleep) -> 85s (shared cache) -> 37s (hard-linked farm).
645 OK, 0 FAIL throughout. The cases that only pass when a tool is
genuinely unfindable -- install_pyroveil's missing-dependency arms,
fix_aggressive without sqlite3 -- still pass, which is what proves the
farm is still hiding rather than quietly returning a full PATH.
lib_test_core.sh hit exactly 250 lines, so _t_hide and the farm move to
lib_test_path.sh (194 + 67).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Committed unformatted in 558883f2, which makes the pre-push prettier
hook fail for anyone pushing afterwards. Formatting only: 9 lines
reflowed, no content touched.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The seams were applied, verified and measured, but never reached the
commit: cb1e8c53 shipped the file with its hardcoded /etc/X11,
/etc/profile and /home/$SUDO_USER paths. ci-mirror caught it by running
the suite from a clean checkout, where the tests read the REAL system
files -- the RenderAccel assertion failed against this machine's actual
xorg config, which says "disabled", and the pyroveil clone path came
back as /home/testuser rather than the tmpdir.
The local run passed only because a stale seamed copy was sitting in the
measurement worktree. Re-applied and grepped back immediately, which is
the documented rule for exactly this failure mode.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
76/76 lines through run_all.sh at --min 100, exit 0. 45 assertions over
15 cases across all four functions.
XORG_CONF and XORG_CONF_D are plain globals ASSIGNED inside
configure_xorg, so the default had to move to the assignment itself --
exporting a value from the harness alone would have been overwritten on
entry. PROFILE_FILE and USER_HOME take the ordinary `${VAR:-default}`
form. The real /etc/X11/xorg.conf.d/20-nvidia.conf keeps its Jun 2025
mtime and ~/pyroveil does not exist after the suite runs.
configure_xorg's RenderAccel value is asserted to be "true" AND asserted
never to be "false": the lib's own comment records that "false" forces
software rendering and pins Xorg above 30% CPU, so that is a regression
worth a dedicated negative assertion.
install_pyroveil's prompt could not be asserted on its text -- bash
writes a `read -p` prompt straight to the terminal, so it never reaches
captured output. The branches are distinguished by the non-interactive
arm's "Auto-installing" line instead, which is observable.
bt_audio.sh re-measured at 127/127 with the new files present.
That is all eight libs from the original handoff: arch_sysctl (banked as
unmeasurable and quarantined), arch_cpu, arch_hardware, thorium_repairs,
hosts_guard_migrate, hosts_guard_rollback, ubuntu_perf_fixes and
nvidia_config. Allowlist 91 -> 84.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Stands up Endurain behind the existing Caddy edge and imports RunnerUp
activities into it. Endurain is a VIEWER: screen-locker's workout gate still
reads the phone over adb and is untouched here, because a WebDAV drop folder is
spoofable by anything that can write to ~/cloud and must not become a
screen-unlock authority.
Primary path is phone -> WebDAV (existing dufs) -> ~/cloud/RunnerUp/ ->
importer -> Endurain, with an adb pull as fallback when a run has not been
uploaded yet. A systemd user timer runs the importer every 15 min.
Findings that shaped the implementation, each verified against the running
system rather than assumed:
- Endurain moved to Codeberg; the GitHub repo is a read-only mirror stale at
v0.17.7 against v0.19.0. The image is pulled from Codeberg.
- postgres:18 moved PGDATA to /var/lib/postgresql/18/docker. Upstream's compose
example still mounts the pre-18 path, which fails as an unexplained
"container is unhealthy"; the real cause is only visible in docker logs.
- Endurain seeds a hardcoded admin/admin via an Alembic migration with no env
override, and endurain.kuhy.duckdns.org already resolved publicly. Publishing
is therefore a separate script that refuses to install the Caddy site while
those credentials still work.
- Endurain performs NO duplicate detection, so the importer owns dedupe on two
axes: a content hash, and an activity key that collapses the .gpx/.tcx pair
RunnerUp writes for a single run (different bytes, same run).
- Ambiguous uploads (5xx/timeout) are never re-queued, since the activity may
already have been committed; only definite 4xx rejections are staged for
retry. The importer exits non-zero so the timer surfaces the problem.
- RunnerUp has no background sync at all, so every upload is a manual tap.
- Its WebDAV synchronizer needs HTTPS with a real cert (Android blocks
cleartext at targetSdk 36; self-signed is broken upstream) and a URL with no
trailing slash, and cannot MKCOL against dufs - so the inbox must pre-exist.
The phone gets a dufs credential scoped to /RunnerUp, verified unable to read
Keepass or gdrive_backups.
Verified: 52 tests at 100% branch coverage; the timer imports 4 real runs and
re-runs create zero duplicates; https://endurain.kuhy.duckdns.org serves 200
with a valid cert and the other edge sites are unaffected; screen-locker's
suite still passes 1090 tests at 100% coverage.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GbYEnfGnf1988cdFdGebot
81/81 lines through run_all.sh at --min 100, exit 0, and
check_shell_coverage.sh passes with the allowlist line already deleted.
59 assertions over 14 cases across all four functions: the sysctl
drop-in with both the "already applied" skip and the documented-default
fallback for an unreadable sysctl; both NVIDIA branches, since the
helper-service path and the standalone fallback write different units
and different undo lines; earlyoom install / already-installed /
config-backup; and fix_failed_sssd for none, all five and exactly one
failed unit -- that last one matters because the detection loop breaks
at the first failure while the masking loop re-checks each unit, so a
naive implementation would mask all five.
The four write targets are now behind overrides defaulting to the real
paths. The add_undo lines deliberately KEEP their literal /etc paths and
are asserted to: the undo script runs on the real system, so seaming
those would have written an undo that removes nothing. Neither
/etc/sysctl.d/99-performance-tuning.conf nor /etc/default/earlyoom
exists after the suite runs.
bt_audio.sh re-measured at 127/127 with the new files present -- that
canary runs after every addition to this suite now, because a test file
silently broke it once already.
The handoff is rewritten for the one remaining lib.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
hosts_guard_migrate.sh 81/81 and hosts_guard_rollback.sh 65/65, both
100.00% through run_all.sh at --min 100, and check_shell_coverage.sh
passes with each allowlist line already deleted.
82 assertions over 36 cases in three files sharing one harness: all four
validate_requirements refusals, plugin installation, the mount-collapse
loop in every arm (clears, fails, hits its 20-iteration bound, dry-run),
per-instance migration for both the bind-mounted and the plugin flavour,
and do_migrate's refusal to retire the legacy pacman hooks when nothing
registered -- the arm that exists so a failed migration cannot leave the
files with no unlock hook at all.
INSTRUMENT FIX: rollback measured 97.01% with lines 69 and 108
uncovered, both `done <redirect>` closing a loop. Reproduced minimally:
in a three-iteration loop ending `done >"$1"`, the body reports hits=3
and the `done` line hits=0, because bash attributes the redirected
compound statement to the loop's OPENING line. That is the same defect
_CLOSING_BRACE_OP already handles for `} |`, so _LOOP_END_OP is its
twin. The redirect is required by the pattern: a bare `done` stays in
the denominator, so genuinely uncovered loops are still caught.
Audited as the instrument notes demand -- every newly excluded line
printed with its text -- and it is exactly those two lines across all of
fixes/lib. Acceptance pair unchanged and exact: dot_resolver_install.sh
39/39, rpi_nc_ca.sh 73/73.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
75/75 lines through run_all.sh at --min 100, exit 0, and
check_shell_coverage.sh passes with the allowlist line already deleted
so the pass cannot come from the exemption.
55 assertions over 23 cases: all five profile repairs plus test_thorium,
including every dry-run arm, the partial singleton-lock case, a corrupt
database and the sqlite3-absent path. The lib needed NO production
change -- it already takes every path from $THORIUM_CONFIG_DIR, so
pointing that one variable at the tmpdir confines it completely.
The gate's per-case timeout moves from the jail's 120s default to 600s.
That default is a hang-catcher, not a performance budget, and this suite
has outgrown it: 76s bare at 25 test files, 93s at 27, and kcov roughly
doubles it. The symptom was a lib FAILING the gate while measuring
perfectly -- bt_audio.sh printed "127/127 lines = 100.00%" and then
"warn: case timed out after 120s", which --fail-on-case-error turned
into a non-zero exit. bt_audio.sh and arch_cpu.sh both pass now.
The cost, recorded in the evidence: a genuine hang takes 10 minutes per
pass to surface instead of 2. test_thorium's `read -r -p` is exactly
that shape, so every case here feeds it stdin explicitly.
This also settles the quarantine in b2779059. Re-adding the arch_sysctl
cases still yields "kcov instrumented no lines" for bt_audio at a 600s
timeout with no timeout warning, so runtime was never that file's
problem and the quarantine stands for the reason its header gives.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
check_shell_coverage.sh was RED on main. bt_audio.sh is off the
allowlist at 127/127, and commit 7c14cfca -- which added
test_arch_sysctl.sh -- made kcov stop registering it entirely, so the
gate failed on a lib nothing had touched.
Bisected by measuring bt_audio.sh at three commits: a2b23521 127/127,
afe6ac3a 127/127, 7c14cfca "kcov instrumented no lines". Removing that
one test file restores 127/127 with every other new file still in the
suite (arch_cpu, arch_hardware x2, thorium x2), and arch_cpu.sh
re-measures at 57/57 alongside it.
Both plausible causes inside lib_test_core.sh were ruled out by
reverting each on its own in a throwaway worktree: `export out` and the
_t_hide farm cache each left bt_audio broken while the file was
present. Neutering the file's body while keeping the source line does
NOT reproduce it either, so the trigger is cumulative rather than one
construct, and it is not understood.
The file keeps all 23 assertions and still passes when run directly; it
just leaves the test_*.sh glob run_all.sh iterates. That costs nothing
measurable, because arch_sysctl.sh has never measured through the jail
in the first place -- it reports "instrumented no lines" on its own,
like nc_php.sh -- and it stays on the allowlist.
The header records the bisect and what to re-verify before anyone
un-quarantines it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
clean_shadow_files() assumed UEFI with an ESP mounted at /boot. On a
BIOS/GRUB system with no ESP, /boot is a plain directory on the root
filesystem holding the only kernel the machine has -- and neither existing
guard catches it: findmnt /boot is empty (GUARD 1 passes) and there is no
EFI/ or loader/ (GUARD 2 passes). boot-repair therefore read the real
vmlinuz-linux and initramfs-linux.img as "orphaned kernel file(s)
shadowing the ESP" and deleted them, leaving the machine unbootable.
Measured 2026-08-22 in a vmbox BIOS/GRUB sandbox, not theorised: after
sudo boot-repair the guest never booted again.
GUARD 3: if /sys/firmware/efi does not exist the system booted via BIOS,
so there is no ESP and nothing in /boot can be a shadow. warn + return 1,
matching the existing guards' tone and exit convention. Overridable with
the existing --esp DEVICE -- naming the ESP by hand asserts the firmware
assumption the guard otherwise checks -- rather than adding a new flag.
The check goes through rpath(), so a --root fixture declares its own
firmware type and the case is testable without a VM. On the live UEFI
machine it resolves to the real /sys path, so behaviour there is unchanged.
Audit of the other BIOS-unsafe paths (none destructive, left alone):
* mount_esp(): resolve_esp_device fails with no /boot fstab entry, so it
reports and does not mount. Non-destructive.
* sync_kernel_image()/rebuild_initramfs(): gated behind boot_is_mounted,
which is false on BIOS, so they never write.
* check_fstab_options(): reports "no /boot entry" -- noise on BIOS, but
it writes nothing.
* run_preflight(): aborts every kernel upgrade on a BIOS box. Fail-closed
and non-destructive, so out of scope here.
Tests: three cases in tests/lib/boot_repair_bios_cases.sh (new file; the
existing case lib is at the 250-line cap) -- the guard refuses and explains,
--esp overrides it, and a UEFI fixture still cleans shadows as before.
Verified failing without the guard (38/4) and passing with it (42/0).
Sandbox: BIOS guest keeps both kernel files, says why, and still boots
afterwards (uname -r 7.1.9-arch1-2 after a clean poweroff). shellcheck clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
com.metrolist.music was on the day allowlist but missing from the night
list, so it was hidden between 23:00 and 05:00. Add it to
night_allowed_packages in focus_owner's policy asset (the list the
installed Device Owner app actually enforces) and mirror it into
phone_focus_mode's NIGHT_WHITELIST so the two sources do not drift.
policy.json ships as a bundled Flutter asset, so the edit is inert until
the APK is rebuilt: deployed via phone_deploy.sh (build 4,
data-preserving install -r, Device Owner intact).
Verified on the phone at 23:1x device time with the curfew live:
hidden=true before, hidden=false after, and Metrolist launches and holds
the foreground across several enforcer ticks.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01U7Wzqp2ub6gW5uVBH5qmhn
94/94 lines through run_all.sh at --min 100, exit 0. 42 assertions over
26 cases across all five functions, including both boot-loader flavours
and the no-boot-loader path.
PRODUCTION BUG, found by writing the tests and confirmed against this
host: tweak_journal's regex required a space before the unit, but
journalctl prints "305.5M in the file system" with no separator. The
size check never matched, so the vacuum was unreachable and a
multi-gigabyte journal would never have been trimmed. This is the third
instance of the same defect -- arch_perf_report.sh:36,65 carried it too
-- so the fix is identical: the separator is now optional. One case
asserts the spaced form still matches, making this a widening.
The mitigation tweak edits the boot loader, so /boot/loader/entries,
/etc/default/grub and grub.cfg are now behind overrides defaulting to
the real paths. This machine's boot configuration is never touched.
_t_hide gets a cache: it symlinks every executable on PATH (13.5k here)
and was rebuilding that farm on every call, which took the suite to 92s
and pushed a jail case past its 120s timeout. Now 77s, and arch_cpu.sh
re-measures at 57/57 so the shared change cost nothing.
The cases are split across two files because a single one was 307 lines
and file-length-cap rejected the first commit attempt.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
57/57 lines, measured through run_all.sh and confirmed a second time at
--min 100 so a shortfall would exit non-zero. check_shell_coverage.sh
passes with the exemption removed.
28 assertions over 14 cases: no cpufreq at all, every core already
tuned, the mixed case where only some cores lag, cpupower absent /
unconfigured / already configured, and all three I/O scheduler targets
(none for NVMe, mq-deadline for SATA SSD, bfq for spinning disks).
The sysfs reads and the two /etc write targets are now behind overrides
defaulting to the real paths, so the tests can present hardware this
host does not have -- a rotational disk -- without touching anything.
One assertion initially passed for the wrong reason: _t_hide keeps
$FAKE_BIN on PATH by design, so the harness's own cpupower stub still
satisfied has_cmd and the "not installed" branch never ran. Unstubbing
before hiding is now explicit, and the case fails if that is lost.
This also settles arch_sysctl.sh's diagnosis: it uses this same
harness, so the harness is not why it cannot be measured.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
23 assertions over 8 cases covering both tweak_vm_sysctl and
tweak_network_sysctl: the skip path, the write-when-the-drop-in-is-
missing path that a naive "nothing to do" check gets wrong, the
value-differs path, an unreadable-sysctl path, and the tcp_bbr guard.
arch_desktop_harness.sh is the third family harness, supplying the
globals and apply_tweak that live in optimize_arch_desktop.sh rather
than in any lib. arch_cpu.sh and arch_hardware.sh will reuse it.
SYSCTL_DROPIN_DIR defaults to the real /etc/sysctl.d, matching the
seam ubuntu_perf_more.sh and arch_perf_fixes.sh already carry; the
suite runs un-jailed in ci_mirror.sh and CI, so a bind mount would
protect nothing. Both real drop-ins keep their Mar 2 mtimes.
_t_run lands in lib_test_core.sh because `out="$(func)"` forks a
subshell and kcov never registers a lib whose first execution happens
inside one -- measured as 0 lines via $( ) against 54 lines via a
direct call.
arch_sysctl.sh STAYS on the coverage allowlist: it still reports "kcov
instrumented no lines" through the jail, while ubuntu_perf_more.sh
measures 32/32 in the same worktree and the same invocation. The tests
are real and green; the coverage number is not yet obtainable, and the
exemption may only come off once it is.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
shfmt runs on every staged shell file and rewrites drifted libs, which
moves coverage denominators and invalidates measurements taken minutes
earlier. Landing the reformat first makes every number measured from
here the number that survives pre-commit.
thorium_repairs.sh converts from 2-space indent to tabs; the other two
only lose runs of blank lines. `git diff -w` shows no token changed,
shellcheck is clean, and the 20-file suite stays green.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Fourth flavour of defect (a) found this session, after multi-line quotes,
`} |` and the backslash-continued assignment. bash reports a multi-line array
assignment at its OPENING line and never emits the element lines or the
closing paren, so kcov instrumenting them inflates the denominator:
+L3 items=('one' 'two') # opened line 3, closed line 6
# lines 4, 5, 6: never emitted
The rule is therefore symmetric with the existing multi-line-quote case: the
opening line is the statement, everything up to the closing paren is data.
_closes_array() finds that paren by recomputing quote state per candidate, so
an element like "a(b)" does not end the literal early.
Measured effect through fixes/lib/tests/run_all.sh:
arch_perf_fixes.sh 93.48% -> 100.00% (43/43)
bt_audio.sh 100.00% -> 100.00% (127/127)
arch_perf_report.sh 100.00% -> 100.00% (59/59)
arch_perf_probes.sh 100.00% -> 100.00% (93/93)
ubuntu_perf_more.sh 100.00% -> 100.00% (32/32)
bt_report.sh 100.00% -> 100.00% (37/37)
bt_pairing.sh 100.00% -> 100.00% (65/65)
One gained, six held, none went DOWN. Acceptance test unchanged:
dot_resolver_install.sh 39/39 and rpi_nc_ca.sh 73/73. arch_perf_fixes.sh
therefore comes off the allowlist in the same commit, confirmed by the real
gate with no exemption.
The fail-open check, since over-detection INFLATES coverage: the old and new
predicates were diffed over every allowlisted lib plus the six cleared ones,
printing each newly excluded line WITH its text. All 91 across 10 libs are
array elements or closing parens -- common.sh's tool tables, packages.sh's
associative arrays, arch_sysctl.sh's sysctl maps, thorium_repairs.sh's four
path lists, ubuntu_perf_fixes.sh's sssd units, and so on. No statement caught.
Tests cover both directions, including three negatives: a single-line
`x=(a b)`, an element containing a parenthesis, and `x=$(echo hi)` which is a
command substitution rather than an array. Module stays at 100% branch
coverage.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Eighth of the 16 untested fixes/lib libs. 22 assertions; 43/46 = 93.48%. The
lib STAYS on the allowlist -- see the artifact below.
Third instance of the same dead branch. Line 16 matched on
`([0-9]+\.?[0-9]*)\ G`, a required space before the unit, while journalctl
prints "4.2G" with no separator, so the vacuum could never run. Fixed to an
optional space, as in arch_perf_report.sh.
This one differs from that earlier fix and the difference was confirmed with
the user before landing: arch_perf_report.sh's vacuum sits behind
--apply-safe-fixes, but THIS one runs whenever fix_arch_performance.sh runs.
So the fix makes an unflagged destructive action live -- the script will now
really vacuum the journal to 300M once it passes a gigabyte. That is the
function's evident intent, since its next act is to write a permanent 300M
cap, and the user approved it explicitly.
Three absolute paths moved behind overrides defaulting to the real ones:
JOURNALD_CONF_DIR, SYSTEMD_UNIT_DIR (fix_media_organizer writes
/etc/systemd/system/media-organizer.service) and ORGANIZE_SCRIPT_CANDIDATES.
Verified afterwards that the real unit file still carries its 2026-07-25
mtime and the journald drop-in its 2026-03-02 one, i.e. a bare un-jailed
run_all.sh touched neither.
The three uncovered lines (68-70) are the element lines of a multi-line array
literal. `PS4` + `set -x` on a minimal case shows bash reporting the whole
assignment at its OPENING line and never emitting the elements or the closing
paren, so they are data, not statements -- a fourth flavour of defect (a),
after multi-line quotes, `} |` and the backslash-continued assignment. The
array is provably used: the loop on the next line iterates it and the
media-organizer assertions pass.
Worth fixing, unlike the bt_adapter one-liner: the form appears in three of
the remaining untested libs (thorium_repairs.sh has four) and in nine
allowlisted libs repo-wide. That lands as its own commit next.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Seventh of the 16 untested fixes/lib libs and the first of the ubuntu-perf
pair, which gets its own harness here. 24 assertions; 32/32 = 100.00%
through run_all.sh on the first measurement, confirmed at --min 100 (exit 0),
then cleared with the real gate and no exemption.
Two absolute paths had to move behind env overrides, defaulting to the real
ones: fix_journal writes /etc/systemd/journald.conf.d/size-limit.conf and
run_undo globs /root for undo scripts. Binding those in the jail was rejected
for the same reason as bt_pairing.sh's udev rule -- run_all.sh runs UN-jailed
in ci_mirror.sh and in CI, so a bind mount protects the coverage run and
nothing else. JOURNALD_CONF_DIR and UNDO_DIR follow the UDEV_RULES_DIR
precedent.
Proof the suite is safe where it runs un-jailed: after a bare run_all.sh the
real /etc/systemd/journald.conf.d/size-limit.conf still carries its 2026-03-02
mtime, a genuine artifact of a past production run, and no
undo_ubuntu_performance_*.sh exists in /root.
Worth noting for the rest of the family: run_undo ends in `exit`, so its cases
run inside an explicit subshell. A command substitution already provides one,
so the exit ends only that subshell rather than the whole test file.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Sixth of the 16 untested fixes/lib libs, and the first with no dead branch to
fix -- the whole job was test-writing. 43 assertions across two files;
93/93 = 100.00% through run_all.sh, twice, the second at --min 100 (exit 0),
then cleared with the real gate confirming and no exemption.
The one genuine gap was check_memory_pressure's high-swap branch (lines
158-159), which fires only when swap is heavily used WHILE RAM is still
available -- a state this host does not happen to be in. The lib reads the
absolute /proc/meminfo, but every read goes through `awk`, so stubbing awk
supplies controlled figures without touching /proc: the stub answers on the
pattern being matched and execs the real awk for anything else. Both
directions are covered, including RAM genuinely exhausted, which must NOT be
flagged because it is not the stutter signature.
Everything else here is branch coverage of probes whose inputs are external
commands: failed systemd units, ACPI error counts, a CPU-hungry Xorg,
governors all-performance vs one-powersave vs unsupported, thermal
throttling excerpts, the TLP/power-profiles-daemon conflict in all three
combinations, and storage health with and without fstrim/smartctl.
lib_test_core.sh gains _t_stub_cat. "Print this fixture file" is by far the
commonest stub shape, and writing it inline forces every call site to embed
a literal ${LIB_TEST_DEV} that must not expand at write time -- SC2016 on
every use. Passing the fixture NAME as data removes the class outright, and
two earlier bluetooth test files are simplified onto it. All 21 test files
and three harnesses now lint clean with no suppression anywhere.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Both were already tested to their reachable ceiling earlier in this session
(a5b6663c, ffcf99f1) and held below 100% only by the closing-brace artifact
that d989cd7e removed. With the instrument corrected they measure:
bt_audio.sh 127/127 = 100.00%
arch_perf_report.sh 59/59 = 100.00%
both at --min 100 with exit 0, and check_shell_coverage.sh returns
"every checked shell library meets the coverage bar" with their allowlist
lines already gone, i.e. with no exemption. bt_report.sh and bt_pairing.sh
were gated in the same run to confirm the two libs cleared earlier did not
regress.
Note that the closing-brace exclusion is now load-bearing: reverting
d989cd7e without also restoring these two lines would turn every subsequent
commit red.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
continuation_lines() excluded the continuation lines of multi-line QUOTED
arguments but nothing else, so a line holding only a group's closing brace
and an operator -- `} |`, `} >>"$REPORT_FILE"` -- stayed in the denominator
while bash attributed no statement to it. That capped four fixes/lib libs
below 100% with no test able to move them.
Why the anchor makes this safe: bash attributes every command INSIDE a group
to its own line, and a pipe's right-hand side to its own line, leaving the
brace-plus-operator line with nothing. Anchoring the pattern on a LEADING
`}` therefore excludes only lines that can hold no statement. A one-liner
`{ echo a; } >file` begins with `{` and is untouched; a bare `}` closing a
function has no operator and is untouched.
Measured effect, all through fixes/lib/tests/run_all.sh:
bt_audio.sh 96.97% -> 100.00% (127/127)
arch_perf_report.sh 98.33% -> 100.00% (59/59)
bt_report.sh 100.00% -> 100.00% (37/37)
bt_pairing.sh 100.00% -> 100.00% (65/65)
bt_adapter.sh 98.55% -> 98.55% (68/69, capped by a different flavour)
Two gained, three held, none went DOWN -- the handoff's regression signal.
That is not luck: _measure computes covered_set = (executed|traced) &
denominator, so for a lib already at 100% the covered set IS the denominator
and removing a subset cannot lower it. The five gated pacman_hook_stall libs
are safe by the same arithmetic.
Acceptance test from the handoff passes unchanged: dot_resolver_install.sh
39/39 and rpi_nc_ca.sh 73/73, both 100.00%.
The fail-open check that actually matters, since over-detection INFLATES
coverage and could promote an untested lib: the new predicate was run over
every lib in the allowlist plus the cleared ones, printing each newly
excluded line WITH its text. All 17 across 10 libs are genuine
brace-plus-operator lines; no ordinary statement was caught.
Tests cover both directions, including the two negative cases (a bare `}`
and a single-line `{ ...; } >file`), keeping the module at 100% branch
coverage. The quote scanner moved into _quote_after() to stay under the
C901 complexity limit rather than suppressing it.
Deliberately NOT fixed here: the backslash-continued ASSIGNMENT flavour that
caps bt_adapter.sh line 97. bash attributes those to the LAST line of the run
but a backslash-continued plain COMMAND to its FIRST, so that predicate needs
a command-vs-assignment distinction. One line in one lib; not worth the
landmine today.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Fifth of the 16 untested fixes/lib libs and the first of the arch-performance
group, which gets its own harness here. 38 assertions across two files. The
lib STAYS on the allowlist at 59/60 = 98.33%, measured twice.
The second dead branch this campaign has found, fixed here. Lines 36 and 65
matched on:
([0-9]+\.?[0-9]*)\ (G|M)
a REQUIRED space before the unit. journalctl actually prints "Archived and
active journals take up 305.5M in the file system." with no separator --
verified against the real binary on this machine. So check_journal_size could
never report a large journal and apply_safe_fixes could never vacuum one:
both were unreachable. The space is now optional, and tests cover the spaced
and unspaced forms so neither can rot.
NOTE: this makes a previously-dead DESTRUCTIVE action live. With
--apply-safe-fixes and a journal over a gigabyte, the entry script will now
really run `journalctl --vacuum-size=300M`. That is the flag's documented
purpose and it stays behind the flag, but it has never actually executed
before today.
Line 20, the one uncovered line, is `} >>"$REPORT_FILE"` -- the closing brace
of a redirected group. A minimal probe writes both inner lines to the file
while bash attributes no statement to the brace, and the report assertions
independently prove the group runs, since they read the text it writes. Same
defect (a) class as bt_audio's `} |`.
That form now caps four libs: bt_audio.sh, arch_perf_report.sh, and
arch_perf_probes.sh and nvidia_config.sh which are still untested. So the
continuation_lines() fix is worth landing after all -- as its own commit,
next, so a denominator change can be reverted without taking these tests
with it.
_t_hide and _t_full_path move from bt_harness.sh into lib_test_core.sh: they
are generic PATH plumbing and the arch harness needs them for the
"nvidia-smi is not installed" branch. All twelve existing suites re-run green
after the move.
One trap worth recording: apply_safe_fixes appends to the ACTIONS array, so
it must be called directly rather than inside "$(...)" -- a command
substitution runs it in a subshell where the appends die with it, and the
assertions read an empty array.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Fourth of the 16 untested fixes/lib libs, the largest at 214 lines, and the
last of the bluetooth family. 47 assertions across three files. The lib
STAYS on the allowlist: 128/132 = 96.97%, twice with the same line set.
The bug the tests found, fixed here. Line 21 read:
if timeout 3 _run_as_user wpctl status &>/dev/null; then
`timeout` is a binary and `_run_as_user` is a shell function that is never
exported, so this always failed with rc=127 -- "timeout: failed to run
command '_run_as_user'". check_pipewire_health therefore reported PipeWire
hung on EVERY run and restarted a healthy audio stack every time
fix_bluetooth.sh ran. The healthy branch was unreachable, which is why the
test for it could not be written without first noticing. Now
`_run_as_user timeout 3 wpctl status`: same bound, inside the sudo'd command.
Why 96.97% and not 100%. An earlier run of this suite read 99/99 = 100.00%,
and that number was wrong -- it measured the file before pre-commit's
`shfmt -w` reformatted it. shfmt rewrites the `{ ...; } \` + `| bluetoothctl`
one-liners into multi-line blocks ending in `} |`, which raises the
denominator 99 -> 132. Confirmed unavoidable by running shfmt against HEAD's
own copy: it rewrites there too, so reverting the reformat is not stable.
The four uncovered lines (90, 113, 137, 157) are each that `} |` -- the pipe
operator joining a block to bluetoothctl, not a statement. They do execute:
a `{ echo a; echo b; } | consumer` block writes every marker to a file even
though bash's xtrace never reports the left-hand side, because a pipe's LHS
runs in a subshell whose xtrace is not captured. kcov instruments them and
cannot attribute them; defect (a) in docs/kcov-under-report.md, third flavour.
Scope, measured: the continuation-pipe form appears ONLY in bt_adapter.sh
(pre-format) and bt_audio.sh (post-shfmt). None of the remaining twelve
untested libs carries either, so this does not block the campaign and the
detector fix stays off the critical path.
Also worth 90.91% -> 96.97% on its own: the bluetoothctl stubs now drain
stdin before exiting. Real bluetoothctl consumes its whole input; a stub
that read one line and exited closed the pipe early, so the writing block
took SIGPIPE partway through and its later lines never ran.
NOTE for whoever edits this file next: bt_audio.sh is now exactly 250 lines,
at the cap. The next change to it needs a split, not a comment trim.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Third of the 16 untested fixes/lib libs, and the first that does NOT come
off the allowlist. 68/69 = 98.55%, up from 45/69. The remaining line is
not a missing test:
PS4='+TRACE:${BASH_SOURCE##*/}:${LINENO} ' bash -x
shows bash attributing the whole multi-line dbus-send assignment in
_services_resolved to line 100, and never emitting line 97 at all. Line 97
is the OPENING line of a backslash-continued statement; kcov instruments
it, bash never attributes a statement to it. That is defect (a) in
docs/kcov-under-report.md -- a wrong denominator -- and no test can move it.
continuation_lines() in shell_coverage_lines.py already excludes such lines,
but only detects them by tracking QUOTE state, so a line ending in an
unquoted backslash slips through. Fixing that is deliberately left out of
this commit: it changes the denominator for every measured lib, and if it
regresses one of the libs now gated with no exemption, reverting a mixed
commit would throw away these tests too.
Scope of the problem, measured rather than guessed -- backslash
continuations are confined to the bluetooth family: bt_audio.sh 10,
bt_adapter.sh 7, bt_pairing.sh 3, pacman_hook_stall_summary.sh 2, and ZERO
in the other 12 remaining libs. So this caps bt_audio.sh next and then stops
mattering.
Also replaces the harness's _t_only_path with _t_hide. _t_only_path dropped
PATH to $FAKE_BIN alone, which hid grep, head and sort as well as the tool
under test -- so the "usbreset is not installed" case died in a cascade of
"command not found" and passed for the wrong reason instead of exercising
the fallback. _t_hide rebuilds PATH from symlinks to every real executable
except the named ones; verified directly that command -v usbreset is then
empty while grep and head still resolve.
bt_report.sh (37/37) and bt_pairing.sh (65/65) re-measured at --min 100,
both exit 0, so neither cleared lib regressed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Second of the 16 untested fixes/lib libs. 65/65 = 100.00% through
run_all.sh, twice, the second at --min 100 (exit 0), then cleared from the
allowlist and confirmed by the real gate with no exemption.
Two findings worth keeping:
- udevadm was NOT in the harness's stub list, so `udevadm control
--reload-rules` was reaching the REAL binary during the suite. A
prepended stub dir cannot hide a tool the host actually has. Caught by
asserting on the recorded call rather than by reading the code, which is
the argument for asserting on calls at all.
- _disable_usb_autosuspend writes /etc/udev/rules.d/50-bluetooth-no-
autosuspend.rules. Binding /etc in the jail would have covered it, and
was rejected: run_all.sh runs UN-jailed in ci_mirror.sh and in the
shell-tests workflow, as its own header says, so a bind mount protects
the coverage run and leaves every bare run writing to a real root-owned
file. Instead the lib gained a UDEV_RULES_DIR override, defaulting to the
real path -- the same env-override seam pacman_hook_stall_harness.sh uses
for PACMAN_LOG/CACHE_DIR/PACMAN_LOCK.
Proof the suite is safe where it runs un-jailed: after a bare run_all.sh,
the real rule file still has its 2026-02-21 mtime (it is a genuine artifact
of a production run six months ago) and the probed power/control values are
unchanged. The jail cannot give that check.
fix_usb_autosuspend globs the real /sys/bus/usb/devices/*/ with no override,
so its branches are driven through stubbed lsusb output: ffff:ffff matches
nothing on any machine and reaches the not-found arm, while 046d:c077
(control=on) and 046d:0825 (control=auto) are read-only probes, each guarded
so the case self-skips with a note if the host state changes rather than
silently asserting nothing.
Split across two test files to stay under the 250-line cap; run_all.sh globs
test_*.sh, so splitting costs nothing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
First of the 16 untested fixes/lib libs. Measured 37/37 = 100.00% through
lib/tests/run_all.sh, twice, the second run at --min 100 (exit 0), then
removed from the allowlist and confirmed with the real gate and no
exemption.
Adds two reusable pieces rather than a per-family harness clone:
- lib_test_core.sh -- PASS/FAIL tally, _t_eq/_t_contains/_t_lacks, a
mktemp -d sandbox wiped on EXIT, a FAKE_BIN stub dir on PATH, and
_t_stub/_t_unstub. _t_unstub runs `hash -r`: deleting a stub file alone
does not drop bash's cached executable path, so the real binary would
never be reached again.
- bt_harness.sh -- sources the core, sources the real scripts/lib/common.sh
for the production log_*/has_cmd helpers (verified inert at source time),
and redefines _btctl and apply_fix verbatim from fix_bluetooth.sh, since
those live in the entry script rather than in any lib. Same trade-off
pacman_hook_stall_harness.sh already makes with log_size.
pacman_hook_stall_harness.sh is deliberately left alone: its six libs are
already off the allowlist, i.e. gated at 100% with no exemption, so
refactoring it onto the new core would risk banked work for no gain. jscpd
reads 1.6% against the hook's own flags, and **/lib/tests/** is excluded
from that scan anyway, so the duplication budget is not at stake.
_t_only_path exists because every tool these libs call (usbreset, pactl,
lsusb, ...) genuinely exists on an Arch desktop: a prepended stub dir
cannot exercise a "not installed" branch, so those cases drop PATH to
FAKE_BIN alone.
The three coverage runs were made in a detached worktree: a second session
was writing to this repo concurrently, and the jail aborts on any
working-tree delta during a run.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Focus Owner hides every package absent from the allowlist, so a newly
installed lyricanki was installed=true but hidden=true: absent from the
launcher and from `pm list packages`, and `am start` reported the activity
did not exist even though dumpsys listed its MAIN/LAUNCHER filter. Nothing
about the app itself was wrong.
policy.json is regenerated from config.sh by python_pkg.focus_policy, never
hand-edited.
Also drops com.android.launcher3 and com.android.camera2 from
SYSTEM_NEVER_DISABLE. That list is prefix-matched and both are strict
supersets of entries already in it, so coverage is byte-identical; this
keeps config.sh under the 250-line cap without weakening the policy.
com.android.settings stays: its only shorter match is com.android.se, an
accidental substring rather than intent.
Verified on the phone: signer matched so the Device Owner upgraded in
place, dpm still reports DeviceOwner, lyricanki flipped to hidden=false,
and workout_app + the dialer are untouched.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RAEChhK3sWGxd6ucsUQmcT
Each measured 100.00% through fixes/lib/tests/run_all.sh, twice, with the
second run gating at --min 100 so a shortfall would exit non-zero rather than
print a number I might misread:
pacman_hook_stall_capture.sh 59/59
pacman_hook_stall_load.sh 25/25
pacman_hook_stall_summary.sh 18/18
pacman_hook_stall_watch.sh 24/24
pacman_hook_stall_usage.sh 1/1
Two matching runs was the bar because this directory's suites drive
watch_forever timing loops, and a timing-dependent suite can vary its line
set between runs -- dwm_config.sh was observed doing exactly that. These five
did not vary.
pacman_hook_stall_setup.sh stays: 26/27 = 96.30%, uncovered line 7.
usage.sh's denominator of 1 is honest rather than broken. The lib is a single
usage() whose body is one `cat <<EOF` heredoc plus `exit 0`, and heredoc
content is data, not statements. Its tests assert on the rendered text, so
the behaviour is exercised even though only one line is instrumentable.
summary.sh has one off-set line (36). That cannot inflate the figure -- an
off-set line is by construction one the trace SAW EXECUTE -- but it does mean
the lib is measured over a subset of its statements.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The handoff recorded fixes/lib as having "no harness at all". It already had
8 test files and 2 harnesses. What was missing is run_all.sh -- the subject
is_covered() actually builds, and the only thing the gate measures through.
All 8 test files pass through it.
No jail_args beside it, deliberately: the harness confines every write to its
own `mktemp -d` and intercepts external commands with a PATH stub dir, and
grep finds no absolute write to /etc, /var or /usr in any of the tests. So it
runs bare, like the transcribe suite.
First measurement through it: pacman_hook_stall_usage.sh is 1/1 = 100.00%.
That denominator is honest, not broken -- the lib is a single usage() whose
body is one `cat <<EOF` heredoc plus `exit 0`, and heredoc content is data
rather than statements.
16 of the 22 libs there still have no tests at all. That is now the largest
block of real test-writing work left in the campaign.
Also rewrites NEXT_SESSION_PROMPT.md, whose central claim ("the instrument is
the bottleneck, not the tests") stopped being true this session. The new one
leads with re-measuring the 103 allowlisted libs on the fixed instrument
before writing any new test.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The pre-push prettier hook rejected the table and list formatting introduced
by the two preceding commits. Formatting only -- no content, number or claim
is changed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Answers "what is billed on every turn regardless of use" by measurement
instead of recall, which is the part earlier audits kept getting wrong.
A surface is reported dead only when unused in BOTH windows. One earlier
audit's seven-day dead list included six servers that had been used within
thirty days, so a single window is not evidence. Shares are weighted with
cache-read weight 0.1, so a cached surface is not billed as though it were
re-read in full each turn.
Verified by running it, not just by tests: the report renders over 44,311
turns, putting rules/ at 4,907 tokens (1.77%), MEMORY.md at 4,312 (1.55%),
memories/ at 3,758 (1.35%), skill descriptions at 3,279 (1.18%) and CLAUDE.md
at 2,057 (0.74%), with per-MCP 7d/30d call counts behind each verdict.
The files were found untracked in the tree from an earlier session; every
module is already under the 250-line cap, so nothing needed splitting, and
the package was already at 100% branch coverage, so nothing needed new tests.
This lands the instrument only -- no surface is unregistered on its say-so.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The report printed a COUNT of traced lines that kcov never listed as
instrumentable. A count cannot be acted on: deciding whether such a line is a
real statement or a function declaration needs the line number. It now lists
them, the way it already lists uncovered lines.
Reading them across every lib with a harness splits them cleanly:
benign dwm_config.sh:27 (`heal_config() {`), transcribe_deps.sh:9
(one-line function declaration), clean_audio_filters.sh:197,200
(`((running++))`) -- correctly not statements
real aw_autostart.sh 68,70,73,79-81,83-84,87-88,94 -- plain echo, local,
if, `sudo -u ... mkdir`, `cat >` heredoc. kcov lists nothing after
the `{ ... } >>"$file"` command group at lines 61-65.
So kcov's LINE SET is incomplete too, not just its hit counts. This does NOT
inflate anything: a line reaches that list only because the TRACE SAW IT RUN,
so folding all 11 into both halves moves aw_autostart from 80/82 = 97.56% to
91/93 = 97.85% -- up. An off-set line is by construction an executed line, so
it can only ever be a missing COVERED line, never a missing uncovered one.
What it does mean is that such a subject is measured over a subset of its
statements. The percentage is honest about the subset, not the whole file,
which is why the lines are now printed rather than tallied.
Separately: nc_php.sh reports "kcov instrumented no lines" on the PRE-FIX
instrument too -- checked at 078d3463 in a throwaway worktree -- so it is not
a regression from the two-pass change. The 84/85 = 98.82% in earlier notes
does not reproduce and is recorded as unverified.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The instrument, not the tests, was the bottleneck. Coverage is now
(PS4-traced lines | kcov hits) & (kcov line set - continuation lines).
The trace deliberately does NOT supply the denominator. A trace reports only
lines that RAN, so using it for both halves would make every subject 100.00%
and arm a repo-wide gate that checks nothing.
Three measured findings, each of which yields a silently wrong trace:
1. kcov's ptrace and bash's xtrace cannot share a process. Under
SHELLOPTS=xtrace every line kcov recorded as hits="1" comes back hits="0"
-- reproduced A/B on one subject. Hence two passes, two namespaces.
2. SHELLOPTS is readonly inside bash, so `export SHELLOPTS=xtrace` fails and
tracing never turns on; and `set -x` does not reach child processes.
3. Decisive: bash under `unshare --map-root-user` runs PRIVILEGED and
DISCARDS an inherited PS4, falling back to "+ " while still honouring
xtrace. The trace loses its file:line prefix and every subject reads 0%.
Fixed by delivering PS4 and `set -x` through a BASH_ENV file, which also
solves child-process propagation.
A fourth trap, in the report: kcov records filename="dwm_config.sh", a bare
basename. The source path comes from the trace's ${BASH_SOURCE} or a unique
repo search; an ambiguous basename yields no exclusion rather than a guess.
Measured through run_all.sh, nothing down:
dot_resolver_install.sh 39/39 = 100.00% (acceptance, unchanged)
rpi_nc_ca.sh 73/73 = 100.00% (was 72/73 = 98.63%)
dwm_config.sh 65/66 = 98.48% (was 35/73 = 47.95%)
rpi_nc_install.sh 85/88 = 96.59% (was 10/88 = 11.36%)
rpi_nc_ca reaching 100% is the load-bearing check: its missing line 141 was
cross-process contagion, and the trace pass has no kcov in it to contaminate.
dwm_config's remaining line 39 is a `done < <(...)`. Left in the denominator
on purpose -- removing lines inflates coverage, the fail-open direction, so a
second exclusion rule needs per-line justification, not a pattern from one lib.
Both new Python modules hold 100% branch coverage (29 tests).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Settles a question the campaign has worked around for three sessions:
rpi_nc_install.sh's 11.36% is an instrument failure, not a testing gap.
Measured against an independent instrument. Driving the lib under
PS4='+PS4:${BASH_SOURCE##*/}:${LINENO} ' with set -x traces 55 distinct lines
as executed, and 45 of them are lines kcov reports as UNCOVERED -- including
61, 62, 63, 66, 68, 102, 118 and 146. The lines run; kcov does not record
them.
That splits the problem in two, and they need different fixes:
(a) WRONG DENOMINATOR -- continuation lines of a multi-line quoted argument
counted as statements, which bash -x never calls executable at all.
Fixable in shell_coverage_report.py. Holds dwm_config.sh down.
(b) WRONG NUMERATOR -- ordinary statements executing unrecorded. Caps
rpi_nc_install.sh, and no test can move it.
The distinction is the point. A denominator filter is the obvious fix and I
nearly built one last session; it would have done nothing for (b).
Also confirmed no test can raise rpi_nc_install.sh: the file is one 227-line
function called twice, an `exit 88` sentinel proves the suite runs to
completion, and its assertions on save_config (line 219) and the password
message (line 227) PASS while kcov calls both lines uncovered.
kcov 43 is the current release, so upgrading is not an available fix.
The kcov material moved to docs/kcov-under-report.md (the 250-line cap), and
NEXT_SESSION_PROMPT.md now leads with the instrument work rather than the
next lib -- 201 in-scope libs, 98 passing, 103 exempt, and the instrument is
the bottleneck.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016sxxsDHWgdRhmdmzNMA6XM
Two defects found in review, both of which made checks PASS more easily, so
neither ever went red.
GATE: is_covered() hardcoded its jail flags and never read the jail_args
marker that ci_mirror.sh and shell-tests.yml both consume. The features/lib
suite needs /usr/local/share and /etc/pki and aborts under set -e without
them -- but the coverage number still printed, because kcov had already
recorded whatever the dead run reached. The gate would have graded 44 libs on
a suite that never finished. It now reads the marker, keeps the historical
defaults for suites without one, fails closed on an empty marker, and passes
--fail-on-case-error.
Negative control: stripping the binds makes the gate exit 1 and restoring
them returns exit 0. Before this commit both cases exited 0 -- the gate could
not fail on an aborting suite.
HASH CACHE: the _t_unstub helper written earlier today for the transcribe
harness was needed in features_harness.sh too. Two assertions were vacuous --
test_dwm_config's "verify still reports when dwm is absent", and
test_rpi_nc_ca's uname removal, where a cached uname left every later branch
running under a faked Darwin. All four `rm -f "$TEST_TMPDIR/bin/` sites in
the repo were audited; the other two are sound (one probes with [[ -f ]], the
other runs in a subshell).
No measured figure moved: dot_resolver_install.sh still 39/39 = 100.00% with
real absence enforced, so its off-allowlist clearance holds.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016sxxsDHWgdRhmdmzNMA6XM
A 38-assertion suite for the last lib in the directory without one. Every
one of its five libs now has a suite; at the start of the session it had
none. 118 assertions across the directory, 0 failures.
build_filters is asserted on the EXACT filter string it emits, because that
string is the lib's entire product. An arnndn chain missing its mandatory
48 kHz resample, or a preset silently falling through to no dynamics, yields
worse audio and never an error -- nothing else would catch it.
Every exit code is pinned: 1 for a missing command, 11 for "RNNoise required
but unavailable" (refusing to quietly fall back to afftdn), 8 for an ffmpeg
whose afftdn lacks md=.
Measured 83/93 = 89.25%; it stays on the allowlist. All ten uncovered lines
are the documented under-report, not gaps: an `exit 66` sentinel proves
require_cmd's lines 37-38 execute, 52-58 are a multi-line array literal (the
confirmed continuation-line artifact), and 129 is an empty `*) ;;` arm whose
branch has a passing assertion. The gate uses the measured figure, so the
exemption stands regardless.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016sxxsDHWgdRhmdmzNMA6XM
A 29-assertion suite covering all six functions. transcribe_venv.sh measures
53/53 = 100.00% through the directory runner, and check_shell_coverage.sh
passes it with the exemption deleted.
Assertions are on recorded ARGV because which index a pip install points at
IS the behaviour: the OpenNMT cu12 wheel index on the NVIDIA path, the
CPU-only torch index for diarization (forced to avoid mismatched CUDA
builds). "pip was called" cannot see either.
generate_test_audio's four-deep fallback chain is covered by failing each
generator in turn. Every stub materialises the wav, because the chain
advances on `[[ ! -s $tmpwav ]]` -- a record-only stub would silently
exercise the branch after the one under test.
One assertion was wrong and got corrected rather than forced: I claimed a
wholly-failing pip still returns 0, but the final `python "$PY_HELPERS"
deps-installed` carries no `|| log`, so it does not. The suite now asserts
what the subject actually guarantees -- all four warnings emitted, no earlier
failure aborting the run.
SC2016 fixed at the source with a quoted heredoc, not suppressed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016sxxsDHWgdRhmdmzNMA6XM
The first allowlist removal of the campaign this session. unity_handler.sh
measures 32/32 = 100.00% through its directory's runner, and
check_shell_coverage.sh passes it with the exemption DELETED -- the real
gate, not just the measurement.
HARNESS BUG found getting there: bash caches executable locations in a hash
table, so `command -v foo` keeps succeeding after foo is removed from the
stub dir. Every "tool is missing" test was therefore running as a "tool is
present" test, asserting the opposite of its own name while looking
plausible. Reproduced minimally: after rm, STILL FOUND; after `hash -r`,
gone.
A _t_unstub helper now removes the stub and clears the table. The two suites
committed an hour ago had the same defect and are corrected here --
test_transcribe_pkgmgr.sh goes from 15 to 18 passing assertions, so three of
its checks had been vacuous. This is exactly the failure the campaign exists
to catch, found in its own test code.
`basename` joins the restricted PATH because the SUBJECT calls it; the suite
avoids needing `tail` via a parameter expansion instead, since every binary
in that PATH is one more thing a stub's shebang can trip over.
transcribe_deps.sh (92.31%) and transcribe_pkgmgr.sh (98.00%) re-measured
unchanged and stay on the allowlist.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016sxxsDHWgdRhmdmzNMA6XM
The directory held five allowlisted libs and no tests. It now has a harness,
a runner, and suites for two libs: transcribe_deps.sh (the fourth
hardest-first target, 48/52 = 92.31%) and transcribe_pkgmgr.sh (49/50 =
98.00%). Both stay on the allowlist. 15 + 18 assertions, 0 failures.
No jail is needed here and no jail_args marker is created: every branch turns
on `command -v <manager>` and every install goes through sudo, both
intercepted by a PATH stub dir. Verified the runner exits 0 when invoked
bare, which is how both gates call it.
The harness REPLACES PATH rather than prepending. This host really has
pacman, and a prepended stub dir cannot HIDE a binary, so every "manager not
installed" case would have asserted nothing. `bash` and `env` are in the
private coreutils dir for a non-obvious reason: every stub's
`#!/usr/bin/env bash` shebang resolves through that restricted PATH, and
without them each stub died with "env: 'bash': No such file or directory"
while looking exactly like a coverage gap.
Assertions are on recorded ARGV rather than on the call happening, because
which packages a branch installs is what differs between distros --
pacman gets python-virtualenv where apt gets python3-venv.
Three assertions aimed at the libsndfile branch were REMOVED rather than left
green-but-vacuous: /usr/lib/libsndfile.so exists on this host, so
need_libsndfile is legitimately 0 and lines 19/52/53/55 are unreachable
without a jail that hides the library. The limitation is written into the
suite instead of asserted around.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016sxxsDHWgdRhmdmzNMA6XM
The previous commit attributed rpi_nc_ca.sh's 72/73 to a heredoc BODY being
counted in the denominator. That was wrong. Line 141 is `cat <<'EOF'` -- the
heredoc's OPENING line, an ordinary statement. The body starts at 142.
The real cause is that the kcov under-report is contagious across processes
sharing one jail. Bisected over the five sibling suites: run
test_dwm_config.sh before test_rpi_nc_ca.sh and line 141 is lost; run
test_aw_autostart.sh or test_dot_resolver_install.sh first and it is kept.
Alone the CA suite reports 73/73 = 100.00%; through run_all.sh it reports
72/73 = 98.63%. Both figures reproduce twice, so this is deterministic
contamination rather than flakiness.
Nothing is untested: with dwm_config running first, the CA suite still exits
0 with every assertion passing, show_help included. The line executes and is
not traced.
A heredoc-body detector was prototyped and deleted once it showed line 141 is
not a body line -- a fix aimed at the wrong defect.
The operational rule this implies is now written down: is_covered() measures
through run_all.sh, so the runner's number is the lib's coverage, and a
suite's own single-file figure must never be reported in its place.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016sxxsDHWgdRhmdmzNMA6XM
