Both pick up the per-instance temp file name, so an app's UI and
headless engines stop losing writes to each other.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two FileLogPersistence instances on one file in one process (a Flutter
UI engine and a headless engine) shared <file>.<pid>.tmp, so one rename
moved the other's temp away and that write threw PathNotFoundException:
wake_alarm's bypass publish was lost to it on the emulator, 2026-11-22.
A random per-instance token now goes beside the pid.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A key=value *.lock file (name/repo/version/sha256) pins one package out of
an Arch-style repo .db. The adapter reads the db (gzip or zstd tar) via a
new http.get_bytes, matches %NAME% in each desc, and the judge compares
with a port of pacman's vercmp (epoch:pkgver-pkgrel), not PEP 440. A right
version with a wrong sha256 is a lock mismatch.
Locks are recognised by content (a repo= line ending in .db in the first
4 KB of a *.lock), so explicit paths from pre-commit and --all both work.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Moves all three onto kuhy-credit 0.3.2 and onto each other's new
tags, so a consumer can pin the newest of each without the resolver
conflicting. gatelock 0.16.1 also ships the 2026-11-21 lock-order
check and the earned-time 0.23.0 pin.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move the earned-time pin v0.22.0 -> v0.23.0 in gatelock
(pyproject.toml) and focus_project (pyproject.toml,
requirements.txt), and v0.21.0 -> v0.23.0 in focus_project's
.pre-commit-config.yaml. 0.23.0 adds the LeetCode review top-up.
gatelock's sickdays pin moves v0.5.5 -> v0.5.6 because the
dependency-freshness gate requires the newest stable tag.
Pin-only: no gatelock/focus_project release, no new tag.
Tests green against 0.23.0 (focus_project 339, gatelock 649,
the latter also against sickdays 0.5.6).
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXYyzSw5S9UTM5k5M1
DOCS-review.md holds the coach row contract, the five pay rules, the
start date 2026-11-23 and every fail-closed path; README points at it
and lists the new API.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXLvi9aq4MxL4XHKJ4
The guard ledger is found by stripping REVIEW.ledger off the review
ledger path; a path that does not end in it now warns and answers
None instead of reading the wrong file or raising IndexError.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXLvi9aq4MxL4XHKJ4
A finished coach review pays 20 shutdown minutes, once a day, as a
top-up: never a rung, never in the floor, added last and capped at the
ceiling (sick days too). A row pays only if signed, sourced, first for
its slug ever, and backed by a signed non-manual accepted submission
in the guard ledger from the 14 days before. From 2026-11-23.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXLvi9aq4MxL4XHKJ4
0.1.6 adds /leetcode review: the review and skeleton events, review_cli,
the coach's own signed review ledger and DOCS-review.md. Not tagged here.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXLvi9aq4MxL4XHKJ4
Replace the placeholder in DOCS-skeletons.md with the real command
and say when it refuses, so the rule text and the CLI agree.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXLvi9aq4MxL4XHKJ4
README lists the review layer, the two new events, the review rule in
progress and the three review_cli commands.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXLvi9aq4MxL4XHKJ4
DOCS-review.md is the /leetcode review flow Claude follows; SKILL.md routes
a first token review to it and points at the skeleton rules, staying at
250 lines. The md-naming gate takes DOCS-* names only, so the text is not
REVIEW.md. test_review_docs.py holds the quoted commands to the CLI.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXLvi9aq4MxL4XHKJ4
Add the review and skeleton session-log events, the review_cli write path
(review-list, review-finish, skeleton-open) and the coach's own signed
review ledger. A review event marks the day solved and starts the +3/+10
re-test clock. The flagged report is read through the guard CLI; a guard
failure is could-not-check, never nothing-to-review.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXLvi9aq4MxL4XHKJ4
The CLI imports kuhy_credit since the logo commit, but install.sh put
only the package itself (--no-deps) in the venv, so every service run
died with ModuleNotFoundError and no game bundle became a task.
test_install.py checks one pip line per pyproject dependency.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
verify_all_lockers now runs six lockers at kuhy's ranks (workout,
book, tutor, diet, focus, LeetCode last), fails when they come up out
of order, and checks each installed locker's rank constant.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Both CLIs import kuhy_credit since the logo commit, but the installers
only put crdt-sync and freedays in the venv, so the deploy's verify
step crashed with ModuleNotFoundError.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The logo commit changed sickdays and app_errors after their tags and
left app_errors/uv.lock without the kuhy-credit dependency; relock it.
kuhy_credit only gained README text but the deploy gate needs its tag
to match HEAD.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A ListView child got the full row width; the painter scaled the dark
field across the row and the glyph left the clip, so the logo was
invisible in punchme's settings. Loosen constraints at the root and
scale by the short side. Release 0.3.1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The 0.5.0 tag passed 'packages/design_system/...' plus package:,
so flutter_svg loaded a doubled key and the logo never rendered.
Lands the logo on main (0.5.0 was tagged off-branch) with tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A repo under ~/.cache (agent worktrees) had every file exempt, so the
local gate passed what CI rejects.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
checkPermission now reports deniedForever after a past request (prefs
flag + no rationale), so an app's resume re-check keeps the state.
currentLocation asks every enabled provider at once: fused alone sat
out its 30 s timeout on the emulator while GPS had a fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Flutter plugin over AOSP LocationManager (androidx LocationManagerCompat):
permission check/request, location switch, one-shot fix and a fix stream.
Replaces geolocator in umbrella and workout_app, whose Android side links
non-free play-services-location even with forceLocationManager.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
\kuhycreditcorner[<inset>] puts the linked logo on every page's
bottom-right corner via the shipout/foreground hook.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Generated latex/kuhy-credit.sty draws the M1 logo as TikZ polygons and
links it to the site; compile E2E in pytest. Version 0.2.0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Flutter example gets its android and web platform dirs (label
"Credit" for the launcher contract); logo painter moves to a part
file to stay under 250 lines.
- Godot click tests size the headless window to 640x360 (the official
build opens 64x64 and drops clicks outside it) and hide GUT's GUI.
- godot/test.sh writes godot output to files, not $(...), and fails
when fewer tests ran than tests/ declares.
- README Python snippet passes ruff format.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Held Enter no longer opens the site (Tk, web), Godot exports
rebuild after _init, strict SVG parser, relative --dir.
Todo: 3a5477b9
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Drop the explain-back exit and the opening-belief question. Once code
passes the examples (or kuhy says done), the coach writes the post-mortem
from the transcript, builds the cards, and appends missing toolbox entries
itself; toolbox_cards add gains --define for new vocabulary properties.
Adds the contribution technique entry from sum-of-all-odd-length-subarrays.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWWTVKrxg1PGVP4wc6W
Sonnet, opus and fable all run the skill as written; only haiku
gets the /model sonnet nudge. Help and DOCS name LEETCODE_MODEL.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWWTVKrxg1PGVP4wc6W
The round-safe master leaves a tab icon's corners empty: at 16 px the
letters spanned 11 pixels and the H lost its crossbar. This variant
fits the same letters to the square (14 of 16 pixels).
Todo: 4a9d9d9a
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The filled Fourier K U / H Y mark kuhy chose, refitted so a round
avatar crop keeps 18 % of its radius as air (it trimmed the K and Y).
Todo: 4a9d9d9a
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The ruff dev pin bump changed both dirs after their tags, so
deploy's tag check failed.
Todo: 2556952d
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
feedback_inbox and app_errors guessed kuhyx/<folder> for a task's repo,
which filed cleanup-plate under kuhyx/cleanup-plate (real:
Sepci0/cleanup-plate) and showed phantom repos on the board. Both now
read the GitHub owner/name from ~/src/<folder>'s origin remote (https or
ssh forms) and fall back to the guess only when there is no checkout or
its origin is not on GitHub. app-errors 0.1.2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWZfgt44aEKxvGBJkwP
kuhy asked for three variants of each candidate type, and for the
K U / H Y grid to step its bottom row to the right. That gives 18
masters, a1..f3. Variant 1 of B-F is the original master, renamed.
The aligned A grid is replaced by stepped ones: a1 a small step,
a2 a half-letter brick step, a3 the small step with a squared U.
The rows now also sit 32 px further apart so they never touch.
All 18 pass make_logo_sheet.py's flat-logo rules, and the contact
sheet is regenerated.
Todo: 4a9d9d9a-0
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sweep.py ignored argv, so --help walked every repo and rewrote SWEEP.md.
It now parses argv first: --help exits 0 and a bad arg exits 2, both
before any walk. --dry-run prints the report to stdout and the summary to
stderr without writing; --out picks the report path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Candidates A-F (grid KUHY, K, kuhy wordmark, KY ligature, ringed-K
glyph, KR) as flat stroke SVGs in ink and accent, one 40/512 stroke.
make_logo_sheet.py validates them and renders the contact sheet at
512/64/16 px in colour, one-colour and inverted.
Todo: f75e1003-0
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
logo_check.py renders one SVG into one PNG with four panels: 16 px,
one-colour, inverted and squint, small panels enlarged nearest-neighbour.
Todo: f75e1003-1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
E2eGate takes an optional recover callback. When the key stays locked
for want of a passphrase (needsPassphrase or wrongKey), the gate asks
it once, saves what it returns into the keystore and opens again. A
bad header never asks; a wrong answer leaves the lock button.
fetchWrapperPassphrase reads <app>/sync-passphrase from the desktop
app's PC wrapper (null on 404, a malformed body, or no wrapper).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWY91NZkk1AH6ugi6ya
kartka now reports through crdt_sync_flutter 0.9.1, so the shared ingest
reads /kartka (old entry shape included) and kartka's own ingest retires.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Without --dart-define=APP_BUILD (phone_deploy, local builds) error
reports carried build 0; the installed versionCode now fills it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A game's own test and smoke gates leaked test errors into the
spool; headless runs and APP_ERRORS_OFF=1 now write nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
RTDB and local spool reports merged by signature into the board's
app-error contract; resolve command; systemd --user timer via deploy.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A day closed by a non-paying charge (escape hatch, outage) never gets a
source=leetcode row, so its LeetCode gate could never be satisfied. The
rung is now pending only while payable: done_today false AND no
verified charge row for today, any source.
Failure paths: charge check unreadable (ledger, key, verify error) ->
rung unchecked -> pending exit 2 (FAIL CLOSED). An unsigned or old
charge never settles a day, so the gate still arms (closed). earned_time
has no public "any charge" helper, so the rows are read from the ledger
and checked with earned_time.verified.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGbnVt12uFXKgNYm8
The tag rtdb-relay-v0.2.0 releases the Python worker and the Dart
client together; the Python code is unchanged since 0.1.1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYeKVg9WihaikbfDSe
Moves the tutor app's generic relay client (RTDB REST/SSE, e2e
sealing per an app layout, request hub, rev watcher, passphrase gate,
Google->Firebase session, desktop credential adoption) into
rtdb_relay/dart. The tutor vectors and a differential test against
the tutor's frozen pre-extraction sealer pin the wire format.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYeKVg9WihaikbfDSe
Add gatelock.evening_chain: `pending` lists the unpaid rungs (book,
tutor, LeetCode), `run` runs each gate in order before night lockdown.
Also RANK_EVENING_GATE = 150, systemd/evening-gates.service and
install_evening_gates.sh (never enables or starts the unit).
Failure paths:
- pending: unreadable ledger or key, reader error, any exception ->
exit 2 on stderr. FAIL CLOSED (root enters lockdown).
- pending: free day, sick day, 23:00 or later -> exit 1 (stands down).
- run: a rung it cannot check is unpaid, its gate stays up (closed).
- run: gate fails, or exits 0 with the rung unpaid -> skipped; two in
a row -> exit 1, never a spin (stops, root enters lockdown: closed).
- run: gate exit 10 -> exit 0; exit 3 (no X) -> exit 3; gate still up
at 23:00 is killed -> exit 0 (the hard time wins).
- earned_time/freedays/sickdays are an optional `evening` extra, so a
locker never fails to start over them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGbnVt12uFXKgNYm8
The tutor's generic relay transport (layout rules, sealer, REST client,
SSE, mirror, unlock) moved here unchanged in format, plus a generic
RequestServer (kind allowlist, sealed-only requests, lease, visible
handler failures) and a presence heartbeat for a second consumer
(todoImplementer). The tutor's contract vectors pin the byte format.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Dated switch GAMING_UNLIMITED_FROM (2026-11-16) and gaming_enforced(day);
Resolution.gaming_enforced; leisure_active False from it; sick shutdown
uses rung pay from it. No public name removed; shutdown unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWViGJJaqBKGr1Hbppr
lint.sh ran gdlint/gdformat via xargs -P; on a cold cache the
processes race in os.makedirs and report a file as unopenable.
Run one process per tool. Pin CI Python 3.15.2, chmod Godot.
Bump version and README tags to 0.1.1.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXs4gEhEUMr9rgYG9V
An open lock window never noticed a credit written by another process:
screen-locker opened at 07:55, a workout was credited at 07:59, and the
window stayed up. Gates read the log once at startup.
watch_credit polls the files that carry the credit on the Tk after loop
(one stat per poll, owner verdict only when a file changed), and hands a
done verdict to the owner's close path. Failure mode is closed: an
unreadable file or a raising verdict is logged and retried.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWWfb63jncj9Lca2jeX
main was red on file length, the noqa ban, codespell, test naming and
dart coverage; this change touches every one of those jobs.
- sickdays: set_enabled is keyword-only; the panel test splits in two
with shared probes; the sync wire format moves to DOCS-sync.md.
- earned_time: _focus types the project as a Protocol, no noqa; the
gate-maturity section moves to DOCS-maturity.md; codespell knows the
`checkin` event name.
- focus_project: real fixes for every noqa (local-day clocks, typed
fakes, logged broad catches, split helpers). The public API that
focus-guard calls keeps its signatures; their PLR0913 exemption
lives in pyproject with the reason, not in noqa.
- crdt-sync: pylint gets nacl/argon2, docstrings, and allows the
public E2eLayout.of; _blobs moves to _e2e_blobs.py.
- crdt_sync_dart: Argon2Params and the header errors move to
e2e_params.dart (re-exported); tests split and reach 100% lines.
- sql_drills: test helpers get a leading underscore, as in earned_time.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
pylint's parallel mode pickles its AST with dill, which crashes on
Python 3.15 (no code.co_lnotab): every Python lib's pre-commit job
was red. Hook rev v4.0.8 -> v4.1.4 and --jobs=1, as leetcode_coach
already did. focus_project imports earned_time but never declared
it, so its CI tests failed at collection; pin earned-time 0.21.0.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every pin dep_freshness reported behind moves to newest stable:
pnpm 12.21.0 (packageManager, with hash), typescript-eslint 8.72.2,
@eslint-react 5.30.2 (5.30.3 is inside the 24 h quarantine), eslint
10.15.0, astro 7.4.12, pylint 4.1.4, ruff 0.17.5, mypy 2.4.1, pytest
9.2.0, numpy 2.5.5, CI node 26.13.1 and the rest.
astro 7.4.12 brings rolldown, whose wasm binding pulls @emnapi 2.0
beta; pnpm then fed that beta to astro2tsx's peer and astro check
died. web_ui pins @emnapi/runtime 1.11.3 so the peer stays on stable.
sickdays 0.5.5 releases its requirements bump so deploy's tag check
holds. The corepack call that broke the gate lived in claude-config
(finish_gate.sh) and is fixed there.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
0.1.3 pinned crdt-sync 0.11.0 exactly, so leetcode-guard could not
resolve against freedays 0.4.3 / sickdays 0.5.4. Also coverage
7.17.3, pylint 4.1.4, ruff 0.17.5. 519 pass, 100%.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sickdays 0.5.3 and freedays 0.4.2 pinned crdt-sync 0.11.0 exactly,
so a consumer on 0.11.1 could not resolve. freedays 0.4.3 and
sickdays 0.5.4 move to 0.11.1 (additive: verified plaintext
cleanup). freedays 125 pass, sickdays 155 pass (1 Tk test needs a
display).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Flutter 3.50 deprecates package:flutter/material.dart. In an app whose
MaterialApp comes from material_ui, the passphrase button's showDialog
found no MaterialLocalizations and threw. The test now pumps a
material_ui MaterialApp: red on the old import, 107 pass after.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
clear_plaintext_prefix + tool/e2e_cleanup: dry run by default, never
creates a header, deletes the old plaintext prefix only when every old
record is in the encrypted copy.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
crdt_sync_dart 0.14.1, crdt-sync 0.11.0: the one-time copy skips empty
sources and never overwrites a destination that is not a sync log
(home-guard's pending upload). Python open_encrypted_client(state_dir=)
caches the derived key (0600) and the copy memo across daemon runs.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
sickdays 0.5.1 pinned crdt-sync 0.10.0 but still freedays 0.4.0, which
pins crdt-sync 0.9.2, so no app could install both: pip reported
ResolutionImpossible for diet-guard and wake-alarm.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
freedays 0.4.1, sickdays 0.5.1, leetcode-coach 0.1.2: crdt-sync
v0.10.0 (E2E), so apps can install them beside the E2E release.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Decodes an RTDB dump and lists plaintext leaves; the E2E done check.
A grep for "title" misses escaped blobs.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Byte-compatible with crdt_sync_dart 0.14.0, proven by golden vectors in
both directions. EncryptingStore, -e2e layout, one-time copy, and
open_encrypted_client for the Python daemons. Dev pins bumped.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Slashless prefixes keep revs/header inside their node; copy walks
nested dirs (RTDB bulk reads are one level), reruns only when old revs
change; derived key cached; bulk reads decrypt; Python golden vectors.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Flutter 3.50 deprecates package:flutter/material.dart for the
material_ui package; its ThemeData is a different type, so apps on
3.50 need the theme builders to return material_ui's.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Flutter 3.50 deprecates package:flutter/material.dart. An app on
material_ui gives flutter/material widgets no MaterialLocalizations,
so sync_settings_ui 0.8.0 and github_device_auth 0.5.0 move first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The text-free sickdays GUI (9d6b17f) was never released: main still
said 0.4.0 while the code moved past sickdays-v0.4.0, so every gate app
kept the worded panel. scripts/deploy.sh installs the sickdays CLI from
the live checkout and fails unless the version tags match HEAD and the
live sickdays is the text-free one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server stores a new_note task as '<repo> todos: <title>' plus a '## <title>' section, so the title no longer started with MANUAL WORK REQUIRED and the entry's own ## heading became a runnable part. After add_task the inbox now rewrites the note with update_task (exact title, section-free body). Marker is written first, format 2 only after the rewrite: a failed rewrite is retried, never re-added; old markers are repaired. install.sh counts folders from the Paths property.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXmfFFHX4fNhyFRqFq
Watches ~/data/*_binaries/feedback via systemd path units and files one MANUAL WORK REQUIRED task per complete bundle. Marker written only after the todo server accepts the task (fail closed, retried by the 5-min sweep). Stdlib only, 47 tests, 100% branch coverage.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXmfFFHX4fNhyFRqFq
A keyring prompt (gcr-prompter) that grabbed input before the
leetcode lock armed left the lock visible but unclickable: every
click went to the prompt hidden under it. Each failed
grab_set_global also froze Tk ~1 s, so the lock barely repainted.
- Raise known credential prompts above the lock after 2 s of a
blocked grab; games and other grabbers stay covered.
- Probe the grab on a side Xlib connection first so a blocked
grab never stalls the Tk loop (worst gap 1.05 s -> 0.05 s).
- Log the Tk grab error and stop blaming our own claim.
- 1.5 s startup grace in wait_for_turn, and a weaker armed lock
steps back (windowless, re-arms later) when a stronger one is
live, so only one lock is ever on screen.
- Real-X checks: verify_prompt_yield, verify_grab_responsive,
verify_all_lockers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWZbpxQZj41WpbQPuxB
Glyph + spoken label replaces every app-written string on the shared
Sync settings screen and the GitHub device-code dialog (kuhy's
textgate rule). Old sentences are semantics labels; tests find them
there. The device user code stays visible (GitHub's data).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Decide by AGP library marker files instead of dex strings:
play-services-base/basement reference GoogleSignInAccount without
linking play-services-auth.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWVQE5pYiASpRGgA3Kg
scripts/check_no_gms_signin.sh fails when a pubspec.lock or android
gradle file pulls google_sign_in / play-services-auth / googleid, or a
built APK (--apk) carries GMS auth classes, and names
google_account_token as the fix. install_repo_contract_gate.sh now
installs its shim and an always-run hook in Flutter Android repos.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWVQE5pYiASpRGgA3Kg
Tests only: sql_drills/leetcode beats back, other-project and browser
beats don't, submissions back only leetcode projects, the cutover
midnight, a tutor chain leaves earlier reading unbacked; earned_time
pays 0 units for a timer-only post-cutover day (real focus_project).
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A timer span earns only if a same-project tutor/sql_drills/leetcode
beat backs it (days from 2026-11-08). status() gains unbacked_minutes.
The 0.8.0 tag shipped only the tutor-chain half.
Todo: fda71ed1
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A score row completes mock only after a contiguous >= 90 min
focus session (600 s gap rule) that ended by the score row.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWZUTC9Zzdzsx7VW7dm
Audit found no external consumers. Drop their refs in md_naming
migrate tooling and the pre-commit comment.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYFhDShgeHByh6yT9L
It has one consumer, kuhyx/tutor (packages/code_runner). Tags
code-runner-v0.1.0 and -v0.1.1 stay so old pins still resolve.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYFhDShgeHByh6yT9L
Clears the dependency-freshness gate (5.0.0 -> 5.1.0); 519 tests
pass at 100% coverage on the new pin.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fresh claude -p (opus-5-5, 2.1.319) loaded vmbox-testing before any
host sudo, sandboxed install+verify+undo, caveated the host. n=1.
Todo: fda71ed1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
jsdom 31 rounds style numbers to 4 decimals; the range
slider test now compares percentages numerically.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYDm7szKeXoi7rinEy
Compile-time dart-define that prefixes every RTDB path with sandbox/<slot>; invalid values throw at startup. Empty keeps prod paths unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
FitGrid drew one dot per page; tunedeck Downloads with 548 files had 46
dots running off the screen. Above 12 pages show "n / N" instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A tutor heartbeat (a graded check-in card in the tutor app) is a beat
of the open timer segment like a checkin: within 600 s it extends the
segment, later it ends it at the last beat; it never opens one.
leetcode, sql_drills and browser heartbeats are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
xvfb-run's 640x480 screen compacts the scale to 0.7 and clamps
body and label to the same 11px floor.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWZd3smuxDGmEgXkYP8
Add gatelock.corner_strip, the top-right always-on-top status strip for a
gate that is stood down but not done. Raise the morning-session wait from
30 s to 180 s: on 2026-11-07 a boot-time DNS race let the 30 s wait expire
and both lockers armed mid morning-session; wake-alarm's refresher now
retries 150 s in-process.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWZd3smuxDGmEgXkYP8
omniroute-gateway start|health|update|backup|combo keeps the local
OmniRoute gateway up. systemd --user timers: health every 10 min
(real free-best chat; restart <=1/30 min, reactivate providers whose
cooldown ended, notify once per outage), update weekly Sun 12:00
(backup, newest stable -web tag, rollback of image + volume on a
failed health check).
Failure modes: health/update fail loudly (exit 1 + desktop + Signal);
all providers cooling = degraded (exit 0, one notice/day); a missing
signal-bot .env logs and exits non-zero on alert paths; health skips
while update holds the lock.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWZSdJmNjGUvLnHJ4WG
free-llm sends big texts to the free-best OmniRoute chain behind a
fail-closed privacy gate (public repos, /tmp, /var/log; credential
scan). claude-free runs Claude Code with the full ~/.claude setup on
free-best. Skill offload-free in ~/.claude says when to use them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYwGi2FJwfZ6QRB1im
Learner SQL piped into the mysql client could run `\! id` as container
root and read every DB. Now: --binary-mode (client commands error out),
a refused-backslash-line guard, learner exec as uid 65534, and a random
root password (0600 file, via env, adopted live on a legacy container).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
An edit that lowers a target or adds a replaced gate
holds from tomorrow, never today. apply_update returns
effective_from.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
Deep or oversized values from the sandbox now fail that example only; compare refuses nesting past 64 levels.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
kuhy 2026-11-02: keep the 180 base cut, pay 2 gaming min per focus min
(240 at 120/120), so a full focus day is 8 h again; an idle day stays 2 h.
FocusEarner.cut_gaming_minutes holds the cut (_resolve._gaming_cut) so the
pay (cut + pure bonuses again) can exceed it. Shutdown unchanged. Carry
rows (focus_project 0.6.0, <project>-<YYYYMMDD>-carry) pay like own rows;
test added. Failure modes unchanged: focus_project unusable = no project
(fail-open, logged); unreadable ledger = None, earns nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
A structured model function such as the tutor's ask_model now works
as `ask` in check, looks_like_code and ask_models (via: "model").
Raises fall through to Ollama, then fail open as before. v0.1.1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
Minutes past a day's target bank (max 60) and pay on the next project day; status shows real minutes (121/120) plus carried_in/carry_out; carry ledger rows.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
Public repo: statements are original, in LeetCode's layout. Real problems are fetched at runtime by network-marked tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
Standalone utils subproject (pyproject, pre-commit, CI, install.sh,
README with the library API). It no longer imports leetcode_guard:
guard_bridge resolves the three read-only calls on demand.
Q5: SQL problems skip req: and ask for an OUTPUT SHAPE (columns,
rows, grouping) before code; algorithm problems keep req:.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
Runs a solution (python3, javascript, typescript) against statement
examples in bubblewrap: no network, read-only root, scratch dir,
rlimits and a per-example timeout.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
Shared MySQL drill engine extracted from testsAndMisc: sandboxed
per-run db and user, exact-match cell diff, predict drills (builtin
theory set incl. an error drill), save_generated validation.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYGW5NZL5uVpnqmNKo
Python 3.14 stdlib package wiring git to the todo MCP server:
- closer (post-commit): runs the done_check of each Todo: trailer
task and close_task's it only on exit 0. It FAILS OPEN: the commit
already happened, so a down server, bad token or no gh prints one
warning and exits 0 instead of failing the hook.
- sweep: nightly drift report (systemd units shipped, not enabled).
- import-pr: a merged PR's '## Todo:' items become tasks, deduped.
- A done_check runs only if its sha256 is in
~/.config/todo-hooks/trusted-checks; 'Check: manual' notes are
never auto-closed.
- Hooks exposed in .pre-commit-hooks.yaml (post-commit, post-merge,
post-rewrite).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
python3 -m focus_project.open opens the tutor focus screen (start=1
when a project is active) and the project's work URL.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXdzckyBH7fEbPywpr
Android now uses google_account_token (no googleid or
play-services-auth). Desktop PKCE sends client_secret.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWY1k4muZfdVJfb7YGJ
Android AccountManager ID token for Firebase signInWithIdp, extracted from todo so the sync apps stop copying the Kotlin.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWY1k4muZfdVJfb7YGJ
A plugin has android/ but no launcher activity. Also rename the sickdays desktop entry to the one-word form the label rule requires; that red gate blocked every pubspec commit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWY1k4muZfdVJfb7YGJ
A focus-project day cuts the whole 180 of gaming base (LeetCode's free hour too); a full day still earns it all back. Shutdown unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWxdzckyBH7fEbPywpr
Ask JitPack metadata last for names no other repo knows and that
are not com.github.* (whose metadata is 401, tags used instead).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWY1k4muZfdVJfb7YGJ
earners_for swaps a focus project's replaced gates for one focus earner
paid per minute from the focus ledger (floor of summed minutes). The gaming
cut equals what the replaced penalised gates cost, so a project day is
neutral; maturity never delays it. An unusable focus_project means no
project (logged once).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYcmik6APnHfc9gEVy
Fail-closed: a failing source logs and adds 0 minutes;
a bad key, ledger or registry exits 1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYcmik6APnHfc9gEVy
RangeSlider.astro + <range-slider> replace the React component.
Pure helpers, tokens.css and range-slider.css are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWYdwRAtMg7ki6TKxuU
Save-as on mobile (SAF), save dialog on desktop, download on web; versioned envelope helper.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWXTUM9dpnCS2wtFkx4
FitGrid lays items out in a grid sized to the box it is given and pages
the overflow sideways with dots, so a screen never scrolls vertically.
First consumer: tunedeck's home tab.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WzxdWZT4baL56wRDmNioxj
http's ClientException prints its request URL, so 'network error ...:
$exc' embedded ?auth=<ID token> (RTDB) or ?key=<API key> (auth) in every
FirebaseSyncError / FirebaseAuthError, which apps show and log. The
messages now use exc.message only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CXYpMKCt4RFSbCMfYyAPCs
kartka's headless agent wrapper serves ~/.config/kartka-agent/
firebase_auth.json once per launch as /sync-credentials, and its weekly
source evaluator reads credential_store_for("kartka-source-eval"), which
otherwise falls back to the retired password. One run with
--app kartka-agent --app kartka-source-eval seeds both from one consent
tap. Both also join the default all-app run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S2brk4NNnRyTnt7cD2QF2f
Drives a Flutter web build in a persistent headless chromium (systemd-run
unit, own profile, CDP port) through the semantics tree: tree, tap by label,
text, key, scroll, wait, shot, console. Never touches the live display;
attaches only to HeadlessChrome. The MCP server wraps the same Driver.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZ4sULGfyav7EEwQ6vhbfg
The sufler PC TTS relay reads ~/.config/sufler/firebase_auth.json via
credential_store_for("sufler"), like tutor, so it needs a seeded session.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Jw52sNWquCp34dy3rj19GD
tool/coach_live_check.sh starts the real Relay (lane, backend, HTTP API,
phone mirror) on a free loopback port, never 8777, with claude, Firebase,
LeetCode, Ollama and Anki faked and every file under a temp dir
(_live_relay.py), then drives it through WrapperServer's /coach/* proxy
with WrapperCoachTransport (coach_live_check.dart):
- flow: start, step-1 partial then pass, a chat turn, code refused before
step 2, three step-2 fails via toolbox, a force, an unknown and an
over-long pick (bad_request), an also_accepted pass, run_code, end,
not_live after it; 404 for an unknown rid.
- slow: a 35 s coach turn through a 30 s proxy timeout answers, with
heartbeat.busy naming the rid and every poll under 2 s.
The flow's raw answers are committed as test/fixtures/coach_relay/ and
parsed by relay_fixtures_test.dart, so the models are checked against
what the relay sends, not what the fake's author believed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
phase.step2_attempts counted every step-2 grade with grade "fail",
and a force repeats the last try's grade, so three named-a-tool fails
plus "Continue anyway" showed attempts: 4 in the state. Measured
against the live relay through the app's desktop path; the contract
(and the snapshot's _coach_grades) already exclude forces.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The RTDB mirror wrote synchronously on the lane's one worker, so a hanging
Firebase (15 s connect timeouts) would have stalled every desktop request.
PhoneLink now only enqueues; relay.outbox writes on its own thread in the
contract's order (chat reset/lines, state, heartbeat, then the result),
keeps only the latest state and heartbeat, never drops a result or a chat
line, and retries a failed write first, with backoff.
The whole-request 280 s cap is enforced again: Coach.handle sets a
deadline and a coach turn gets what is left of it (at most 240 s), so an
ambiguous message's code check plus the turn cannot outlive the phone's
300 s "No reply" wait. The contract's caps line is back to that wording.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
leetcode_coach.relay implements DOCS-coach-relay.md: one paced serial lane
(1 start / 3 s, 5 queued, done-map re-answers a rid), the loopback API on
127.0.0.1:8777 (Host check, JSON only, 16 KiB), the RTDB request/result/
state/chat/heartbeat mirror over SSE with 60 s rescans and 24 h cleanup, the
five-step state machine derived from the session log, graders (steps 1/5
haiku -> Ollama, fail open "unverified"; step 2 through PhaseGate.screen;
step 3 mechanical against answer_key.json), the CLI's sonnet chat with the
log in the child env and the local hand-over, run-code on node --permission,
and step 4 through the harvester's read-only fetch_recent_ac once a minute
(three-valued: unknown is never "not yet").
New log events session_start/grade/chat/session_end land in session_log,
_coach_log and _well_formed together; the snapshot rows gain grades and
accepted (_coach_grades). A forced step-2 grade opens code (phase.req_state).
leetcode-coach-relay.service + an install.sh step (not run).
Contract fixes in the same commit: step 3 counts also_accepted and combined
entries' parts, step-2 grade events and the force event shape, next_step,
per-call model caps instead of a 280 s watchdog, heartbeat on busy changes,
the statement fetched in-process, enable + restart on reinstall.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
The exporter used to publish coach: null before the first session, so the
tab printed "unknown" everywhere with no way forward. Now a missing or empty
sessions dir exports empty problems/re-tests, every cluster at zero with its
problems (slug, number, title, URL) and next, and a how_to block (commands,
the 4-step method, first command). null stays for a real failure.
how_to lives once in leetcode_coach/how_to.py; DOCS-how-to.md quotes it and a
test fails on drift. The app's Coach tab shows a How to start card (monospace
command box with copy), one card per cluster with tappable problems, re-tests
with their command, and "unknown" + why only for a null section. Also fixes
the done/solved key mismatch (the app read a key no exporter wrote).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
The incomplete-list note shows meanings before code, so "a ^ x ^ x = a"
or "x & -x" handed over the answer. Rewrite the six leaking meanings plus
two that said "sorted", and pin it with
test_no_property_meaning_names_an_entry.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
The fetched statement carries LeetCode's topic tags and site hints ("Hash
Table", "Two Pointers"), so exempting it pre-cleared the answer. Some real
toolbox meanings illustrate with the operation (self-inverse: "a ^ x ^ x"),
so the incomplete-list note shows those properties by name only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
The gates read the toolbox through toolbox.load_safe (typed, fail-open)
beside main's raw load(); the session log follows LEETCODE_COACH_SESSIONS_DIR,
stamps LEETCODE_COACH_CLUSTER, and the card builder logs postmortem on
delivery. leetcode report dispatches beside cluster_cli; client notes move
to DOCS-clients.md to keep SKILL.md under the cap.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
"merge intervals" is #56's own title, so the leak bounce would have bounced
the coach's restatement of it; "dictionary" was a bare common word. A test
now pins that no entry name or alias equals any clusters.json title.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
- toolbox.json: 20 entries over a fixed 40-property vocabulary; no entry's
property set nests in another's, so every property card has one answer.
- clusters.json: decision 10's six clusters, ids plus slugs.
- progress.py (stdlib only): solved = code + postmortem, re-tests due at
+3/+10 days, cluster progress, all derived from the session logs.
- leetcode due / --cluster <name> / --retest <slug> (cluster_cli.py): due
re-tests of the cluster first, then the next problem with no code event,
skipping LeetCode-Accepted ones (live check fails open). A cold re-test
logs `retest` only once the session reached a post-mortem.
- Post-mortem cards are property -> operation only, generated by
toolbox_cards (shared slug `toolbox`); `add` appends an entry only after
kuhy confirms its properties. The local client accepts {"toolbox": [...]}.
- _coach_log publishes retests_due and clusters from the coach.
- requirements-sync treats sibling repo packages as first-party.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
The REPL passes os.environ to `claude -p`, so a PYTHONPATH set by the
launcher reached every hook and every Bash call of the coached session. The
old launcher did not do that. The launcher now runs `/usr/bin/python3 -P -m
leetcode_coach.leetcode_repl` and relies on install.sh's editable install,
the same way the Stop-hook shim does. A test now pins that the launcher
sets no PYTHONPATH.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
Phase A of TODO-coach.md (decisions 1-3). The skill code is now a package here
with the same behaviour as before. Nothing new is added. ~/.claude/skills/leetcode
and ~/.local/bin/leetcode become symlinks into the package, created by
scripts/link_leetcode_coach.sh, which install.sh runs.
- Package imports replace the bare sibling imports. To stay under the 250-line
cap, the Stop-hook detector is split into gate_detector.py and stop_gate.py,
the card rules move into card_rules.py, and the prompt text moves into
local_prompt.py.
- local_coach uses gate_detector directly. It no longer loads the hook file
with importlib.
- make_cards runs on the system python: a genanki pin (pyproject and
requirements) replaces the `uv run --script` header. The note guid is
passed to genanki.Note explicitly, so CardNote is no longer subclassed;
the guid values are unchanged.
- stop_gate also accepts the package path as a skill-load marker, because the
harness may print the resolved symlink path.
- Every swallowing except now logs a warning. A NullHandler on the package
logger keeps stdout/stderr the same as before.
- Modules whose prints are their interface (REPL, local session, card builder,
fetcher, eval, Stop hook) get T201 per-file ignores, and their tests assert
on stdout.
- hook_shims/ holds the 3-line ~/.claude/hooks shim. It is excluded from
ruff, mypy, pylint and coverage, and a test runs it end to end.
- Pre-commit: a requirements-sync entry for each package, leetcode_coach and
genanki added to the pylint hook, and the name-tests-test exclude widened.
coverage and ruff are bumped to the latest stable releases because the
dependency-freshness gate requires it.
Verification: 111 coach tests pass with 100% branch coverage. ruff, mypy
strict and pylint are clean (10.00). On the 370 assistant replies in 39 real
/leetcode transcripts, the old hook and stop_gate give the same verdict
(0 mismatches, 20 blocked). fetch_problem output (live two-sum) and
make_cards --dry-run output are byte-identical to the old skill, and so are
the note guids.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PjvF5DZG9JduSBYKxqyysi
Flutter repos now fail the repo contract when a launcher icon's ink is not on
the shared 340px circle (png_ink), the launcher label breaks the naming rule
(capital, ASCII letters, <= 10 chars, <= 522 wide in GoogleSansFlex), or an
applicationIdSuffix variant still carries the LAUNCHER category. hook_sync
rewrites a repo's stale repo-contract `files:` line to the widened template.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017k9j5dbjenpfZj7dh2faPN
In CI the gate defers to the workflow and returns 0, so the real-checker
test saw 0 instead of 1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NBScELfMUiieK9ManFbS3Y
pre-commit 4.7.0, pylint 4.1.3 and ruff 0.17.2 across crdt-sync,
earned_time, freedays, gatelock, music_theory and sickdays; numpy 2.5.4 in
music_theory (requirements, pyproject and the mypy/pylint hook deps). The
ruff-pre-commit rev follows to v0.17.2 in those configs and vmbox's, so the
hook lints with the version requirements.txt pins. No Python lockfiles exist.
All six suites at 100% coverage; every package's pre-commit config green.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NBScELfMUiieK9ManFbS3Y
- sickdays: normalise_keep raises TypeError (not ValueError) for a bare
string -- ruff 0.17.2 TRY004; it is a caller type error, the CLI always
passes a list.
- sickdays: remove_sick_day and set_kept_lockers share require_sick(), which
clears pylint R0801 (duplicate-code, 9.99/10 under --fail-under=10).
- repo_contract/web.py: tomllib sorts as stdlib under ruff 0.17.x (I001).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NBScELfMUiieK9ManFbS3Y
The checker shipped with 100% branch coverage but nothing ran it: add it to
the CLAUDE.md test/lint/coverage commands, the shared-gates-tests workflow
(--cov-fail-under=100, same shape as repo_contract/coverage_gaps) and the
test_changed.sh fallback gate list.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NBScELfMUiieK9ManFbS3Y
New agent_ready/ package (dist and live-url modes, 100% branch coverage)
checks robots, sitemap, llms.txt, ai-catalog, 404, no-JS text, markdown
mirrors and JSON-LD. repo_contract gains web-repo detection and installs the
agent-ready pre-commit hook, CI workflow and check_agent_ready.sh shim;
scripts/check_agent_ready.sh is the shared entry point.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NBScELfMUiieK9ManFbS3Y
toolchain:python's "latest" was platform.python_version(), so the verdict
depended on the machine running the gate: setup-python "3.x" gave kuhyx/spellcode
3.15.0 on one runner and 3.14.8 on another, making its transitive:pyodide
allowlist entry dead (exit 2) on one and required (exit 1) on the other.
Now registries/toolchain.python_latest() reads python.org's release API
(is_published, pre_release=false), keeps only exact "Python X.Y.Z" names (the
feed also lists "Python install manager 26.4"), and takes the newest stable.
The never_cached() carve-out is gone: it existed because a cached
*interpreter* version from an older runner image (3.13.15) was served to a
3.14.7 job in screen-locker's CI. A registry answer is safe to cache, so
python now goes through the normal TTL cache, prefetch, and offline fallback
(expired cache -> DEGRADED; no cache -> unavailable), never the local
interpreter.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A4VrLZn9YjhPuoxTj54WKk
resolve()/base_for(leisure=) is now tri-state: None (default) decides by
the switch file on `day`, True forces leisure pricing (the shadow budget),
False forces the old base and cap. A consumer billing on the 06:00 gaming
day passes leisure_active(gaming_day), so 00:00-05:59 on the go-live date
no longer prices the eve with the +2 h leisure base/cap.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dsh4Znwh2KnC7q6vbGpX1N
Contract C6 of the leisure rollout (leisure-watch TODO-leisure-rollout.md).
leisure_from(path=LEISURE_FROM_PATH) -> date reads the switch file afresh on
every call (O_RDONLY|O_NONBLOCK|O_NOFOLLOW, fstat S_ISREG, <= 64 bytes,
strict YYYY-MM-DD with one optional newline); any failure is the parked
LEISURE_FROM. Never raises. leisure_active(day, path=...) = day >=
leisure_from() is the one way consumers ask "is Phase 2 on?".
gaming_base_minutes_for / gaming_ceiling_minutes_for / base_for / resolve
use it; leisure=True still forces (now a boolean, so a file date later than
2027-01-01 cannot drop the shadow budget back to 5 h). resolve reads the file
once, so base and cap never disagree mid-call. Tests redirect the path via
an autouse fixture and never touch the real /etc file.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dsh4Znwh2KnC7q6vbGpX1N
A `pkg @ git+https://github.com/<owner>/<repo>@pkg-vX.Y.Z` pin outside
kuhyx/utils was asked of utils, which has no such tag, so
steam-backlog-enforcer's leisure-watch pin was undeterminable (exit 3).
Foreign pins are now named `<host>/<owner>/<repo>:<pkg>`; utils pins keep
their bare names (allowlist labels and cache keys unchanged).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dsh4Znwh2KnC7q6vbGpX1N
A live claim was not "input is locked": it is published before the window
grabs, and leetcode-guard's study mode releases the grab while the claim
lives on. leisure-watch (contract C2) must pause shadow billing only while a
grab is really held.
- Claim.grab_held (always serialised; missing key from an old writer reads
as grab != "none"). Arbitration still uses the declared grab only.
- Arbiter.set_grab_held(held=) rewrites the claim and holder copies in place
through the owner's flocked fd (O_APPEND cleared, pwrite at 0, padded so
the file never shrinks) -- same inode, flock liveness untouched. Never raises.
- Set true only after grab_set/grab_set_global succeeded; RecoveryLoop
reports each tick's real answer (a stolen, un-retaken grab reads false).
LockWindow.report_grab_held(held=) is the hook for deliberate releases.
- gatelock.held_grabs(claims_dir=None): live claims with grab_held and a real
grab. Root callers pass /run/user/1000/gatelock/claims.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dsh4Znwh2KnC7q6vbGpX1N
Contract C1 of the leisure rollout (leisure-watch TODO-leisure-rollout.md).
New _leisure.py: LEISURE_FROM = 2027-01-01 (parked; Phase 2 sets it to the
day after the shadow week), gaming_base_minutes_for(day) (7 h from it, else
5 h) and gaming_ceiling_minutes_for(day) (10 h / 8 h). base_for and resolve
use them; leisure=True prices any day with the leisure base and cap (the
shadow budget). Pay, penalties, shutdown and the sick rule untouched; every
day before LEISURE_FROM resolves byte-identically (full day still 480).
resolve() now has six parameters (the contract fixes the signature), so
ruff/pylint max-args is 6; the term loop moved into _term() for
too-many-locals. test_ladder's shutdown-defaults-to-today test compared to
a literal that broke when UNIFIED_FROM passed; it now compares to today.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dsh4Znwh2KnC7q6vbGpX1N
Code-generated sound effects for spellcode (no AI audio, no sfxr port):
seeded white/pink noise and grain gates (noise), a phase-accumulating
oscillator over glide/vibrato/arpeggio/jump/retrigger pitch curves plus
ring mod (sweep), exponential and punch envelopes (envelopes), resonant
RBJ low/high/band-pass with a moving cutoff (biquad), and an sfxr-style
layered parameter set rendered at 44.1 kHz by default (sfx).
Presets: cast/hit/impact for arcane, fire, water, lightning, earth and
wind, plus dodge, parry_normal, parry_perfect and fizzle, following R1's
per-element recipes (arcane's is new). `python -m music_theory sfx --out
DIR` writes 16-bit mono WAVs and a SHA-256 manifest; manifest.entry now
takes any Recordable (Recipe or Sfx). analysis gains band_energy and
spectral_slope, which the tests use as the oracle.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016o4CqWFZwpfiDj8csAPgLa
They were only reachable through test_changed.sh. Add them to the
shared-gates workflow (with pre-commit, and apt PyYAML for the
/usr/bin/python3 hook-id helper) and to CLAUDE.md's test command.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PAhbNszs5JKdtnD4uBi3xp
`full` ran every manual-stage hook in one pre-commit call, which takes
over 60 s in screen-locker and leetcode-guard. `step <hook-id>` runs one
manual hook on --all-files and records "<tree> <id>" in
$GIT_DIR/gate-steps (per-worktree); the step that completes the set
writes the stamp exactly as `write` does. Dirty trees and non-manual ids
are refused with exit 2. Hook ids come from scripts/gate_stamp_hooks.py
(PyYAML, /usr/bin/python3), which treats an absent default_stages as
"every stage", matching pre-commit, so the steps cover what `full` runs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PAhbNszs5JKdtnD4uBi3xp
sync() reads the pool, waits on the network, then writes its merged copy
back without a lock. A second write landing inside that window was
silently overwritten. Marking day A and then clearing day B within a
second or two lost the clear, so B stayed a sick day.
Every background sync now holds a module lock, and each write tries it
without blocking. A click that lands mid-sync shows "Syncing -- try again
in a moment." in the panel and calls on_saved with nothing. The phone's
picker avoids the same race by making the grid inert while it syncs.
The red-first test pins the lost write with a sync stub that reads, waits,
then writes back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PAhbNszs5JKdtnD4uBi3xp
`sickdays gui` opens a Tk window modelled on the phone's Sick days screen:
a Monday-first month calendar (cat-2 fill, a dot for kept lockers, past
days faded and inert), a five-switch panel for the clicked day, and the
"Marked sick" list with "Kept: ..." lines. A free day is refused with the
phone's wording. Each write starts sync_quietly() on a non-daemon thread.
The window class is "sickdays" and Escape closes it.
sickdays.gui.KeepPanel is the panel on its own, a plain tk.Frame with no
root, mainloop, grab or wm_* call, so screen-locker can embed it under its
lock-screen grab. `import sickdays` and the rest of the CLI never import
tkinter: the CLI loads the gui module by name only for `gui`, and a
subprocess test checks this.
The tests drive real Tk on a withdrawn root, and CI now runs them under
xvfb-run, as gatelock's CI does. Adds a sickdays.desktop launcher, which
install.sh installs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PAhbNszs5JKdtnD4uBi3xp
A sick day may keep some lockers active: a locker stands down iff the day
is sick and the locker is not in that day's keep set. New record field
`keep` = [sorted, de-duplicated locker ids, hlc].
- add/remove write state, actor and keep on one tick (keep=[] on a clear),
so a stale keep cannot survive a clear + re-add.
- set_kept_lockers re-stamps only keep and actor; state keeps its clock, so
a keep edit never re-asserts sick over a concurrent clear. Ticks are
issued after every clock already on the record (clock-skew safe).
- Reading: no keep (0.1.0) or keep older than state -> keep nothing;
malformed keep -> every locker kept (fail closed, logged); unknown ids
ignored.
- API: stands_down, kept_lockers, set_kept_lockers, add_sick_day(keep=),
SickDay.keep, LOCKER_* / LOCKERS. is_sick_day unchanged.
- CLI: add --keep, keep DAY [LOCKER ...], list shows keep, check --locker.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PAhbNszs5JKdtnD4uBi3xp
Also fixes a ~5% flake in the hmac round-trip test (freedays and sickdays):
load_key strips whitespace, so a random key edged with a whitespace byte
never compared equal to the raw bytes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Xfj1sxKH9VBPqpcAni8Jry
sickdays: per-date pool, no cap or reason, add/remove today or later only,
import_legacy for the migration. LWW state flip per date (no tombstones),
Firebase RTDB sickdays-sync/devices/<uuid>/sick_days.json, HMAC audit trail.
Readers fail closed (unreadable pool = not sick, logged); writers refuse a
corrupt pool. Free-day cross-check via a lazily imported freedays predicate.
earned_time: resolve(..., sick=True) floors the final shutdown at 20:00
(capped at the ceiling); gaming, terms and base unchanged. Result dataclasses
moved to _resolution.py for the line cap.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Xfj1sxKH9VBPqpcAni8Jry
From TUTOR_QUOTA_FROM (2026-10-12) credit_units scales the tutor's deduped
day minute sum to min(60, minutes * 60 // quota); reward and penalty are
unchanged, every quota <= 60 so the switch is raise-only. Earlier days pay
one unit per minute as before. WORKDAYS is a literal copy of freedays'
(importing freedays pulls crdt-sync; earned_time is stdlib only), guarded by
a drift test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01E57caQF4KXwEvqEnKtJiHM
Rows carry no earner name, so only the ledger path moves; the tutor copies
its data dir (old rows included). Pre-cutover AUTOMATION (anki quota),
LADDER and the waiver are unchanged. Same minutes, cap, ladder, TUTOR_FROM
and confirmed_on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YADHKXzZhhePzQ2rNoXFcK
AUTOMATION_TUTOR: 1 gaming min per unit (minute), max_units 60; TUTOR_LADDER
automation Rung(1, extra=1), rungs still sum 250 (23:00 / 18:50).
credit_units sums each row's units (ROW_UNITS: tutor detail.minutes, a row
without it = 15); a repeated entry_id pays once at its smallest count; an
invalid minutes field is refused by tutor_match. An all-block day resolves
exactly as 0.7.0.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N9cLaZkZaH7Zi2K63uWuoB
teatr-pw and teatr-pw-pytania (created ad hoc 2026-10-09) had no shim, hook or
'## Commands' section: the gate was only ever rolled out by hand. The
installer writes shim + pre-commit hook + CI workflow and bootstraps the
Commands section and scripts/test_changed.sh, exiting 1 while <FILL IN>
remains.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TjyHmRrjZssaxXDWcGuCeQ
A repo with `.dep-freshness-no-workflow` at its root keeps the local pieces
(delegate script, pre-commit hook) but the installer no longer creates or
rewrites `.github/workflows/dependency-freshness.yml` there. teatr-pw
deleted its workflows on purpose (no GitHub Actions minutes on a private
repo; its pre-push hook runs `--all --strict`), and without this every
fleet sweep would put the workflow back.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01H8tUQzqQmiidbg8ZXmeQAg
TUTOR_FROM = 2026-10-10 (the day after the deploy). The automation tutor
keeps confirmed_on=None, so it stays "new".
Ladder floor: a rung comes off the floor only for a pure bonus (no
penalty_from) or an earner whose penalty has started -- the same
_penalised() rule (first_credits -> penalty_start, confirmed_on fail-closed
fallback, penalty_from without first_credits) the gaming base uses. Before
this, a never-credited tutor would have lowered 2026-10-10's floor by its
full 50-minute rung (19:50 -> 19:00) even with first_credits. Maturity can
only raise time: new property test against the 0.6.0 floor formula.
ruff 0.17.0 (dependency-freshness): CPY001 ignored -- personal repo, no
copyright policy.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WdfmsGMht9ypyvxGPmNF1V
CI's earned_time pre-commit flagged tests/_maturity_rows.py: the local
--all-files run predated git tracking of the new files. Same exclude as
steam-backlog-enforcer's tests/_*.py helpers. No package code changes.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WdfmsGMht9ypyvxGPmNF1V
How far to trust a gate is computed from its signed ledger: new / maturing /
mature (>= 14 verified credit days, first >= 21 days old), with a read-only
CLI table. Only signed, non-manual, paying credits count; a gate's own
re-evaluation grant (detail.grant_of, book-guard bonus: rows) counts at
amount 0.
Earner.confirmed_on (hand-set) is new. penalty_start = max(penalty_from, day
after first real credit, day after confirmed_on); resolve/base_for take an
optional first_credits map and can only delay a penalty, never add one.
Omitted, every day resolves exactly as in 0.5.0.
Fail closed: a gate with confirmed_on and no credit on record (unreadable
key/ledger, deleted or emptied ledger) falls back to max(penalty_from, day
after confirmed_on), so deleting a ledger cannot lift a penalty.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WdfmsGMht9ypyvxGPmNF1V
freedays 0.3.0 pinned crdt-sync v0.9.1 exactly, so any consumer that
moved to crdt-sync 0.9.2 (screen-locker) could not resolve both in a
clean venv. No code change.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018RHwVPyiiQMPh4M3JQwQku
**FileCredentialStore.save wrote every save through one fixed `<cache>.tmp`.**
Two processes share a credential cache (a long-running stream and a periodic
sync), so one could rename the other's half-written temp file into place and
leave a truncated refresh token behind. Each save now gets its own mkstemp
file in the same directory -- created 0600 from the outset, so the token is
never world-readable -- and replaces the cache atomically. A failed write
unlinks its temp file and leaves the previous cache untouched.
Tests cover: no temp file left after a save, distinct temp names per write,
and an interrupted write keeping the old credentials with no leaked temp.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018RHwVPyiiQMPh4M3JQwQku
~/src/daily-limits' user timer publishes today's limits for its phone app
and answers its requests, with a Firebase session of its own.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AcxgxMaQHMSGWzLRpNSkuP
From LADDER_FROM (2026-10-10) shutdown is read down from sleep: ceiling =
WAKE_MINUTES (07:00) - 8h = 23:00, floor = ceiling - every earner's rung
(workout 110, LeetCode 50, reading 30, Anki 25, Automation 25) = 19:00.
A second workout earns nothing more. Earlier days resolve exactly as in
0.3.0 (frozen 23:00 ceiling); gaming is unchanged for every input.
WORKOUT gets a signed ledger (~/.local/share/workout_locker/ledger.json,
written by screen-locker) and workout_match; a rest_day row counts only if
declared before its day began. New API: first_credit_at, credit_units,
day_window, shutdown_minutes_for, extra_shutdown_minutes_for,
shutdown_ceiling_for, on_ladder, LADDER/Rung, LADDER_FROM, WAKE_MINUTES.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QrjUTo1yhgAsUUasdw4bpG
subosito/flutter-action's cache is the whole SDK, 1.6 GB: ~38 s to restore
on every Flutter job even on a hit, which alone kept them over a minute.
This composite action caches a depth-1 clone of the release tag with only
the host and web artifacts (393 MB zstd'd); verified locally that test,
analyze --fatal-infos and build web run on it without lazy downloads.
Release APK builds keep subosito, which they need for the Android engines.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FBbkN4ZB4aUPJPzR2prRan
screen-locker and leetcode-guard each carried a near-identical copy of the
reader for wake-alarm's signed morning_session.json; book-guard is about to
need it too. The policy stays in wake-alarm -- this only answers "signed,
dated today, now < exempt_until". Path and retry budget are arguments so
each consumer's test redirect keeps working.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FBbkN4ZB4aUPJPzR2prRan
blocker split blocked_by on the first "@", so "transitive:@scope/name@1.2.3"
read the package name as "" and left "scope/name@1.2.3" stuck to the
version -- the format check then rejected the entry outright. rpartition
splits on the last "@" instead, which is always the version separator
even when the package name carries its own leading "@".
Found adding a typescript-eslint entry (blocked by its own
@typescript-eslint/scope-manager not having published 8.70.1 yet) in
signal-bot.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014Zg7ULrGyB1gMcKC1mkbLb
Two forks per path (`git ls-files --error-unmatch`, `grep -qiE <<<`) made
the gate a 5 s step on a 2053-file repo. The tracked set now comes from a
single `git ls-files -z -- "$@"` and the extension match is bash's own
`=~` on the lower-cased path (same case-insensitive semantics as `grep -i`).
0.2 s on the same tree; verdicts unchanged (blocked / allowlisted /
untracked-ignored checked on a throwaway repo).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016P4KZw1E2sXJjYqfcZWf1x
Three findings from proving the JitPack fix on the TachiyomiSY gate:
- The first repository with a stable answer used to win. Central still
carries com.github.ben-manes:gradle-versions-plugin 0.11.1 (2016) while
the plugin portal has 0.54.0, so a 0.51.0 pin graded "current". Every
repository is now asked and the newest stable across the pooled lists is
the answer.
- ~900 lookups, eight at a time, each resolving the registry host (A+AAAA)
per connection: the LAN router's forwarder started dropping queries for
every client on the LAN and the gate degraded itself. socket.getaddrinfo
is memoised per (host, port, hints) for the run, and the per-host probe
runs under a lock so eight workers make one probe, not eight.
- A resolver EAI_AGAIN ("temporary failure") during the probe marked the
registry unreachable for the whole run. The probe retries it
HTTP_ATTEMPTS times; a permanent failure still fails immediately.
Both retry loops lost their unreachable `return` sentinel (http.py is at
100% branch now); the resolver tests live in test_http_resolver.py to keep
test_http.py under the 250-line cap.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016P4KZw1E2sXJjYqfcZWf1x
Since ~2026-09-14 https://jitpack.io/.../maven-metadata.xml answers HTTP 401
"No access token" (artifact POM/AAR downloads still 200), which turned every
com.github.* coordinate into "undeterminable" (exit 3) in --strict CI.
JitPack is dropped from MAVEN_REPOS. A com.github.<owner>:<repo> name that no
HTTP repository knows is now answered from the GitHub repo's tags via
`git ls-remote --tags --refs` -- JitPack's version IS the tag name, so the
tags are compared as they are (v prefix included). The tag list goes through
the same stable-first selector as maven-metadata, so a newer pre-release tag
does not win. Not an allowlist entry, not a token.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016P4KZw1E2sXJjYqfcZWf1x
The packageManager field was bumped after the last install, so CI's
--frozen-lockfile refused it (ERR_PNPM_FROZEN_LOCKFILE_WITH_OUTDATED_LOCKFILE).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DnAYyNH1Sm4iJNidfYbrZJ
The ruff bump surfaced one ANN001 in test_window.py. ts-core / web-ui /
ts-config workflows gain workflow_dispatch so a workflow-only change (like the
last commit, which touched no ts_core/** path and therefore never ran) can be
re-verified on demand.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DnAYyNH1Sm4iJNidfYbrZJ
ts_core and web_ui depend on `link:../ts_config`. pnpm links the directory
and installs nothing inside it, and a flat config imports its plugins from
the preset's own node_modules, so CI died on "Cannot find package
'@eslint/js'" while the local tree (where ts_config had been installed by
hand) passed. Both workflows now `pnpm install` in ts_config first, and
trigger on ts_config/** too so a preset change re-verifies its consumers.
The same push turned the dependency-freshness job red on five pins that had
aged underneath it (ruff 0.16.8 in four requirements.txt,
flutter_secure_storage 11.2.0 in crdt_sync_flutter); bumped, pubspec.lock is
ignored in that package so nothing else moves.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DnAYyNH1Sm4iJNidfYbrZJ
The two shared packages are the first repos on the preset (link:../ts_config
inside the monorepo; consumers outside use the git tag). Each eslint.config.js
is now the defineConfig() call plus one override with its reason: a public
API name (`nextChance`, `fuzzyMatch`) the boolean-prefix rule would rename,
which is a breaking change for the three repos importing it.
Everything else the preset flagged is fixed in code, not disabled:
key order (perfectionist), hex casing, sort compare functions in tests,
Iterator#map().toArray(), explicit returns on multiline arrows.
web_ui: react-refresh/only-export-components was a real finding --
range-slider.tsx exported `steppedValue` next to the component, which breaks
Vite HMR for the module. The keyboard contract moves to stepped-value.ts as
`steppedValue` + a new pure `keyTarget(key, values, current)`, which also
resolves unicorn/no-late-event-control: the handler now decides the target
first and calls preventDefault() only when it owns the key. Its tests move
with it (the 250-line gate caught the combined file). Minor bump for the new
export. dist/ rebuilt in both packages.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DnAYyNH1Sm4iJNidfYbrZJ
Surfaced by making ts_core and web_ui the first consumers:
- base({ project }) uses an explicit tsconfig list instead of the project
service. Both packages exclude tests from tsconfig.json so they never land
in dist/, and widen the include in tsconfig.lint.json for lint only; the
project service would have reported every test file as "not found".
- unicorn/no-barrel-files is off for src/index.ts: a package entry point IS
a barrel, that is its job. Stays on everywhere else.
- unicorn/no-keyword-prefix is off in the React layer: `className` is
React's prop name and fires on every styled element.
- pnpm-workspace.yaml: allowBuilds for unrs-resolver set to false explicitly
(pnpm 12 had written a placeholder); the resolver's prebuilt binding is
used, no install script runs.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DnAYyNH1Sm4iJNidfYbrZJ
Evil Martians' "ten anti-AI-slop moves" as a package: strict tsconfig base
(noUncheckedIndexedAccess, exactOptionalPropertyTypes), type-checked
typescript-eslint strict+stylistic, unicorn/sonarjs/perfectionist, behaviour
rules for tests (expect-expect, no-focused-tests), React (hooks, @eslint-react,
jsx-a11y strict, you-might-not-need-an-effect), Playwright, and layer
boundaries via eslint-plugin-boundaries 7 with the article's default policies.
Knip / jscpd / Stryker bases; Stryker carries no break threshold on purpose.
Plugins are dependencies, not devDependencies: a flat config imports them from
this package's own node_modules, so a consumer installing from the git tag
gets them or the lint cannot run. No build step, so the tag needs no prepare.
test/fixture is a mini repo with one planted violation per rule family;
test/expected.js names the rule that must catch each, and the suite runs the
real CLI over it (in a child process -- v8 coverage merges natively-loaded
and transformed copies of the same module into "uncovered"). 35 tests, 100%
branch coverage on the factories.
eslint/fast.js is the type-free subset for a per-edit hook: 0.52 s cold on
one file, measured against 1.28 s for base() without types -- importing a
plugin is the cost, so it loads none, nor base.js which imports them.
Probed under ESLint 10: jsx-a11y 6.10.2 (peer says <=9) and boundaries 7.2.0
both run; the fixture pins that.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DnAYyNH1Sm4iJNidfYbrZJ
QEMU user-mode networking (libslirp 4.9.4 / qemu 11.1.1) freezes 2-4 of 5
concurrent 60 MB streams at ~10 MB until they time out, while one stream
does 45 MB/s. pacman's ParallelDownloads = 5 hit this on every guest
upgrade: a 90-package -Syu crawled at "500 KB/s" for minutes and then
finished in 12 s once set to 1. Window scaling and rmem do not help; a
tap/bridge NIC would but needs root, which vmbox avoids by design.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SS8M5wk88nJTXxDWz61KKk
The golden image logs in with agetty+startx and has no RGB controllers,
no NVIDIA card and no sound card, so the night-lockdown scripts — which
mask/stop lightdm.service and getty@.service and shell out to openrgb,
nvidia-smi and amixer — had nothing real to act on in a sandbox and
were being "verified" on the host instead.
- `vm lightdm <name>`: installs lightdm + lightdm-gtk-greeter into the
live overlay (pacman -Syu, not -S: the overlay's package database is
the image's and mirrors have rotated those versions out), configures
autologin -> i3 exactly like the host, turns tty1 back into a plain
getty, reboots the guest and proves lightdm.service is active with an
Xorg it owns. Never rebuilds the base: `vm build --force` re-seals a
new checksum while every existing sandbox still backs onto the old
file's name, which is silent qcow2 corruption.
- `vm shim <name> <tool> [exit[:N]|hang|sleep:S]`: a fake tool ahead of
the real one on PATH, every call logged with an epoch to
/var/log/vmbox-shim/<tool>.log so the host can assert what ran and in
what order. `hang` is the openrgb 1.0-2 that never returned on
2026-09-12 and defeated the curfew.
- The DISPLAY/XAUTHORITY resolver `vm run` sources becomes its own file,
guest/vmbox-x11.sh, installed by the image build and re-installed by
the lightdm switch; it now tests readability, not existence, and tries
~/.Xauthority before the startx serverauth glob, because lightdm's
Xorg -auth file is root-only.
- README: the worked example uses MIDNIGHT_SHUTDOWN_CONFIRM=y (the
installer refuses to run unattended now; `echo y |` no longer works),
and the "i2c RGB stays on the host" line is gone.
- bats: four host-side tests for shim.sh (exit, exit:N, hang, refusals).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015mLnUeoJ3mUwGRTowTUH6d
Dependency-freshness CI went red on the gatelock 0.8.2 push because
@types/node moved under it; bump both pins and lockfiles.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BxkZDHDmoVSxUT4w72xSiE
The 60 s deadline was the 2026-09-13 leetcode-guard miss with a smaller
number: a Persistent=true catch-up starts one second into a boot, before X
exists, and a bounded wait that expires exits the unit while systemd has
already stamped the slot as fired. No X server means the user cannot use
the machine either, so an unbounded wait is not a bypass. The wait now
re-states itself at WARNING every five minutes so a long wait never reads
as a healthy, silent unit. `timeout_s` stays as an opt-in for callers that
want a bound.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BxkZDHDmoVSxUT4w72xSiE
The palette-drift check parsed LockConfig's colour defaults; they live
on LockPalette in _theme.py since 0.8.0, so it read zero tokens and
failed closed as designed. The two new LockWindow handles had no test
reaching them, which CI's 100% bar caught and the local run did not.
gatelock/prompts/TODO-pylint-to-ten.md described this job with the
older design (raise the score, leave the gate); the gate is at 10.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HHCbkJWXHjtTUaqvnHRAqA
leetcode-guard's study mode has to stop the re-assertion loop and the
output-change watcher before it releases the grab, and could only reach
them as LockWindow._recovery / _detector -- attributes 0.8.0 folded into
the arming bundle, which broke that documented private dependency.
They are public read-only properties now, next to `surfaces`.
_window no longer re-exports the LockConfig family; it was the one
thing keeping it over the 250-line cap, and every importer now takes
those names from gatelock._config (or the package) directly.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HHCbkJWXHjtTUaqvnHRAqA
LockConfig was one 35-field record: the lock's behaviour, the grab
policy, nine colours, a type scale and a spacing scale, flat. It is
now the seven behavioural knobs plus four nested records --
GrabPolicy(kind, disable_vt, retry_ms, log_every, preempt_weaker_holder),
LockPalette, LockTypography and LockSpacing -- each overridden as a
unit. Breaking for every constructor that passed grab=/disable_vt=/
grab_retry_ms=/preempt_weaker_holder= or a colour, and for every read
of config.<colour>; hence the minor bump. screen-locker, leetcode-guard
and diet-guard follow in their own commits.
The two fail-open guards in the recovery loop carry per-line
suppressions: catching everything is the property they exist for, and
the check stays live for the next careless except. That was the last
finding, so the gate now says 10.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HHCbkJWXHjtTUaqvnHRAqA
The three import-error findings were the hook not having crdt-sync in
its own venv; it is now pinned there as in pyproject.toml. The one
source finding is the background sync's documented fail-open net, which
must catch everything by design, so it carries the one suppression.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HHCbkJWXHjtTUaqvnHRAqA
Same config move as crdt-sync (ignore-paths, init-hook, the three
implicit-booleaness disables the reference repos carry). The gate stays
at 8.0 for now: LockConfig's 35-field record and the two fail-open
guards in the recovery loop are still open.
Everything else pylint had left on source is gone. The optional
python-xlib dependency is loaded through one importlib helper instead
of two function-scope import statements, keeping the "not installed"
arm reachable for branch coverage; RecoveryLoop and LockWindow hold
their collaborator bundles rather than unpacking them into a second
set of attributes; the claim types are imported from _claims, where
they are defined, so _arbiter no longer needs an export list that
mirrored the package's; runtime imports sit above TYPE_CHECKING
blocks; the RandR Protocol stubs say which Xlib call they mirror.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HHCbkJWXHjtTUaqvnHRAqA
settings.gradle.kts can declare any number of catalogs via from(files());
TachiyomiSY has mihon.versions.toml and sy.versions.toml beside libs.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Yc4TiPjhbQLfH35QBPfrbP
The gate had no way to read a Gradle build, which left the two Kotlin
forks (tachiyomisy, syncyomi-src) outside the fleet-wide rule.
- parsers/gradle.py reads gradle/libs.versions.toml ([libraries] and
[plugins] in every shape: inline string, module + version.ref,
group/name/version, rich {strictly|require|prefer} tables) into
`maven:<group>:<artifact>` deps, pointing the fix at the [versions]
line when a ref is used; BOM-managed entries without a version are
skipped. gradle-wrapper.properties becomes `toolchain:gradle`.
- registries/maven.py asks Google Maven, Central and the plugin portal
in that order, reading the full <versions> list -- AndroidX writes
alphas into <latest>. Plugins resolve through their marker artifact.
- registries/toolchain.gradle_latest reads services.gradle.org.
- http.get_text shares the transport with get_json (XML bodies).
- Templates: the pre-commit `files:` and the CI cache key know the
catalog. 100% branch on every new file; fixtures only.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Yc4TiPjhbQLfH35QBPfrbP
All ten import-error findings were pylint failing to find the package
from the monorepo root; the init-hook puts the package dir on the path
the way pytest already does, and the tests-scope ignore-paths leaves
pylint grading source only (ruff keeps the tests). That alone is 10.00,
so the gate can say so.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HHCbkJWXHjtTUaqvnHRAqA
The 8.0 bar tolerated two points of debt and let the score regress
unnoticed. Adopting the fleet's source-only pylint scope needed
ignore-paths, not ignore = ["tests"]: ignore matches only the basename
of what pylint is handed, and pre-commit hands it explicit file paths,
so a directory name there never fired under the hook.
What the scope change left in source: runtime imports moved above the
TYPE_CHECKING blocks, the consent-flow CLI plumbing shared between
google_id_token and seed_session instead of duplicated, the token
exchange split out of google_id_token to stay under the 250-line cap,
and http.server's do_<METHOD> names admitted by method-rgx since the
stdlib dispatches on that exact spelling.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HHCbkJWXHjtTUaqvnHRAqA
The TachiyomiSY fork's first CI run died with AttributeError on the
runner's Python 3.12. Globs now go through a small tested translation
(** spans directories, * and ? stop at a slash) that runs on any 3.x.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Yc4TiPjhbQLfH35QBPfrbP
A fork of a third-party project can carry data that is code only by
extension (TachiyomiSY's 6000-line exh/eh/tags/*.kt string tables). The
reason belongs in that repo next to the data, so the gate now reads a
.file-length-exempt of '<glob> # <reason>' lines from its cwd. A glob
without a reason is a suppression and exits 2 instead of passing.
The package had no tests at all; it now has a suite at 100% line+branch,
wired into shared-gates-tests.yml with --cov-fail-under=100.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Yc4TiPjhbQLfH35QBPfrbP
A --fail-under=8.0 gate tolerates two points of lint debt permanently and
lets a score regress from 9.9 to 8.1 without failing anything.
Surveyed every repo under ~/src and ~/praca_magisterska: ten pylint hooks
across six repos still sit at 8.0, while five sites already run at 10 --
so this adopts an existing standard rather than inventing one. The prompt
carries the exact file:line of all ten, including the second pylint hook in
screen-locker that a naive sweep would miss.
Measured the sizing rather than guessing at it, because the raw numbers are
misleading. These configs run enable = "all" with disable = [], so the
findings are dominated by test-file idioms -- 147 redefined-outer-name in
gatelock is pytest fixtures, not defects. The repos already at 10 resolved
this and documented why: pylint grades source, ruff grades tests, via a
`tests` basename ignore. Applying that reference config plus the existing
init-hook takes the four utils packages from 8.54-9.59 to 9.42-9.93, leaving
~47 genuine source findings instead of ~600 -- and music_theory's 10
import-errors look like the missing init-hook rather than code.
The done-condition is two commands with exit codes, verified to report
exactly the ten outstanding sites today.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019kMvSMtsVMB3d2uoZ56pbp
Profiled the 23s hook step per hook. It is not evenly spread: mypy ~10s and
pylint ~9s are the whole of it, and the remaining 21 hooks share ~4s. Local
timings with a warm cache put mypy at 0.25s against 5.7s cold, so nearly all
of its CI cost is re-analysing the stdlib and third-party stubs from scratch
into a 19MB cache dir that was then thrown away.
Keyed on the config rather than on the sources. The mypy cache metadata
already records the interpreter, the flag set and a hash per source file, so
mypy re-checks whatever actually changed and a restored cache cannot let
stale code pass. The config hash is in the key regardless, so bumping the
mypy rev or editing its args busts it.
Deliberately not keyed on github.sha, which is the other common pattern:
that re-saves 19MB x3 on every commit and would evict the 68MB hook caches
under the 10GB LRU budget for the repo. A fixed key keeps the dependency
analysis, which is the part that costs the 10s.
pylint needs no change -- it already runs --jobs=4.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019kMvSMtsVMB3d2uoZ56pbp
Two follow-ups measured on the warm-cache run of d56b936, where the three
pre-commit jobs were 39-44s split as: 4s runner setup, 9s pip, 2s cache
restore, 23s hooks.
The 9s pip was pure waste. Every linting hook runs in its own isolated env
-- ruff, mypy, pylint, bandit and codespell each come from their own
repo+rev, mypy and pylint declaring their own additional_dependencies -- and
the only local hooks in the pre-commit stage need bash and python3. The
pytest-coverage hook, the one thing that would need the real dependency set,
is stages: [pre-push] and never runs here. Verified by running all three
configs green in a venv containing pre-commit and nothing else. The pin is
grepped out of requirements.txt so it cannot drift from the
dependency-freshness gate.
restore-keys is added because the exact key includes the config hash, so the
next hook-rev bump -- this session did one -- would have been a total miss
and reinstalled everything. A prefix hit rebuilds only the changed hook.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019kMvSMtsVMB3d2uoZ56pbp
Measured on 6154dc0: the three pre-commit jobs took 41-59s, of which 38-49s
was pre-commit installing hook environments and ~11s was an uncached pip.
The cache was not merely cold, it was miskeyed. pre-commit/action@v3.0.1
keys on hashFiles('.pre-commit-config.yaml') -- the repo ROOT's config --
but the hooks it installs come from the per-package config passed in
extra_args. Four different configs, one cache key: the three package jobs
raced to overwrite each other's hook sets, and editing a per-package config
never invalidated the key, so the restored cache was silently wrong rather
than missing. Three 68MB entries in the cache list are the symptom.
So the action is replaced with a direct `pre-commit run`, which keeps the
same --config/--all-files behaviour plus --show-diff-on-failure, wrapped in
an actions/cache keyed on the config that actually drives the install.
pre-commit is already an exact pin in each requirements.txt.
setup-python gains cache: pip keyed on each requirements.txt, for the test
jobs too.
The two crdt-sync workflows now list their own file in `paths`, which the
other four already did -- a workflow edit should re-run its own gate.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019kMvSMtsVMB3d2uoZ56pbp
The three Python packages pinned ruff twice at different versions:
requirements.txt at 0.16.7, .pre-commit-config.yaml at v0.15.2. Only the
older one ever ran, so `select = ["ALL"]` was enforced two minor versions
behind. Measured before aligning: v0.15.2 reports 0 findings, every 0.16.x
reports CPY001 on each file -- 70 crdt-sync, 23 freedays, 71 gatelock, and
no other rule.
Resolved by adding the notice rather than ignoring the rule, so the gate
keeps its teeth.
Two gatelock modules sat at exactly the 250-line cap, so a one-line header
tipped them to 251. Both are split along a real seam rather than shaved:
_widget_fills.py colour arithmetic over LockConfig, no tkinter import
_surfaces_focus.py selects among existing surfaces, never mutates them
The focus split keeps _surfaces.py's documented containment (it stays the
only module that creates, moves or destroys windows), and the static guard
in test_recovery_invariants.py globs _surfaces*.py, so the new sibling is
covered with no list to extend. widgets.py re-exports the fills helpers,
which keeps gatelock.widgets the single public surface its tests import.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019kMvSMtsVMB3d2uoZ56pbp
Latent drift, not a regression: the crdt-sync pre-commit workflow is path
filtered on crdt-sync/**, so it had not run since the file was written. The
requirements bump in 2c195cd triggered it for the first time and it failed
on I001 immediately.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019kMvSMtsVMB3d2uoZ56pbp
Clears the last of the dependency-freshness backlog, so the gate exits 0
with only the pre-existing npm:typescript exception.
vitest 5's text reporter no longer prints a per-file table; the summary and
coverage-final.json still carry the real numbers, and both packages hold
their 100% thresholds.
CI's pnpm/action-setup moves 10 -> 12 to match the regenerated
packageManager field, rather than relying on pnpm 10 self-switching to
12.4.1 at install time.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019kMvSMtsVMB3d2uoZ56pbp
New subproject porting Luke Haas's "Music Theory for Programmers" section
by section into typed Python: pitch (MIDI <-> Hz), tuning (harmonic
series, the Pythagorean comma, brute-forced "why twelve"), scales as step
arrays with modes as rotations, chords stacked in thirds with Roman
numerals and the V7 -> I resolution, named progressions, a numpy synth
(waveforms, ADSR, low-pass), a bar-exact sequencer, and a seeded Recipe ->
WAV composer. `degrade` is the same piece ruined by a 0..1 amount, so an
ending can be the opening destroyed on the same seed; filtering runs LAST
because filtering first measured brighter than the clean render.
17 demos (`python -m music_theory demo all`) each print the article's
numbers -- the comma is 1.013643, the tempered fifth is 0.113 % flat,
twelve is the first acceptable division -- and the FFT of the render, so
the claims are tested rather than asserted. 157 tests, 100 % branch
coverage, ruff ALL, mypy strict, pylint 10.
Consumed next by roadside-assistance for its three world-state beds.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VC17zymxkeiUcHrMvt8W5W
Run through Claude Code's `!` there is no stdin to paste into, and the
console dialog's copy button has already put the value on the clipboard.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019pBgaXSJKyY3DarYpDatDF
seed_session failed with invalid_client on 2026-09-11: the secret in
~/.config/crdt-sync/oauth_client_secret had worked on 2026-08-12 with the
same client id, so it was rotated in the console since. There is no API
for it (no gcloud command, none in the Firebase CLI), so the irreducible
step is copying it from the console page. refresh_oauth_secret.sh opens
that page, reads the paste without echo, rejects anything that is not a
GOCSPX- secret before writing, and runs the seed at once so a wrong paste
fails here rather than on the next timer tick.
xdg-open is detached from stdin: the spawned browser inherited the
terminal and swallowed the paste.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019pBgaXSJKyY3DarYpDatDF
sync_quietly exists so a background timer stays green when the pool
cannot be synced. It caught only SyncUnavailableError, which meant the
contract held only because of an upstream class hierarchy: today
FirebaseAuthError happens to subclass RemoteSyncError, and requests'
JSONDecodeError happens to subclass RequestException, so an expired
refresh token or a malformed payload does get wrapped and caught.
That is a coincidence, not a promise to this package -- and it is exactly
the kind that goes quiet the day upstream reorganises its exceptions. The
symptom would be a unit going red every 15 minutes, which is the outcome
this function's own docstring says the design exists to prevent.
It now also catches anything unexpected and logs it with a traceback, so
a real bug is loud in the journal while the timer stays green. The
narrow handler stays first and stays at debug level: offline is normal
on a laptop and must not fill the journal.
Two tests, neither of which the old suite could have failed on -- the
existing one monkeypatched get_client to raise the one exception already
known to be handled. One raises FirebaseAuthError *mid-sync* (a token
expiring during the tick rather than at client construction, which is the
realistic case per seed_session.py's own docstring), the other raises a
RuntimeError that no hierarchy covers.
111 tests, 100% branch coverage.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two things the pool needed to be real rather than merely correct.
**freedays no longer depends on gatelock.** It imported
gatelock.log_integrity for ~15 lines of HMAC, but importing gatelock runs
its __init__, which imports tkinter for the lock-window machinery. So a
headless policy library had acquired a hard dependency on a GUI toolkit,
and `import freedays` failed anywhere without Tk:
ImportError: libtk8.6.so: cannot open shared object file
A vmbox run caught it. The host masked it completely -- Tk is installed
here for the lockers -- and so did CI, whose own comment claimed "no
display needed" while the job would have needed libtk. freedays/_hmac.py
now does the signing with hashlib and hmac.
The format is deliberately unchanged (HMAC-SHA256 over compact
sorted-key JSON), so nothing has to be re-signed: all 8 entries already
in the real audit trail, written by gatelock's implementation, verify
against the new one. That was checked, not assumed.
This also dissolves the version diamond that shaped the app rollout:
freedays now pins only crdt-sync, so it can never conflict with whatever
gatelock tag an app carries.
**The pool now syncs on a schedule.** A `sync-quiet` subcommand plus a
systemd user timer every 15 minutes, because "it syncs" was otherwise
only true if you typed `freedays sync` yourself -- a day marked on the
phone would sit there. sync-quiet exits 0 when there is no network or no
credential, on purpose: a unit that goes red every time the laptop is
offline is a unit you learn to ignore, and a real bug still reaches the
journal through the library's logging.
install.sh gives freedays its own venv rather than borrowing an app's --
the pool is shared by five apps, so binding its CLI to one of them would
make "is freedays installed?" depend on something unrelated, and
uninstalling that app would take the command with it.
Verified in vmbox first (systemd), then on this host: the timer is
enabled and scheduled, and the service *finishes* rather than failing
when there is no remote.
seed_session.py gains freedays, without which the one command that can
give the pool a Firebase session would reject it outright (--app is
choices-validated) and the pool would stay local-only with nothing
saying why. That file was already exactly at the 250-line cap, so the
app list and its rationale move to tool/_seeded_apps.py.
freedays: 109 tests, 100% branch coverage. crdt-sync: 433 pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Five apps (diet-guard, wake-alarm, screen-locker, leetcode-guard,
home-guard) each enforce something daily, and none of them had a way to
say "not today". screen-locker was the only one with any notion of a day
off, and it had two overlapping ones: a budgeted sick day designed to be
hard to take, and an undocumented scheduled_skips.json that no other app
could see.
This is the second thing generalised, not the first. Free days are
pre-declared, carry no justification, allow consecutive runs, and are
capped only by an annual budget -- 35 per calendar year, resetting Jan 1.
Sick days stay in screen-locker exactly as they are: their rolling
windows (1/7d, 3/30d, 10/90d) exist precisely to forbid the consecutive
runs a holiday is made of, so merging the two policies would have had to
ruin one of them.
One global pool: a free day is free for every app at once, so there is
nothing per-app to mark and nothing per-app to forget.
The integration is one call, `freedays.is_free_day()`, which reads a
single local JSON file and never the network -- a gate fires from a
systemd timer and a hung sync must not be able to hold one open or shut.
It returns False for anything it cannot parse, so an unreadable pool
means the normal rules apply rather than everything switching off.
Two design points worth keeping:
- Un-marking flips a `state` field rather than tombstoning the record.
crdt-sync deletes are monotonic, so a tombstoned date could never be
marked free again.
- `consumed` is a sticky flag. Releasing a day that has not arrived
refunds it; releasing today does not, because those hours were already
unguarded. It is OR-ed against the stored value on every write so two
devices either side of midnight cannot hand a spent day back.
Timezone is local, decided in _day.today() and nowhere else. The apps
disagreed about this before: screen-locker's lock chain computes the day
in UTC while add_scheduled_skip.py uses local time, so a skip added near
midnight could name a date the chain never matched.
94 tests, 100% branch coverage, no suppressions.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Cargo's bare `"0.19.1"` is a caret range; its exact operator is a single
`=`. The shared EXACT regex only knew pip's `==`, so the one form that is
actually an exact pin in Cargo was reported as "not exact-pinned" against
itself (platformer/proto-bevy, 2026-09-06). Strip the operator in the Rust
parser; the constraint string keeps it so the report shows what is written.
Co-Authored-By: Claude Code <noreply@anthropic.com>
The fifth gatelock sibling (github.com/kuhyx/home-guard) calls
firebase_client_for("home_guard") from its systemd-run gate, so it needs
~/.config/home_guard/firebase_auth.json seeded like every other daemon.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
8.69.0 still declares `typescript: ">=4.8.4 <6.1.0"`; only the version in
the reason moves, so the text the gate prints on every run stays true.
Co-Authored-By: Claude Code <noreply@anthropic.com>
Every repo should give the same answer to "may this file be committed?",
and until now only testsAndMisc had a checker. signal-bot is public, had
three binary SQLite files in its history before anyone noticed, and
holds a private message database one `git add -f` away from the same
fate -- so .gitignore is evidently not the whole defence.
Two changes from the testsAndMisc original: sqlite3-wal and sqlite3-shm
are blocked by name, because a live SQLite database is three files and
blocking only the first blocks nothing useful; and the error message no
longer names a testsAndMisc-specific directory to move things to.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015d48t4qsQLBJnDwDeo2Nom
Both libraries had already moved `requires-python` to 3.14, but ruff, mypy and
pylint were still grading 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), a dead `timezone` import left behind by it, and ruff's
stable formatter applying PEP 758 -- `except (A, B):` becomes `except A, B:`.
That last one 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).
These two are libraries the apps install, so the floor matters more here than
anywhere else in the fleet -- every consumer already pins >=3.14.
Per project: ruff, ruff-format, mypy --strict and pylint clean, crdt-sync 432
tests and gatelock 389 tests, both at 100% line and branch coverage.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SfnJmxxsZ534AKk16rC2Gr
WUT_Computer_Science/Programming/TRAK/sightpy is the ray tracer from
rafael-fuente/Python-Raytracer (BSD 3-Clause). TRAK's assignment was photon
mapping built on top of it.
Verified against a fresh upstream clone with line endings normalised (the
copy in the archive is LF, upstream ships CRLF): 30 of 32 .py files are
byte-identical; sightpy/ray.py and sightpy/scene.py carry a 12- and 13-line
local patch that adds ray counters for the report.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0172dj34FDpao6wsix8ZuCzE
The coursework archive carries third-party content that the cap would
otherwise demand be split: the ASP.NET scaffolder's jQuery/Bootstrap dist
bundle (30,875 lines of JS, 43,027 of CSS -- ~98% of all JS and CSS in that
repo), stock Unity package/template assets, manufacturer-supplied Altium and
PADS component libraries, and the EOPSY course simulators.
The EOPSY entries are the load-bearing ones, because "provided by the
department" is normally a judgement call. It is not here: work/ is the
extraction target of `make setup`, and extracting the committed tarballs and
diffing shows 14/14 java files byte-identical to the originals --
ControlPanel.java (1852 lines) and Kernel.java (511) included. Nothing under
work/ is student-written, so splitting it would diverge the source from the
assignment it shipped with.
A data change to the existing extension point, so md_naming inherits it via
the shared file_length._tables import. The authoritative per-entry reasons
live in gates/vendored.txt in that repo; this tuple mirrors it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0172dj34FDpao6wsix8ZuCzE
`pinned:` is a required field precisely because an entry is an argument about
ONE version. Matching on the package name alone turns a narrow, reasoned hold
into a blanket exemption -- the fleet-wide typescript entry pins 6.0.3 for a
reason specific to 6.0.3, and dufs-cloud/web sat on `~5.8.3`, two majors
older for no stated reason, with the gate reporting it current.
An entry that exists but does not cover the pin in front of it still counts
as USED for the rot check. The hold is real; this repo is simply not on the
held version, which is a stale pin to fix rather than an allowlist line to
delete.
322 tests, 100% branch coverage.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
`.fvmrc` is a declaration nothing in this fleet reads. `subosito/flutter-action`
takes its own `flutter-version:`, and that pin is what the runner installs, so
the two drift in silence and the failure mode is the worst kind: green
locally, red in CI, with no diff to explain it.
kuhylog is the live case. Its runner is on Flutter **3.44.9**, which predates
the `@awaitNotRequired` annotation, so `unawaited(Navigator.push(...))` is
`unnecessary_unawaited` on 3.47.2 here and `discarded_futures` there -- the two
SDKs demand opposite code and CI went red on a commit that is clean locally.
The fleet is spread across 3.44.6, 3.44.8, 3.47.0 and 3.47.1 with no one
tracking it.
So a workflow's `flutter-version:` / `node-version:` is now a dependency like
any other, judged against the same latest stable. That needs no separate
"these must agree" rule: two values both required to equal latest already
agree. A channel name and a `${{ }}` expression are skipped -- one is not a
version, the other resolves at run time and guessing is worse than silence.
Found real drift on its first run: both TypeScript workflows here pinned
`node-version: 24`, a floating major that lets two runs of the same commit
install different Node. Now 24.20.0.
321 tests, 100% branch coverage on every production module.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
**The gate could not see the shared Python libraries at all.** They are
consumed as PEP 508 direct references --
`crdt-sync @ git+https://github.com/kuhyx/utils@crdt-sync-v0.5.1#subdirectory=crdt-sync`
-- which parsed as an unpinned PyPI package that does not exist on PyPI. That
answer is undeterminable, not stale, so wake-alarm sat on gatelock-v0.4.1
against v0.7.1 and crdt-sync-v0.5.1 against v0.9.0 with a green gate. They are
git-tag dependencies now, judged the same way the Dart libraries are.
A ref that is a branch or a SHA still parses as before: there is no version
in it to compare.
The line is found textually when the key indexer cannot help. A TOML
dependency array holds bare strings with no `key =`, so these findings would
otherwise report line 0 and make the fix a search.
Both libraries also had their dependency upgrades land after their newest tag
(81d2493), so anything moving to "latest tag" would have taken the pre-upgrade
constraints. Cut 0.9.1 and 0.7.2; version bump only.
315 tests, 100% branch coverage.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Every Flutter app here declares `flutter_launcher_icons` twice: as a dev
dependency, and as a TOP-LEVEL configuration block for the same tool. The line
lookup broke ties by indentation alone, so the top-level block won and the
gate told you to fix a line that carries no version at all -- lyricanki
reported `pubspec.yaml:64` while the caret being complained about is at 48.
The pubspec parser now says which top-level keys may hold a version, and a key
found outside them is only a fallback. `environment` is in that list on
purpose: without it the `sdk:` inside a `flutter: {sdk: flutter}` block would
outrank the SDK constraint it is meant to find.
Only the pubspec parser passes the hint. JSON and TOML indent their section
headers differently, so they keep the old behaviour unchanged.
310 tests, 100% branch coverage.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Most of these configs interleave local and remote hook repos -- diet-guard,
screen-locker and wake-alarm carry nine `- repo:` blocks each. Appending at
end of file lands the hook under whichever block happens to be last, and a
local hook definition handed to a remote repo fails with an error naming an
unknown id rather than the real problem. diet-guard's last block happens to be
`local`, so this would have worked by luck there and broken elsewhere.
A config with no local block now refuses rather than guessing.
308 tests, 100% branch coverage.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
The gate is three files per repo -- a delegate, a pre-commit hook entry and a
CI workflow -- and Phase 3 needs them in sixteen more repos. Hand-replication
is precisely how the 250-line cap ended up shipping under four different hook
ids, two of which were independent reimplementations rather than delegates, so
the copy is now a script and the templates have exactly one home.
scripts/install_dependency_freshness_gate.sh <repo> [--check] [--fvm]
Idempotent, and it rewrites a copy that has drifted from the template rather
than leaving it. `--check` reports without writing. `--fvm` pins the Flutter
SDK, and only in a repo that actually holds a pubspec.yaml -- an .fvmrc in a
Python repo declares a toolchain nothing there builds with. An unresolvable
SDK version leaves the pin alone and exits 1: guessing would pin every repo to
a number nobody chose.
The pre-commit block is APPENDED rather than parsed and re-emitted. Every one
of these configs carries comments explaining why its hooks exist, and a YAML
round-trip through the standard library drops all of them.
306 tests, 100% branch coverage on every production module.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
sync_settings_ui 0.4.1 and crdt_sync_flutter 0.3.1 both pinned the v0.11.0
tag, which predates crdt_sync_dart's exact pins -- so anything taking these
libraries resolved the shared package to its pre-upgrade constraints.
`flutter pub get` re-resolves cleanly in both; 88 tests green in
crdt_sync_flutter, 35 in sync_settings_ui, 321 in crdt_sync_dart.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Two things, because the first is what blocked the second from committing.
**The rot check was scope-blind.** pre-commit passes only the staged
manifests, so a run that sees no package.json cannot learn anything about
whether the fleet-wide typescript hold is still needed -- yet it concluded the
entry was dead and exited 2. Staging two pubspec.yaml files in utils was
enough to make the repo uncommittable. The dead-entry check now runs only
under --all, which is CI and the manual whole-repo run; a subset scan has seen
a subset by construction.
**crdt_sync_dart 0.11.1 and design_system 0.2.1.** Both had dependency
upgrades land after their newest tag, so a consumer moving to "latest tag"
would pick up the pre-upgrade constraints (crdt_sync_dart-v0.11.0 still
carries `crypto: ^3.0.7`, `http: ^1.6.0`, `lints: ^5.0.0`) while carrying its
own new exact pins -- the unsatisfiable-resolution case exact pins create.
Version bump only; the dependency work is already in 922fb36 and 704c67e.
The allowlist end-to-end tests moved to their own file for the 250-line cap,
and `run`/`canned` moved to conftest so both files share one definition.
`canned` is no longer autouse: the resolver tests patch `_fetch` themselves.
293 tests, 100% branch coverage on every production module.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Two things that were fine in utils and would not have been in the next forty
repos.
**The allowlist is now fleet-wide plus per-repo.** `typescript-eslint` blocks
TS 7 in EVERY repo that pins typescript -- eight of them and counting -- and a
repo-only allowlist meant hand-copying one entry into eight files that then rot
independently. That is the four-way fork of the 250-line cap, reproduced inside
the mechanism built to avoid it. `~/utils/dependency-freshness.allowlist.yaml`
is inherited by every repo; a repo entry for the same `ecosystem:package`
overrides it.
Inherited entries are exempt from the no-longer-needed check everywhere except
the repo that owns the file. Without that carve-out a fleet-wide hold on
typescript exits 2 in every Flutter app -- it excuses nothing there, which is
the normal case, not rot -- and those repos become uncommittable. Confirmed
against punchme, which now exits 0 while utils still reports the entry as in
use.
**`quarantine.py` bypassed the cache entirely.** It ran for every npm finding
and fetched the FULL packument -- 7MB for react, 16MB for typescript -- with no
TTL and no reuse, inside a pre-commit hook. One finding in utils, but the JS
repos ahead have 16 to 37 each. Now cached under the normal 6h TTL, keyed
`<name>@<dist-tag>`: the tag has to be part of the key rather than clamped
afterwards, because clamping returns the ceiling and the ceiling is frequently
the very release the quarantine is excluding.
**Both cache and shared allowlist are now redirected by AUTOUSE fixtures.** A
quarantine test had already seeded a fake answer for package "x" into the real
~/.cache/dep-freshness/registry.json, and the next run read its own pollution
back and passed for the wrong reason. Removed the entry; the fixture makes it
unrepeatable.
292 tests, 100% branch coverage on every module.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Deliberately the last step, not the first. All eleven manifests here are now
on latest stable -- four Dart libs, two Python packages, two TypeScript
packages -- so `--all --strict` exits 0 and this can be a gate rather than a
permanent red light. Installing it any earlier would have taught everyone to
ignore it.
Unlike every other repo's copy, this one runs `scripts/check_dependency_freshness.sh`
directly instead of checking kuhyx/utils out into `.utils`: this repo IS the
gate.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Finishes the last two of the eleven manifests in this repo. Exact-pins every
devDependency and takes three majors -- @testing-library/jest-dom 6 -> 7,
jsdom 29 -> 30, pnpm 10.13.1 -> 11.24.0 -- with lint and coverage green in
both (ts_core 42/42, web_ui 102/102, 100% branches).
**TypeScript stays at 6.0.3, and this is the campaign's first real
unsatisfiable conflict.** TS 7.0.2 is latest, but typescript-eslint 8.68.0 --
itself the latest release -- declares `typescript: ">=4.8.4 <6.1.0"` and throws
"does not support TS 7.0" as eslint loads its config. There is no 9.x. Taking
TS 7 would mean dropping every type-aware lint rule from both packages, so the
compiler waits for the linter. Recorded as a `transitive:` allowlist entry,
which is a predicate rather than a date: the gate re-checks the blocker each
run and the entry clears itself when upstream ships support. First use of the
allowlist, and the loud banner now prints on every run here.
**pnpm 11 also made the gate wrong, and the gate was fixed rather than pnpm.**
pnpm 11 quarantines packages published inside a 24h window -- a supply-chain
measure against the period when a compromised publish is live -- and refused
@testing-library/react 16.3.3, published 18h earlier. Reporting a repo as
behind a version its package manager will not install is a finding nobody can
act on: the hook fails, and so does the fix it asks for. `quarantine.py` now
narrows npm's "latest" to the newest release old enough to install, bounded
above by `dist-tags.latest` so a maintenance-line release never outranks the
tag. It fetches the FULL packument (7MB for react, 16MB for typescript) and so
runs only after a finding already exists -- a clean run pays nothing, which a
test asserts by making the fetch raise.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Two npm-specific corrections found by running the gate over web_ui.
**`peerDependencies` must not be exact-pinned.** The gate was reporting
`react: ">=19"` as "not exact-pinned", and acting on that would have been
actively harmful: a peer range declares what a CONSUMER may bring, so pinning
it forces every consumer of @kuhyx/web-ui onto one exact React. Peers are now
judged the way toolchain ranges are -- does the range still admit latest? --
and only reported when it does not.
**`packageManager` is a real dependency.** corepack enforces it, so
`pnpm@10.13.1` is an exact pin that nothing was checking. It is now read and
compared, which immediately surfaced that both TS packages sit on pnpm 10.13.1
against 11.24.0.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Sixteen `>=` floors become exact pins across the two Python packages, taking
five majors with them: mypy 1.13 -> 2.3.1, pylint 3.3 -> 4.0.7, pytest 8 -> 9,
pytest-cov 5 -> 7, bandit 1.7 -> 1.9.4, plus ruff 0.15 -> 0.16.5 and
requests/types-requests. A `>=` floor is satisfied by everything newer, so it
told a checker nothing about whether these repos were actually current.
Verified in a clean venv per repo, which is what both CI workflows do
(`pip install -r requirements.txt`): gatelock 389 tests at 100% coverage under
xvfb, crdt-sync 432 tests at 100%. No test change was needed for any of the
five majors.
**Open, and deliberately not in this commit:** `.pre-commit-config.yaml` pins
these same tools AGAIN via `rev:` (ruff v0.15.2, mypy v1.13.0, pylint v3.3.2),
independently of requirements.txt, and the gate does not read `rev:` pins at
all -- so the linters CI actually runs are still on the old majors. Bumping
them is a separate and much larger job: ruff 0.16 raises 71 findings here
(CPY001 alone accounts for many, since this repo selects ALL) and mypy 2.x
raises 60, mostly stricter implicit-re-export rules. One major per commit, and
that is several.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
`types-requests==2.33.0.20260712` was reported as "not exact-pinned" against
its own value: the regex capped the release segment at three components, and
PyPI stub packages date-stamp a fourth. PEP 440 places no such limit.
Found by running the gate over crdt-sync's newly pinned requirements.txt,
which is the only manifest in the fleet with a four-part pin.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Both were the oldest floors left in the fleet -- `http: ^1.2.0` while every
sibling was on ^1.6.0. Neither moved a resolved version: pubspec.lock is
untouched, so this is a declaration change, and analyze / 24 tests / 75-line
100% coverage are all unchanged.
This is the first real repo to exercise the three-way manifest / lockfile /
registry check, since it is the one library here that tracks its pubspec.lock
while its four siblings gitignore theirs. Both directions confirmed against a
copy of this repo: a lock resolving 1.5.9 under a manifest pinning 1.6.0 is
reported as "manifest and lockfile disagree", and staleness correctly takes
precedence over it when both apply.
Version 0.3.0 -> 0.3.1, tag cut. Four consumers pin it by git tag
(home_inventory, todo, diet-guard, wake-alarm) and move when their turn comes.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Every CI workflow clones kuhyx/utils into `.utils/` *inside* the repo being
checked, so `--all` walked utils' own manifests and failed the consuming repo
for staleness in a directory that repo does not own and cannot fix.
Caught by punchme's very first CI run: locally `--all --strict` exited 0, and
the same command on the runner reported twenty findings, all of them in
`.utils/web_ui/package.json` and `.utils/ts_core/package.json`. Exactly the
kind of local/CI divergence the gate exists to prevent, in the gate itself.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
Exact-pins every registry dependency (crypto 3.0.7, google_sign_in 7.2.0,
http 1.6.0, shared_preferences 2.5.5, url_launcher 6.3.2, uuid 4.6.0) and takes
flutter_secure_storage 10.3.1 -> 11.0.0, a major.
The major is safe here for a specific reason, not by assumption. 11.0.0 removes
the algorithms deprecated in v10 and makes data written by them unreadable --
but this package has been on ^10.x throughout, and v10 migrated that data
automatically. It also removes `encryptedSharedPreferences` and
`sharedPreferencesName` from `AndroidOptions`; keystore.dart constructs a bare
`FlutterSecureStorage()` and passes no AndroidOptions at all, so neither is
reachable from here. The remaining break is minSdk 24 / compileSdk 37, which
consuming apps inherit from `flutter.minSdkVersion` (Flutter 3.35 already
raised its own floor to 24) and which fails a Gradle build loudly if wrong.
Analyze clean, 88 tests green, coverage held at 202/202 (100%).
Also installs the dependency-freshness gate in pre-commit -- the first repo to
carry it. Pre-commit only, not CI: the other ten manifests here are not
upgraded yet, and `--all` in CI would be permanently red until they are. The
`files:` regex means a source-only commit makes zero network requests.
Proven in both directions through the hook itself, not just the CLI:
crdt_sync_flutter/pubspec.yaml passes, github_device_auth/pubspec.yaml fails
and names http ^1.2.0 and url_launcher ^6.3.0, and a .dart path is skipped
by the filter.
Version 0.2.1 -> 0.3.0; consumers move their git pins after the tag.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
A caret range's *floor* is the only comparable number it carries, and all four
libs sat at ^10.2.0 while 10.3.0 was current. The Q13 carve-out lets these lint
packages keep a caret; it does not let the floor rot.
Deliberately its own commit, separate from any dependency bump: a lint-package
major/minor lands new rules across four packages at once, and tangling that
with a resolution change would make an analyze cascade indistinguishable from a
version conflict. `flutter analyze` is clean in all four, and all four suites
pass unchanged (88 / 68 / 24 / 35).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
"Are we on the newest version?" was not mechanically answerable across these
repos: constraint styles had drifted per-ecosystem and a caret range is
"newest-compatible" by definition, which gives a checker nothing to compare.
This gate makes it answerable and then enforces the answer.
Checks two things at once -- that every dependency carries an exact pin, and
that the pin equals its registry's latest STABLE release -- across pub.dev,
PyPI, npm, crates.io, the Go proxy, kuhyx/utils git tags, and the Flutter /
Node / Python toolchains.
Three deliberate design points:
* Pre-releases are rejected per-ecosystem, because every ecosystem leaks them
differently and each leak fails GREEN. npm lets a maintainer point
`dist-tags.latest` at an rc; crates.io's `max_version` includes them; PyPI's
legacy `info.version` includes yanked releases. Unit-tested with fixtures,
since an end-to-end run cannot surface a wrong "up to date".
* Offline, the gate serves an expired cache entry and PASSES with a DEGRADED
banner (exit 3 only under --strict, which is CI). `--no-verify` is banned
here, so a hook that hard-failed offline would leave no way to commit at all.
* The allowlist has two entry classes. `blocked_by: transitive:<pkg>@<ver>` is
a predicate that clears itself when upstream moves and must NOT carry a
date; everything else is discretionary and requires `expires`, capped at 90
days. Exact pins plus latest-everywhere is provably unsatisfiable when a
dependency internally constrains a shared transitive, and expiring the
predicate class on a calendar would turn the gate red for a reason no commit
could fix. An entry with nothing left to excuse is itself an error.
One hook id, one shim, one shared implementation -- always delegating, so it
cannot fork the way the 250-line cap forked into four ids, two of which were
independent reimplementations.
256 tests, 100% branch coverage on every module of the gate. Also adds the
workflow that runs them, which incidentally gives md_naming's 71 tests their
first CI run.
Not yet installed in any repo: a repo that gets the gate before its deps are
bumped is instantly and permanently red, so the gate and the upgrade land
together, per repo.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CRykabMwPgDzFHcARoqpaK
connectWithPassword() had no try/catch around firebaseFactory(), unlike
connectWithGoogle() right next to it. A thrown FirebaseAuthError (wrong
password, 400 from identitytoolkit) escaped uncaught out of the button's
async onPressed, so _busy/_status never updated and the screen sat on
"Signing in..." forever with no error shown -- reproduced live via the
browser console (confirmed "Uncaught Error" immediately after the 400,
not a hung Future).
Also reports real progress stages ("Saving account...", "Signing in to
Firebase...") instead of one static string, clears the bad password on a
confirmed rejection (previously only the null-client branch did, letting a
background sync tick keep retrying it), and surfaces the actual failure
reason in the status text instead of a generic "rejected" message.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XATAXuMzpYggWyTBPGyiSN
The failure message named refactor_claude_todo.md, which the markdown
namespace migration renamed to TODO-file-length-250.md. A gate telling
you to read a file that no longer exists is worse than one that says
nothing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A47aX5DJwmA3QFn3pkEWi7
Markdown across every repo had accumulated a dozen ad-hoc names
(refactor_claude_todo.md x23, DESIGN_AUDIT_TODO.md, NEXT_SESSION,
HANDOFF, PLAN, SPEC) with nothing distinguishing a live task from a
finished record. An audit found docs asserting work was pending that
shipped months earlier, and one asserting completion while the work
was open.
Four namespaces now: README.md, CLAUDE*.md, DOCS*.md, TODO*.md. Task
files carry the literal "REMOVE ME AFTER FINISH" and are deleted when
the work lands.
Enforced bidirectionally -- a TODO without the marker fails, AND any
file carrying the marker that is not named TODO*.md fails. Without the
reverse rule a task can be renamed into the permanent namespace and
quietly live forever, which is the failure being fixed.
md_naming/ mirrors the file_length topology and imports its tables
(EXCLUDED_DIRS, THIRD_PARTY_REPOS, VENDORED_ANYWHERE, GENERATED_PATTERN)
rather than restating them, so the two gates cannot disagree about the
same file. 71 tests, 100% branch coverage.
stale.py warns on TODOs past 90 days but never fails: no machine can
prove a task is finished, so the honest mechanism is a nudge.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01A47aX5DJwmA3QFn3pkEWi7
Item 5 was recorded as "PASS (partial, reset check pending)". Now verified:
`chattr +i /etc/hosts` in a sandbox gives `----i-----------------`; after
`vm reset` the same file reads `----------------------` and accepts a write.
Worth having measured rather than assumed. Immutability is exactly what makes
the hosts guard hard to undo on a real machine -- `rm` fails EPERM -- so
"reset discards it wholesale" is the strongest form of the reset guarantee and
the one most worth being sure of.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wfxvsYxKM5xDXoDbkKi5u
Two fixes to DISCOVERY_TEST.md, both found by re-checking claims instead
of trusting them.
The .claude remote: `git branch -vv` + `ls-remote` confirm main DOES track
origin/main on kuhyx/claude-config-local, and `gh repo view` says that repo
is PRIVATE. The correction was right but stated the wrong load-bearing
fact -- what matters when telling a future session to push there is the
visibility, not merely that a remote exists. Now says so, and says to
re-check rather than trust the line.
The contamination: NEXT_SESSION_INSTALLER_FIX.md, added to the root of
~/testsAndMisc the same day, is about hosts/install.sh -- the exact file
the discovery prompt names -- and contains "Sandbox first, always ...
~/utils/vmbox". One `ls` at that repo root leaks the answer, which is a
likelier path than the ~/utils grep already noted, and this session created
it. Records it and offers the pacman wrapper as an uncontaminated
substitute prompt.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
The sandbox-first rule and the vmbox-testing skill only pay off if a
session with no memory of building them reaches for the sandbox on its
own. This is the test for that: a pasteable prompt (install the hosts
guard and confirm it enforces) that never mentions vmbox, sandboxes, VMs
or the skill, plus a Pass/Partial/Fail rubric and the follow-up.
Deliberately NOT executed by the session that wrote it -- that session
knew all about vmbox, so its behaviour would prove nothing, and a subagent
either inherits that context or gets the task framed by it.
Chose the hosts guard over the shutdown timer because a Fail should be
cheap to lose: it leaves an immutable /etc/hosts rather than a surprise
poweroff. The rubric carries the undo commands next to the Fail
definition so a failed run is not discovered weeks later.
Records that ~/.claude DOES have a remote (kuhyx/claude-config-local),
correcting an earlier note that claimed main there is remote-less; the
fix on a Fail is the skill's description: frontmatter first, then the
CLAUDE.md block, and both need committing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Replaces the handoff prompt with the next three pieces of work.
Job 1 lands the firmware guard for the bug the sandbox found: boot-repair
deletes the only bootable kernel on a BIOS/GRUB box because its "orphaned
kernel shadowing the ESP" heuristic never checks /sys/firmware/efi. The vmbox
guest IS such a box, so the fix is verifiable in the same place the bug was.
Job 2 writes the install_core_system.sh findings into a prompt in the repo
that owns that code. The two dead module paths resolve to ~/screen-locker and
~/steam-backlog-enforcer (checked on disk, not assumed), which makes the
installer cross-repo and is itself a question for kuhy rather than a guess.
Job 3 is a test of whether a fresh Claude discovers the sandbox rule on its
own. It is written, deliberately NOT executed: running it in the session that
built the tooling, or as a subagent briefed by that session, measures nothing.
The prompt says so up front, since it constrains the whole session.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
The previous commit skipped the poweroff settle when `ssh_rc != 255`, on the
reasoning that a successful poweroff kills the ssh connection and ssh reports
255. That holds for the ssh transport and for nothing else. `_run_transport`
falls back to the serial console whenever ssh does not answer -- and the
README calls serial "the transport of record for shutdown tests" -- where the
exit code comes from serial_exec.py and is 0 even while the machine is already
on its way down.
Result, reproduced: `VMBOX_TRANSPORT=serial vm run <vm> 'sudo systemctl
poweroff'` reported "still running" for a guest that had cleanly powered off,
with SHUTDOWN{guest:true} in events.jsonl and "reboot: Power down" in the
serial log. The one thing this tool exists to get right, silently wrong. My
five verdict re-tests all went through ssh, so none of them caught it.
Grepping the serial log for shutdown markers instead does not work either:
they land a second or two after the command returns, so the check loses the
race. The fix waits a short grace period (default 3s) for any sign of a stop
before committing to the 30s wait. Warm runs go from 0s to 3s; that is the
honest price of a correct verdict.
Re-tested: clean poweroff, reboot and panic on BOTH transports, plus
tests/destructive_demo.sh 7/7.
Also, per "a fix that needs a human to remember a flag is not a fix": the X
session wait is now automatic on the branch that actually boots the guest --
the only one that can race tty1's autologin -- instead of requiring
VMBOX_X_WAIT. Warm runs never reach that branch. The env var remains an
override. README numbers corrected to the measured 14s/3s and ~100s.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
`vm run` took 41s cold / 30s warm to run `true`. Both were vmbox waiting on
itself. Now 12s / 0s, with all five shutdown verdicts re-tested unchanged.
- _run_settle waited up to 30s for the kernel's "reboot: Power down" marker
UNCONDITIONALLY, including on the majority of runs that leave the guest up,
where that marker is never going to appear. It now settles only when the
guest has actually stopped, or when ssh returned 255 (what a successful
poweroff looks like from the host).
- The ssh readiness polls used a 5s ConnectTimeout. QEMU's user-mode
networking ACCEPTS the forwarded connection while the guest still has
nothing on :22, so each probe blocked for the full timeout instead of
failing fast: a guest reachable at ~10s was reported up at ~45s. Fixed in
both the sandbox and build paths, via an env override -- ssh honours the
FIRST value for an option, so a second -o could not have worked.
Making runs fast exposed a race: sshd now answers BEFORE tty1's autologin has
started Xorg, so the X environment could be missing on the first run after a
boot. `VMBOX_X_WAIT=<seconds>` waits for the session; it defaults to 0 so an
ordinary run never pays for a graphical session it is not using.
Build: the host's /var/cache/pacman/pkg is offered to the build guest
read-only (`readonly=on` AND an `ro` mount, because it is the real host cache)
as an extra CacheDir after the guest's own writable one, so a package the host
lacks still downloads rather than failing the build. This removes the 396 MiB
the build used to fetch. Host cache verified byte-for-byte identical across
four builds: same file count, same size, same mtime.
Build time itself moved only 103.8s -> 100.3s. Per-phase timings (also added
here) show why: 47s of it is the build guest's cloud-init first boot, and that
is NOT datasource probing -- `cloud-init analyze blame` totals 2.27s. Left
unexplained rather than guessed at; SESSION_RESULTS.md records the evidence.
Also: `vm scp`, which lib/ssh.sh could always do but the CLI never exposed,
and provision.sh split at the 250-line cap.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Testing boot_recovery produced a guest that cannot boot: boot-repair read
/boot/vmlinuz-linux and initramfs-linux.img on this BIOS/GRUB guest as
"orphaned kernel files shadowing the ESP" and deleted them, leaving the
sandbox at a GRUB prompt. That is a legitimate sandbox outcome -- destroying
the guest is what the tool is for -- but vmbox handled it by hanging.
Two independent unbounded waits, both fixed:
- vm_wait_ssh counted only its `sleep 2` calls, ignoring the up-to-5s
ConnectTimeout each failed probe spends inside ssh. A nominal 150s wait ran
past 400s (measured, then cut short by the harness). It now works to a wall
clock deadline: the same guest is now reported unreachable in 156s.
- _run_recover_rc reboots the stopped guest to read the recorded exit code,
and fell through to an unbounded serial console read: 180s for a prompt
that never comes plus 300s for a command that never runs, all AFTER the
verdict had already been printed. Recovery is best-effort by definition, so
both bounds are now short (60s/30s) and overridable.
serial_exec's timeouts become env-overridable to make that possible.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
README documented `bats tests/test_vmbox.bats` covering "name validation,
meta handling, serial rotation and the full verdict table", and the handoff
made it a done-condition, but the file did not exist. Written to that spec:
39 tests, no VM boot, well under a second.
VMBOX_HOME is redirected to a per-test mktemp dir, so the suite can create
and delete VM state without any chance of clobbering a live sandbox's overlay
or pidfile; one test asserts every path helper stays inside it.
The verdict cases encode the distinctions the design turns on rather than just
exercising the function: guest:true with a truncated serial log is DIRTY (3)
and not a clean poweroff, POWERDOWN alone proves nothing and must read as
"none", an unrecognised shutdown reason fails closed at 6 rather than passing
silently, and events.jsonl is read last-event-first so a previous boot's
poweroff is never mistaken for this run's outcome.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
The image ships Xorg + i3 on tty1 precisely so locker and i3 tests can run,
but `vm run` executes through `sh -c`, which reads no profile. Every X tool
therefore landed with no DISPLAY and no XAUTHORITY and failed with
"Could not determine i3 socket path" -- indistinguishable from "this sandbox
has no X server". testsAndMisc's i3-configuration installer failed that way;
with the environment resolved it exits 0 and i3-msg reports {"success":true}.
XAUTHORITY is the awkward half: startx generates /tmp/serverauth.XXXXXXXX
with a fresh random suffix every boot, so it cannot be hardcoded. Resolve it
from the running Xorg's own -auth argument, falling back to a newest-first
glob.
The `[ -f ]` guard on the source line is load-bearing, not defensive
dressing. In POSIX sh, `.` on a missing file is a SPECIAL BUILTIN error that
exits the shell outright, and `|| true` does NOT rescue it -- written that
way, every `vm run` against a base image predating this snippet silently
produced no output at all. Found by doing exactly that.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Three changes from kuhy's follow-up:
- An explicit standing instruction that when vmbox breaks the next session
FIXES it -- not works around it, not reports it as a limitation, not quietly
falls back to testing on the host. Bugs found by real use are the return on
having built this.
- Guidance to attach to an existing sandbox or create one as appropriate,
rather than tearing down and rebuilding per command (`vm reset` is ~2s).
- A build-speed job with measurements rather than guesses: the last build
pulled 197 packages / 396 MiB over the internet and spent 45s waiting on
cloud-init, while the host's 114 GB pacman cache already holds every one of
the packages checked. Exposing that cache read-only is the obvious first
move; trimming base-devel and shortening the cloud-init wait follow.
The smoke gate stays regardless -- a faster build that fails open would be a
bad trade.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Two jobs, deliberately ordered. First: run real linux_configuration work
through the sandbox (the 8 Arch-only CI tests, night-lockdown, boot_recovery,
the hosts guard) and fix vmbox wherever it breaks -- it has only ever been
driven by one session against a handful of scripts, so the ~1090-file surface
will find things.
Second, and only if the first says vmbox is trustworthy: the CLAUDE.md rule
that routes script testing here by default. A rule pointing at a flaky tool is
worse than no rule, so that ordering is the point. Four questions have to be
settled before writing it -- what triggers it, whether it replaces or precedes
the host run (see feedback-verify-real-deployment-path), the hardware escape
hatch, and keeping it short since CLAUDE.md is always-on context.
Carries forward the traps that cost real time this session: diagnose an
unreachable guest with `systemctl list-jobs`, never grep a bash -x trace for
the failure, and fix the image live over the serial console rather than
rebuilding once per hypothesis.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Captured a real 1280x800 PNG of a sandbox's i3 session straight from qemu's
framebuffer -- no VNC or SPICE viewer installed on the host. This is the path
locker and fullscreen-gate tests use, and it is safer than testing those on
the host, where a fullscreen locker can hijack input and steal focus.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
`vm reset` killed the VM and immediately recreated the overlay, without
waiting for the process to actually exit. The next boot then read a
half-detached file and failed with "sandbox did not become reachable" -- a
symptom that appears one step AFTER the cause and says nothing about it.
Reset itself was always correct (verified by hand: after
`rm -rf --no-preserve-root /etc /usr`, a reset guest reports
NAME="Arch Linux" and has /usr/bin/bash back). This only bit when a reset
followed a still-shutting-down VM closely enough to race it, which is exactly
what the destructive demo does.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Measured between two sandboxes: 3 packets transmitted, 3 received, 0% packet
loss, 0.23 ms average.
The previous commit called this broken. It was not: the dgram transport worked
all along, but the odd/even pairing computed the wrong peer port for even
indexes, so a VM pointed `remote.port` at a port nothing had bound. Both ends
came up, addresses were right, and zero frames crossed -- indistinguishable
from a dead transport, and nothing in any log said otherwise. The tell was
`ss -lunp`: the bound ports were 12349 and 12350 while the guest was sending
to 12348. The test then compounded it by using indexes 4 and 5, which belong
to DIFFERENT pairs (4 pairs with 3, 5 pairs with 6).
Pairing is (1,2), (3,4), (5,6)... Two sandboxes created back to back are wired
together, which is the common case.
mcast= remains genuinely unusable here and is documented as such so nobody
re-tries it: lo has no MULTICAST flag and there is no multicast route, so qemu
accepts it and drops every frame.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
The peer-address service read a `vmbox.peer=` kernel argument that the
launcher never passed -- the reader was written and the writer forgotten -- so
eth1 came up DOWN with no address on every guest. It now derives the address
from the NIC's own MAC (the host allocates 52:54:00:be:ef:<index>), which
removes the coupling entirely: nothing has to be passed in, so the launcher
and the guest cannot disagree about which index a VM has.
That fixed the addressing but NOT the connectivity. `mcast=` cannot work on
this host at all -- `lo` carries no MULTICAST flag and there is no multicast
route, so qemu accepts the option and silently drops every frame, which is the
same "flags accepted, nothing happens" trap as POWERDOWN. A listen=/connect=
socket hub over loopback TCP was tried next: the hub port listens, both
guests bring eth1 up with the right addresses, and ARP still fails.
Documented as a known limitation rather than papered over. Sandboxes work
fully on their own; two guests cannot talk to each other yet.
Also corrects the destructive demo: a running Linux SURVIVES `rm -rf /usr`
because processes hold their binaries through open file handles, so sshd keeps
answering. The honest proof of destruction is that the machine cannot boot
again, so the demo now reboots it and asserts it does not come back.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Ran kuhy's real setup_midnight_shutdown.sh inside a sandbox -- the installer
whose own output says "There is NO disable option - this is protected by a
monitor service". It created /etc/shutdown-schedule.conf under guard-lib's
immutable protection and enabled shutdown-timer-monitor.service,
day-specific-shutdown.timer and shutdown-timer-monitor-watchdog.timer.
The host was provably untouched afterwards: base image checksum OK, /etc/hosts
still carrying its +i flag, and the host's own shutdown-schedule.conf still
stamped 11:21 that morning rather than the 21:19 when the guest wrote its copy.
Two properties of that installer are now documented, both found by running it:
it is interactive (`read -r -p 'Do you want to proceed?'`, so pipe `echo y |`
since vm run closes stdin deliberately), and it is subcommand-driven, so a
bare invocation prints usage and changes nothing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
9p exports a directory TREE and does not follow a symlink pointing out of it.
The share directory held `testsAndMisc -> /home/kuhy/testsAndMisc`, so the
guest mounted /mnt/hostrepo successfully, listed the symlink, and then got
"No such file or directory" following it -- the target path does not exist
inside the guest. Measured in a live sandbox, not assumed.
`vm share <path>` now bind-mounts the directory into the share and remounts it
read-only, so the guest sees real contents while still being unable to write
to the host copy. `vm unshare <name>` reverses it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
The destructive demo asked the guest to confirm its own destruction, one step
after deleting /usr -- which removes bash, sshd and agetty. Nothing would have
been left to answer, so the check would have FAILED on a step that actually
succeeded. It now proves breakage from the host (the overlay grew, and the
guest can no longer answer at all), which is the same evidence the shutdown
verdict relies on and for the same reason: a corpse cannot report its own
death.
Step 3 also no longer recreates the sandbox. `vm reset` in step 4 is the claim
under test; recreating first would have quietly tested something easier.
README's verdict table now carries the MEASURED qmp events per guest command
rather than the inferred ones, and states the finding that motivates the whole
design: ssh returned an identical "Connection closed by remote host" for the
poweroff, the reboot AND the panic.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
`_run_exec "$name" "$@"` sat on its own line under `set -euo pipefail`, so a
non-zero return aborted the script before the verdict was ever computed. ssh
returns 255 when its connection dies -- which is exactly what a SUCCESSFUL
guest poweroff causes. The single case this tool exists to report on was the
one case that silently printed nothing.
Caught by the acceptance demo, not by the unit tests: verdict_report was
correct in isolation the whole time and passed 8/8 there. Only running the
real CLI against a real poweroff exposed the wiring.
Verified after the fix, end to end:
==> Running in 'fixtest': sudo systemctl poweroff
Connection to 127.0.0.1 closed by remote host.
ok VERDICT: clean poweroff -- guest powered itself off and reached
'reboot: Power down'
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
shutdown_verdict_demo.sh exercises every branch of the verdict table against a
real guest -- clean poweroff, reboot-instead-of-poweroff, kernel panic, and a
command that leaves the machine running. The failure branches matter as much
as the success one: a verifier that has only ever returned "pass" is untested.
destructive_demo.sh is the acceptance test for the tool as a whole: destroy
/etc and /usr in the guest, assert the guest really is broken, reset with one
command, assert it is pristine again, then assert the HOST is untouched
(/etc/hosts attributes unchanged, base image hash unchanged, /etc and /usr
still present).
Measured on the first real run, which confirms the design's central claim:
QMP reported SHUTDOWN {"guest": true, "reason": "guest-shutdown"} and the
serial log reached "reboot: Power down", so the verdict was "clean poweroff"
with exit 0 -- while ssh saw only "Connection closed by remote host", which
would have been indistinguishable from a crash or a dropped link.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Root cause, confirmed causally rather than inferred: masking
systemd-timesyncd (which vmbox does so `-rtc base=` stays pinned, since
timesyncd otherwise resets the guest clock to real time) left
systemd-time-wait-sync waiting for a sync that can never arrive. That unit sat
in "start running" forever and every other job queued behind it as "start
waiting" -- sshd included -- so the guest never reached multi-user.target. It
booted to a login prompt and looked perfectly healthy while being permanently
unreachable. The fix is one line: mask both units together.
Four earlier hypotheses were each disproven by measurement. What finally
worked was asking the guest instead of reading absence-of-log-lines:
`systemctl is-system-running` returned "starting", and `systemctl list-jobs`
put systemd-time-wait-sync at the head of the queue. Stopping it made sshd
active and bound :22 within seconds, and ssh from the host then succeeded.
Rule for next time: a unit that is enabled, has satisfied dependencies, and
neither starts nor fails is QUEUED -- check list-jobs first.
The smoke gate added earlier did its job here: it caught the unreachable image
and refused to seal it, keeping the sandbox so its logs survived.
Also fixes serial_exec markers to be plain alphanumerics -- shell
metacharacters get echoed and line-wrapped by the tty, which made them
unreliable to match and produced spurious timeouts.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Committed by accident: running systemd-ssh-generator --help during
investigation made it treat '--help' as its output path and write generator
units there. Not part of vmbox.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
The Arch cloud image leaves sshd enabled but never started -- no start line,
no failure line, dependencies satisfied. Four hypotheses (NTP, cloud-init
clean, a socket-disable, a blocking start) were each disproven by measurement,
so rather than keep paying a ~10 min rebuild per guess this adds the transport
the tool should arguably have used all along.
For the case vmbox exists for, serial beats ssh on the merits: it keeps
printing straight through the poweroff that kills an ssh connection, and it
works when sshd does not. `vm run` now picks ssh when it answers and the
serial console otherwise, and recovers the guest exit code either way. Guests
get a second serial port (ttyS1) on a unix socket with a root autologin, kept
separate from ttyS0 so driving a shell never interleaves with the append-only
log the verdict is read from.
serial_exec.py waits for an explicit per-command end marker rather than a
fixed sleep. That is not a detail: an earlier probe used a 4s window, the next
command was typed while the previous one still ran, and the tty echoed that
input -- which reads exactly like output in a transcript and produced a
confident but wrong conclusion. It also strips ANSI CSI and systemd's OSC 3008
shell-integration blocks, verified against a real captured boot log.
Also: a vmbox-sshd-kick unit that forces sshd up after multi-user.target as
belt and braces, and the smoke-test sandbox renamed to a name that actually
passes validate_vm_name -- "__smoke" was rejected, so the gate silently never
ran.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
Much of testsAndMisc/linux_configuration cannot be tested on the host: 15
files use `chattr +i` (so a bad test is not simply revertable), the shutdown
installers write units that power the machine off, and boot_recovery installs
pacman hooks. The repo's own CI says as much -- its Ubuntu job runs only the
side-effect-free tests, noting the rest "drive pacman, /etc, systemd or adb".
vmbox runs them in a throwaway Arch guest and verifies the result from the
HOST, which is the part that is not obvious: a successful shutdown kills the
connection carrying the report, and ssh returns 255 for a clean poweroff, a
crash and a dropped link alike. So the verdict comes from two artefacts that
outlive the VM -- a QMP event log written by a recorder attached at launch
(attach later and qemu has already exited), and the serial console log.
Five outcomes are distinguished: clean poweroff (SHUTDOWN guest:true plus a
serial tail reaching "reboot: Power down"), dirty stop, guest-reset (rebooted
instead of powering off), GUEST_PANICKED, and hung. A POWERDOWN event proves
nothing -- it means the request was delivered, not that anything acted on it.
Reset is a qcow2 backing-file trick: the golden image is built once and sealed
(chmod 444 plus a sha256 sidecar verified on every launch, since booting the
base would silently invalidate every overlay), and each sandbox is a thin
overlay, so reset is rm + recreate. No root anywhere -- user-mode networking
with per-VM forwarded ssh ports, and a socket multicast segment for VM-to-VM.
The guest clock is pinned and re-applied on every launch, because the shutdown
installers gate on a 21:00-05:00 window; NTP is masked in the image or it
silently undoes the pin. Repos are cloned from a read-only share scoped to an
explicit directory, never $HOME, and credentials are sentinel values.
`vm build` ends by booting a throwaway sandbox off the sealed image and
refusing to ship if ssh does not answer. Several distinct faults produced an
identical "boots fine, connection reset" symptom; checks inside the build VM
only prove state at seal time, so the gate has to be an end-to-end boot.
Tests: 8 bats cases over the verdict table and name/meta/serial handling;
shellcheck and ruff clean; every file within the 250-line cap.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VdyMBGCrGVGDWtJ9xe48f3
database.rules.json carries YOUR_FIREBASE_UID on purpose -- this repo is
public and the uid is the one value that must not be in it. But nothing
stopped anyone publishing the template as-is, and rules pinned to a uid that
matches no account deny every read and write. The symptom is every device
failing at once with a bare permission error, which reads like an outage
rather than a bad paste.
render_rules.sh substitutes the uid from ~/.config/crdt-sync/firebase.json
and fails closed on anything unsafe: a surviving placeholder, missing
`auth != null`, a missing uid pin (that one matters -- `auth != null` alone
lets any stranger who self-registers through the public Web API key read
everything), or output that is not JSON.
Verified: its output is byte-identical to the deployed
~/.config/crdt-sync/database.rules.json, and the live database still answers
an unauthenticated read with 401. All three failure paths exercised.
Also corrects the README, which described the in-repo copy as "the deployed
rules, with the real uid filled in" -- it never was.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CFSGeWxHPY5rHUGrPqhcmG
lyricanki's Linux desktop cannot use google_sign_in (no Linux
implementation) and its loopback replacement is dark until a Desktop OAuth
client exists. Hiding the button there makes the screen silently offer less
on one platform, which reads as a bug.
googleUnavailableReason renders a *disabled* button plus the reason. That is
not the "control that can never succeed" trap: the trap is a live-looking
button that reports a fake cancellation, and a null onPressed cannot be
tapped at all. Null stays the default, so every existing host is unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CFSGeWxHPY5rHUGrPqhcmG
An optional '' default reads as "this parameter is usually irrelevant", and
it made `avoid_redundant_argument_values` fire in the first app to pass ''
explicitly -- pushing that app toward deleting the argument, which is exactly
the line that must stay for when the Desktop client id arrives.
Requiring it forces each app to state which halves it configured. Passing ''
still means "not configured"; it just has to be said out loud.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CFSGeWxHPY5rHUGrPqhcmG
Five apps had copied google_sign_in_backend.dart; home_inventory's copy was
byte-identical to todo's. lyricanki was about to become the sixth, which is
the mistake this package exists to prevent. So the plugin call moves here.
Desktop is the new half. google_sign_in ships android, ios and web
implementations only -- there is no google_sign_in_linux -- so lyricanki's
real GTK build could never have used the plugin, and "one tap" would have
meant "one tap on the phone, type a password on the PC". The OAuth loopback
flow covers it: bind a socket, open the system browser, catch the redirect,
exchange the code with PKCE. Both halves return a Google ID token of the same
shape, so signInWithGoogle cannot tell them apart.
googleSignInSupported takes the client id rather than reading a const. In
todo the emptiness check was a second conjunct at the call site, and moving
the fetcher here without it would have let a forgetful app ship a button that
always reports "cancelled" -- todo's own file documents that failure. Taking
the value as an argument makes the check impossible to drop.
The decision (which flow, or none) is pure and covered; the two one-line
delegations that reach a platform channel and a real browser are the only
suppressions. 83 tests, 100% coverage.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CFSGeWxHPY5rHUGrPqhcmG
Two different git refs for the same package name cannot coexist in one
resolution, so an app taking both sync_settings_ui and crdt_sync_flutter --
which is the intended pairing, and what lyricanki now does -- failed version
solving while this stayed pinned to v0.10.0.
v0.11.0 only widens what FirebaseAccount.tryParse accepts, so nothing here
changes behaviour: 29 tests green, analyze clean.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CFSGeWxHPY5rHUGrPqhcmG
Deleting prompts/dopamine-ux-01-motion-tokens.md (the completion protocol) left
00-INDEX.md's "Distributed copies" table citing a path that no longer exists, so
verify_prompts went from 3 pre-existing failures to 4. Exactly the dangling
anchor 44bc12b fixed after the doc split.
Asserts the path ABSENT rather than dropping the row, which keeps the check
two-directional: a finished prompt reappearing in a repo is itself a failure,
because the next session would re-run work that is already done. Verified both
ways -- recreating the file fails the check, removing it passes.
This recurs for 02-09, so the comment says to move a row as each completes.
Back to the same 3 failures as before this session, all in ~/screen-locker and
all pre-existing (prompts 04/05).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-22 17:31:03 +02:00
1938 changed files with 159396 additions and 6182 deletions
- deploy: `git -C ~/src/utils pull --ff-only && ~/src/utils/scripts/deploy.sh` (installs the sickdays CLI + launcher from the live checkout, then exits 1 unless every released library's version tag is on origin and matches HEAD and the live `sickdays` runs the text-free checkout; `--verify-only` for the health check alone)
AgentReady defines no catalog schema of its own ("AgentReady does not
introduce its own bespoke manifest"), so the checker uses this one.
- llms.txt: the spec's informative "llms.txt shape" example, and <https://llmstxt.org/>.
## Requirements (quoted from spec.md)
| ID | Level | Requirement | Checker |
|---|---|---|---|
| AR-FIND-01 | MUST | "The site **MUST NOT** block agent and crawler access to public content, in `robots.txt` or at the network layer." | robots.txt is 200 `text/plain` and not HTML. It has a `User-agent: *` group, and no group has a blanket `Disallow: /`. In url mode the homepage returns 200 with no challenge page for ClaudeBot, GPTBot, Claude-SearchBot, OAI-SearchBot, Claude-User and ChatGPT-User. |
| AR-FIND-02 | SHOULD | "Answers agents are expected to cite **SHOULD** be published on one canonical, current, first-party docs page." | SKIP (manual) |
| AR-FIND-03 | SHOULD | "The site **SHOULD** list its pages in a `sitemap.xml` referenced from `robots.txt`." | robots.txt has a `Sitemap:` line. The sitemap is XML (by content type in url mode), and every `<loc>` is on the site's own host. |
| AR-FIND-04 | MAY | "...publish an AI Catalog at `/.well-known/ai-catalog.json` listing every agent-facing resource it ships ... linked from the homepage head with `<link rel="ai-catalog">` and an HTTP `Link` header." | The catalog is valid (see the schema below) and lists `/llms.txt`. The homepage has `<link rel="ai-catalog">`. In url mode the homepage also sends a `Link: <...>; rel="ai-catalog"` header. |
| AR-READ-01 | MUST | "The answer **MUST** be present in the initial HTML — server-rendered or prerendered, not injected by a single-page app after load." | The homepage and each sitemap page return 200 `text/html` with at least `--min-text` characters of visible body text. A page that is byte-identical to the homepage fails. An empty sitemap is a FAIL. |
| AR-READ-02 | MUST | "...accurate HTTP status codes: `404` (or a valid redirect) for missing paths ... never a `200` for a missing page" | url mode: two random `/agentready-probe-<hex>` paths must end, after redirects, on a 404 or 410. dist mode: a `404.html` with text must exist. |
| AR-READ-03 | SHOULD | "The homepage **SHOULD** be served as raw HTML with real links to docs, product, and pricing." | The homepage HTML has at least one `<a href>` to another page on the site. |
| AR-READ-04 | SHOULD | "Docs pages **SHOULD** answer the tasks users bring to agents." | SKIP (manual) |
| AR-READ-05 | SHOULD | "Code snippets **SHOULD** be fenced and language-tagged." | Every `<pre><code>` carries `language-*`, `lang-*` or `data-language`. Every fence in a markdown mirror has an info string. |
| AR-READ-06 | SHOULD | "Discovery files (`llms.txt`, `.well-known/*`, `openapi.json`, `AGENTS.md`) **SHOULD** be linked from content agents already visit" | `/llms.txt` and the catalog (plus `/openapi.json` with `--act`) exist and are linked from the homepage, by `<link>` or `<a>`. |
| AR-READ-07 | SHOULD | "The site **SHOULD** ship an `llms.txt`: a lean, described index of its key pages." | 200 `text/plain` or `text/markdown`, not HTML. It needs an H1 first, a `>` summary, and an `## H2` with a `- [name](url)` list; other prose is allowed. Every on-site link resolves. |
| AR-READ-08 | SHOULD | "Key pages **SHOULD** carry JSON-LD — the homepage above all" | The homepage has a parseable `application/ld+json` block with `@context`. |
| AR-READ-09 | MAY | "Pages **MAY** advertise a markdown mirror via `<link rel="alternate" type="text/markdown">` or an equivalent HTTP `Link:` header." | Every checked page advertises a mirror. The mirror is 200 `text/markdown` or `text/plain`, and is not HTML. |
| AR-READ-10 | n/a | "`AR-READ-10` is retired." | not emitted |
| AR-READ-11 | SHOULD | "Pricing **SHOULD** be published on a first-party page" | SKIP (n/a: no paid plans) |
| AR-ACT-01 | SHOULD | auth completable by an agent; "OAuth 2.0 with discoverable metadata under `/.well-known`" | `--act`: either `oauth-protected-resource` has `resource`, or `oauth-authorization-server` has `issuer`. |
| AR-ACT-02 | SHOULD | "A public API **SHOULD** be described by a machine-readable OpenAPI contract." | `--act`: `/openapi.json` has `openapi` and `paths`, and every operation has an `operationId`. |
| AR-ACT-03 | SHOULD | "...expose an MCP server over Streamable HTTP at `/mcp`." | `--act`, url mode: a POST `initialize` gets a JSON-RPC result, or a 401 with `WWW-Authenticate`. |
| AR-ACT-04 | MAY | "An MCP server **MAY** ship a server card" | `--act`: the catalog has an `application/mcp-server-card+json` entry. |
| AR-ACT-07 | SHOULD | "...register WebMCP tools on `document.modelContext`." | `--act`: `modelContext` appears in the homepage HTML or in its same-site scripts. |
| AR-ACT-10 | MUST | "...**MUST** expose at least one interface an agent can act through: an MCP server (`AR-ACT-03`) or a CLI (`AR-ACT-06`)." | `--act`: same verdict as the `/mcp` probe. |
## kuhy's choices that are stricter than the spec
- Every in-scope item is required, whatever its strength. That includes the
MAY items (FIND-04, READ-09).
- The spec says a site "**MAY** block a training crawler". kuhy chose to allow
training crawlers too, so a blanket `Disallow` for GPTBot, ClaudeBot or any
other agent is a FAIL.
- The act stage is off by default. Turn it on with `--act` only for sites with
user actions. The spec allows this: "a purely informational site conforms on
`AR-FIND-*` and `AR-READ-*` alone".
## AI Catalog schema (ai-catalog.md, specVersion 1.0)
- The media type is `application/ai-catalog+json`. When fetched by link
discovery, plain `application/json` is also acceptable.
- At the top level, the object MUST have:
- `specVersion`: a string in "Major.Minor" form, for example `"1.0"`.
- `entries`: an array, which may be empty.
- The top level MAY have `host`, `extensions` and `signature`.
- If `host` is present, it MUST have `displayName`. It MAY have
`identifier`, `documentationUrl` and `logoUrl`.
- Each entry MUST have:
- `identifier`: a URI or URN. The recommended form is
`urn:air:{publisher}:{namespace}:{name}`.
- `type`: a media type, for example `application/mcp-server-card+json`.
- **exactly one** of `url` or `data`.
- `displayName` is optional on entries (ai-catalog ADR-0016).
- Link discovery uses `Link: <...>; rel="ai-catalog"` or
"description":"What an app writes for the errors loop (DOCS-errors.md): `~/.local/state/app-errors/<app>.json` is a `report`; kartka's own state.json is mapped to the same `record` by an adapter. One record per identical error (signature).",
"$defs":{
"record":{
"type":"object",
"description":"One identical error or slow wait. The loop makes one ticket per `app:signature`.",
"message":{"type":"string","description":"Masked message (no secrets, no user data)."},
"top_frame":{"type":["string","null"]},
"count":{"type":"integer","minimum":1},
"first_seen":{"type":"string","minLength":1},
"last_seen":{"type":"string","minLength":1},
"build":{"type":"integer","minimum":0,"description":"Monotonic build of the newest report (Android versionCode); 0 when the source does not know it."},
"description":"One entry of settings.json `error_sources`. `resolve` and `build` are argv lists; `{signature}`, `{repo_dir}` and `{commit}` are replaced.",
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.