The chat picker's unread count is headed NEW and its last-message age
WHEN; the channel list's unread count is NEW; the live feed's timestamp
lane is WHEN. Short words over a badge and an age, which UNREAD outgrew
by a cell and had to borrow the gap for. The command line's columns keep
their names: they are a documented contract of their own.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
On the PicoCalc the main menu's middle chip, Quit?, opens the quit
confirm, and the confirm's own lane was empty, so leaving took a move
to the Quit button and Enter. The confirm now carries Quit! on that same
F3, sending the same quit, which with the question up is the leaving, as
a second ^Q is. A button dialog can take a lane for that one case; every
other prompt keeps its empty one.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A ── Label ── heading drawn in the border's own colour read as part of
the frame it sits in. Grouped lists' headings and the chat's day
dividers now wear their own theme style, heading: a bold dark grey on
the desktop, slot 8 on the PicoCalc. The frame keeps the accent.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Beside its square the shorter word is unmistakable, and it brings the
whole key to 52 cells, so the legend under a path graph or the contacts
table is one line on the PicoCalc and the Cardputer instead of two. The
type's name stays room server everywhere else, the CLI's answers
included.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
meshterm.net/map and www.example.org are links now, as surely as their
https forms: a host starting www., or ending in a top-level domain IANA
has delegated, read from the registry itself, kept verbatim in
assets/tlds.txt. So file.txt stays text, txt being no TLD, and so do a
decimal, an abbreviation, an address, and a path. A scheme-less link
opens, and its QR code encodes, as https://. Link finding moves out of
the chat screen into core/links.py.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A URL too long for a message's hanging lane steps out to the timestamp's
column on a line of its own, whole, since a terminal opens a link, and a
reader copies one, only while it sits on one line. Only a URL wider than
the screen still folds. ^U on a picked message opens its links on the
share screen, one code at a time, and with several, the arrows step
through them, the URL line flanked by a lit or dim arrow each way. On the
PicoCalc the lane carries it as QR, behind Day down.
CLAUDE.md also states the shortcut rule: a bare letter only on a screen
that takes no typing, a Ctrl chord everywhere letters are text.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The bordered frame pads its body one column in from each side border,
which is air for a row of text and two columns of drawing given up on
the map. A screen can now be flush (Screen.flush), and the map is: its
body is cols - 2 wide on the desktop, base_body_size agrees through the
shared panel_inset, and the braille edge scrub drops to the one border
column. The borderless platforms already drew the full width.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Every URL in a channel or direct message, any scheme (a meshcore://
share as surely as a web link), now gets a braille code under the
message, hung at the body's indent. Several sit side by side and wrap
onto a new band when the line is full, so a link read on the handheld
opens on a phone instead of being typed out. Platform.url_codes keeps
it to the PicoCalc: half-block codes would swamp a desktop transcript,
and the Cardputer's 14 rows can't spare 9 under every link.
A picked message now keeps its whole block in view, codes included,
rather than only its first row, so walking the pick down onto a link
no longer leaves its code below the fold.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Braille was a fallback for codes too big for half blocks; on the
PicoCalc it is now the grid every code is drawn on, eight modules to a
cell, share screens and page fences alike. A channel share is 23 cells
by 12 rows there, a contact card 29 by 15, each with its link under it.
The regular platform keeps half blocks, its fonts' braille being dotted.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Now that the PicoCalc's fonts draw braille as solid tiles, a braille
cell is a true 2x4 pixel grid, so it can hold a QR code at half the
module size. The fit ladder tries every half-block fit first (a lighter
error level and a narrower quiet zone cost a camera nothing on a
screen), then the same fits in braille. That puts the contact card on
the 53x26 panel whole: the standard code in 29 cells by 15 rows, with
its link under it, where half blocks had to send the link a page down.
Platform.solid_braille says where braille is solid; desktop fonts draw
it dotted, so the regular platform never offers it. Codes are cached
per (data, error level) since the share screen refits every paint.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
With Esc inert at the main menu, the PicoCalc had nowhere to say how to
leave. Its title bar now ends on ^Q quit there (Esc clear while a filter is
typed), and the menu's F-key lane puts Quit? on F3, which asks first, with
Quit! behind it on F8 to leave without asking. Follows up issue #22.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Esc-mashing back to the menu used to land on the quit confirm, and ^Q, one
key from ^W, quit with no question at all. Now Esc at the main menu only
clears a typed filter, and ^Q floats the same quit confirm the Quit row
opens, from anywhere and without stopping what runs underneath; a second
^Q while it asks leaves at once. The main menu's hint names ^Q quit, the
only place the chord is shown. Option D from issue #22.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Over Bluetooth every command is a write-with-response, and a companion that
restarts on the reboot command can drop the link before acknowledging the
write. The write then hung until the link supervision timeout, holding the
reboot dialog back by a second or more. The reboot now returns once the
write has had a quarter second to fail, and leaves the rest to the teardown
that follows.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The reboot flow used to send the command and then wait for a liveness poll
to see the link drop. A quick reboot behind a port that never goes away
(a USB-UART bridge) was never seen, so the app went on sending commands to
a board mid-restart. Sending the reboot now declares the link down itself:
the session watcher wakes on the announcement, the Device rebooting dialog
takes over, and the first reconnect attempt waits a moment for the board to
actually begin restarting.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Those firmwares guard only the UART write characteristic, so the
refusal arrives on the handshake's first write, which meshcore's send
swallows. The empty handshake read as "not a MeshCore companion", and
the PIN pairing, the stale-bond repair and the macOS Passkey wait never
ran. The write is now wrapped to keep its refusal, and an empty
handshake behind one raises it, so both firmware shapes take the same
pairing path.
bleak no longer gets the PIN on Windows either: its WinRT pair() is
Just Works only, and after a mistyped PIN it left Windows holding an
unauthenticated bond that the next, correct attempt reused and was
refused over.
Verified against a T-Deck on MeshOS 1.3.0: a planted Just Works bond
heals in one run, and an unbonded connect without a PIN now says the
device needs one.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
s (F2 on the PicoCalc) steps all → unscoped → each region heard in the window
A–Z → unknown scope, beside w's window cycle. A narrowed page charts that
scope's floods per day (per hour at 24 h), their time-of-day rhythm, and its
own ledger; nodes per day and arrivals leave, being read off adverts, which
carry no scope. A window's stored floods are read once (Repository.flood_frames)
and resolved against the known region names for the page's life.
The node and own-node pages offer no cycle: a node's page is its adverts and
telemetry, and what we send is traces, which go direct.
The ring's order, words and chip move to ui/scopering.py, shared with the
dashboard.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
An art editor spells white as a reset then a bare bold (ESC[0m ESC[1m): the
console's default grey, brightened. The loader forgot the colour on a reset,
so it dropped the bold with nothing restated and the run fell to the
terminal's own default grey. After a reset the grey is now the colour in
force, and a bold lifts it to bright white like any other.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
s (F3 on the PicoCalc) steps all → unscoped → each region A–Z → unknown
scope → all, and the whole screen follows: the activity chart, the pulse
rate, the traffic tallies and the SNR/RSSI medians narrow to that scope's
floods, while the radio's own numbers stay whole. A narrowed view reads the
screen's 2-hour window, since naming a frame's region is a per-frame check
the database can't run; the headings say so, and a direct frame, which has
no scope, counts only under all. The title names the view, falling back to
"Dashboard · scope …" where a long region name won't fit.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Traffic listed what the device reported (advert, telemetry, message, ack)
beside the frames those reports arrived in, so one advert was counted twice
and its frame row had to be renamed "raw advert" to tell the two apart. It
now counts frames heard on the air only, named by the same function as the
live feed's class lane (payload_marks), and every bar takes the colour the
feed draws a frame's class in.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
"Contacts · 198 known + 123 archived", falling back to "Contacts · 198 + 123"
where the frame's rule can't hold the words. The fallback is a new
Screen.short_title the frame picks through fitted_title, on both the bordered
panel and the PicoCalc title bar.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
records --width now refuses anything but 1, 2, or 4 with a usage error, the
way --category already refused an unknown discipline. A width that doesn't
exist used to come back as exit 5, indistinguishable from an empty database,
and --width 0 was falsy and so listed every width. (#15)
A device with no port is labelled by its name alone. The simulator fell
through to the serial branch and came out as "… (nothing transmits) ()". (#14)
platform and specimen pass their own one-line help, so Typer stops printing
their docstrings — notes for whoever reads cli.py, Sphinx roles and source
paths included — and their summaries fit in meshterm --help. A new test runs
--help on every registered command and refuses those markers anywhere. (#16)
Built from the issues alone; the pull requests filed against them were
closed unreviewed and their code was never read.
Closes#14, closes#15, closes#16.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
MeshCore firmware sets its clock only forward: CMD_SET_DEVICE_TIME with an earlier time is answered ERR_CODE_ILLEGAL_ARG, which read as 'the device rejected the request as malformed'. Since 2026-09-24 the companion's clock has been ahead of this computer's (itself within 0.7 s of NTP), so every connect's sync was refused. set_clock now reads the radio's clock first: within a minute ahead is in sync and nothing is written; further ahead raises ClockAheadError, which says by how much and that a reboot resets it. The device layer also maps the firmware's refusal to that error, and the simulator's clock now only moves forward, like the firmware's.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Measured against a T1000-E from a uConsole (BlueZ 5.66, btmon on the side).
A connect that BlueZ had to retry stranded its link. bleak retries a link
attempt that failed to synchronise (le-connection-abort-by-local, 3-7 per
connect on a Pi radio) inside connect(), and each one fired meshcore's
disconnect handler, which resets BLEConnection.client to None. bleak's
next retry then connected a client nobody held, meshcore gave up as "not
established", and the live link kept the companion from advertising, so
the retry scanned for 30 s and reported it out of range. Three runs out of
three. While connect() runs, a disconnect is now only recorded, and the
teardown keeps the last live client if meshcore let go of it anyway.
A companion given a PIN after it was paired unauthenticated answers the
subscribe with ATT 0x0E, Unlikely Error (bond Authenticated=0). That is now
read as a pairing refusal, so the stale-bond message and the PIN repair
reach it; with the PIN it re-paired and connected.
BlueZ's Pair left its link up (the companion then stopped advertising to
the connect), and failed at once on the same link flake bleak retries. It
now puts the link down after pairing and retries a flake a bounded 4 times.
And BlueZ starts a pairing of its own when the subscribe is refused, asking
the default agent; with none (any headless machine) it waited on a yes/no
until the companion's 30 s SMP timeout. For the length of a connect
MeshTerm is now the default agent for that one device, KeyboardOnly: with a
PIN it answers, without one it refuses at once, and it never approves a
pairing without the passkey, which is how a bond ends up weak.
Unpaired, no PIN: 33 s -> 4 s to "requires a PIN". Unpaired with the PIN:
paired and connected in 11 s. Paired: connects in 7 s, where it failed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The BlueZ fixture patched os.getpid to return a string, and bluez.os is the
global os module, so Python 3.14's event loop (which opens a pidfd for its own
pid at startup) failed before any test ran. The agent path is now derived from
the real pid instead.
The end-to-end chat test piped Esc beside Enter, so a slow runner left the
screen and cancelled the spawned send before the row was written. It now sends
Esc only once the message is recorded.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
On Linux the PIN never reached the radio: bleak's BlueZ pair() calls
Device1.Pair and nothing else, BlueZ asks a pairing agent for the passkey,
and bleak registers none. Without a desktop agent BlueZ falls back to Just
Works, the firmware refuses a bond without MITM, and the picker looped on
"That PIN was rejected" for a correct PIN. MeshTerm now registers its own
Agent1 for the length of one pairing (core/bluez.py) and calls Pair from
the same bus connection, so BlueZ routes the passkey request to it and
never to the desktop's agent. The Windows repair contract carries over:
an existing bond is reused, force removes it and pairs afresh, and the
outcome is recorded so a refusal names it. Unpair & quit and the stale-
bond message work on Linux too, and the remedies there name Linux and
`bluetoothctl remove` instead of Windows' Settings. bleak no longer gets
the PIN on Linux, where it could only start a second pairing.
A serial port that refuses to open was reported as "didn't answer as a
MeshCore device", with nothing logged. Permission denied now names the
dialout group, a held port names the other program (and ModemManager on
Linux, also named when the port opens but never answers), and a vanished
port says so. A failed startup probe is logged at warning.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
One "requires a PIN" sentence covered every GATT refusal, and a link that
never opened read as "didn't answer as a MeshCore device". A Heltec V4 user
whose Windows bond had gone stale was told to pair a device Settings showed
as paired. Each failure point now has its own message and remedy: link never
opened, stale Windows bond, wrong PIN, device refused to pair (with the
WinRT status), paired but still refused, a BLE probe timeout by stage, and
link up but no identity reply. The PIN dialog shows the diagnosis instead of
calling every refusal a rejected PIN, and each case is logged at warning so
a log sent in from another machine names it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Both labels were right-aligned over their numbers, so each ended
mid-value with the unit trailing past it; every other label on the row
starts at its lane's left edge. RSSI also borrowed the SNR's five-cell
field for a reading never wider than four (-120), which set the two
readings three or four cells apart instead of the row's usual two.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
dir messg, chan text, chan data and anon req replace direct message, channel text, channel data and anon request, so the longest class (telemetry, chan text, multipart) is 9 cells and the live feed's class lane gives 5 of its 14 back to the row. The dashboard's traffic tallies and the viewer's headline read the same names from the same table.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The class name was dropped below the width the whole row needed, leaving only its icon. It is now drawn at every width, like the scope; a row wider than the screen is read by scrolling the highlighted one with ←→. The class icon beside the name is a family cue once the name is always there, so it follows Platform.menu_icons: on the PicoCalc, where it could only be a one-glyph stand-in, the lane is gone.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
reset_flood_scope and force_unscoped arrived late in meshcore 2.3.x. The uConsole's 2.3.7 satisfies meshcore>=2.3 without them, so every scoped channel send would have crashed on restoring the scope. set_flood_scope now sends the firmware's command 54 itself ([54][0][key], [54][0], [54][1]), as the default scope's command 63 already was. Every other library call MeshTerm makes exists in 2.3.7.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The radio hands a message over as text, not as the frame it rode in, so a 💬 row opened from the live feed had no scope to show. Its card now finds the message's flooded copies in the packet log, the way the message paths dialog does, and reads the scope off them into a scope row. The copies found are kept per row; the region is resolved on every paint, so a name learned while the card is open names it. Direct messages get no row: their copies are encrypted end to end, and a routed one has no scope.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The firmware parks its home mark on the wildcard when no region is home, so *^ means no home region set. The wildcard row's note now reads 'unscoped · no home' then, and 'unscoped floods' once a region is home.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
With a message picked, ^R (the Resend chip on the PicoCalc) offers that one: a scoped message of ours is resent unscoped, and a pick that isn't one offers nothing rather than reaching past it to another. With nothing picked it is still the newest. The pick clears once the resend lands at the tail, and the picked-row hint names ^R when it would act.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
It came and went with the terminal's width: hidden between 70 and 83 columns, where the class label won, and always on the PicoCalc. It is now drawn at every width, and the class label is what gives way on a narrow screen, the icon still saying what the packet is. Beside the subject, the lane is on screen even at 53 columns; there it is the readings after it that scroll off.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The region rode the route row as a trailing atom (flood · scope harbour), easy to miss on a card opened from the feed's SCOPE lane to see it. Route now says how the packet was routed (flood, direct) and a labelled scope row underneath says where: the region, 'unknown region · code 3fa1', or 'unscoped'. A direct packet has no scope row.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The chat title reads "#devtest · onqc" rather than "#devtest · scope onqc". `channels scope N` on a channel with no scope prints - (null in --json) and exits 0: sending under the device default is an answer. Only an empty slot still exits 5.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
SCOPE moves ahead of the SNR and RSSI readings, since where a flood was allowed to go is a fact about the packet like its subject; the readings are about how it reached us. A plain flood draws a muted dash instead of the word, so the lane is quiet until a row is scoped, and the lane narrows to 10 cells. The Channels list's column test rides along.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The meshcore library does not serialize commands, and every command
waiting for "OK or ERROR" takes the first ERROR to arrive. At connect the
channel prewarm runs beside the clock sync, which this companion refuses
as malformed; a slot read handed that refusal took it for "no such slot",
ended the probe there, and cached a list missing every channel after it
for the rest of the session.
Only the firmware's NOT_FOUND (or a code-less refusal from old firmware)
now ends the list. Any other refusal is asked again once, and a second
one, or no answer at all, is a failed read the probe reports as
unfinished and does not cache.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The region each channel's messages are flooded into, blank for one that
sends under the device default. The lane is only as wide as the longest
scope, and shares a budget with the name lane so that at 72 columns it
never cuts the activity sparkline: with a name at its ceiling it gets
eight cells, what the row had spare. On the PicoCalc the row was already
cut inside the activity lane, so there it costs the LAST ages.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
On Windows, Click expands any argument holding `*` into the files it
matches, quoted or not, so `chat send --scope '*'` arrived as the working
directory's file names. The entry point turns the expansion off; nothing
MeshTerm takes is a glob.
`channels scope N REGION` on an empty slot did nothing and exited 5, which
tells a script the scope was set and read back empty. A write there is
now a usage error; a read still exits 5, as `share` does.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
"scoped · 3fa1" put a `·` inside the reading, so in the message paths
title it chained as two status atoms. It is now "unknown scope 3fa1",
and "? 3fa1" in the live feed's SCOPE lane, whose heading already says
what the column holds.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The node page of a repeater shows the regions it was last heard to relay
floods for, whether it relays unscoped floods too, and when it said so, and
asks it once on request. The region store keeps that unscoped flag and the
answer's time beside the regions.
Repeater admin gains a Regions page: the region tree read over the remote
CLI (core/region_admin.py parses the dump, drops a line the 160-byte reply
cut in half, and recovers the rest from the flat lists), flood allowed or
denied per region and for the wildcard, home, default scope, add, remove
and save. Each edit is one command applied at once; the title counts what
a reboot would undo. meshterm regions NODE asks from the command line.
The simulator transcribes the firmware's region CLI, reply cap included.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The firmware has no per-channel scope, so a channel send is now a window
on the companion's session scope: set it, send, restore it, under a
transmit lock every flood-capable send also takes (DMs, adverts, admin
commands, logins, regions/neighbour requests), so nothing interleaves.
Ack waits stay outside the lock. A scope the radio can't take stops the
message (FloodScopeError naming the firmware it needs) rather than sending
it unscoped or under the default; * checks the firmware version, since
before 1.16 the override would quietly fall back to the default scope.
A failed restore is repaired by the next transmission.
Channel scopes are set from the channel page's Send scope picker (known
regions with their carriers, or a typed name) and from the CLI (channels
scope, chat send --scope). The chat's title carries the scope as a status
atom, ^R/Resend sends the newest scoped message again unscoped after an
amber confirm, and the scope each sent message went under is recorded
(messages.scope, schema v17) so the paths dialog can say what it went
under and when no known repeater carries that region. Device config now
teaches the region store the default scope it reads or sets.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The packet viewer's route row reads "flood · scope harbour" (or
"scoped · 3fa1" for a region nobody here has named, or "unscoped")
instead of the library's bare "tc flood"; a direct frame keeps its
route word and states no scope. The live feed gains a SCOPE lane where
the width holds it, 72 columns included (the subject lends it one
cell); the PicoCalc's row already scrolls at 53 and leaves the scope to
the viewer. The message paths dialog states a message's scope once, as
a title atom, read off its arrivals' frames. monitor streams a SCOPE
lane before NAME, and the scope object (or null) under --json.
The viewer and the feed take the region store's scope_of by injection,
falling back to reading the frame alone, and the viewer's card cache
keys on the scope so a region learned while it is open renames it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The groundwork for region scopes. core/regions.py holds the firmware's
maths (key = SHA-256("#"+name)[:16] over UTF-8, code = HMAC over type byte
+ payload with the reserved values stepped off) and the Scope reading of a
received frame. core/region_store.py keeps the regions known by name,
which repeaters carry them, and each channel's send scope, and memoizes
resolution. Observations gain transport_code and scope_body (schema v16)
so a scoped frame resolves against a name learned after it was heard.
Device gains set_flood_scope (the session override, * forcing unscoped)
and request_regions (the anonymous REGIONS request). The default scope
is now framed by MeshTerm: the library stored it with a # the firmware
and apps leave off, and padded by characters rather than bytes.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
--spi used to take whichever SPI profile config.toml listed first, a
coin toss that transmits once two radios are configured. It now counts
the radios actually attached: exactly one is used whether or not a
profile names it, which keeps the common single-radio case to a bare
--spi. Two attached, or two SPI profiles with nothing attached to count,
is a selection error (exit 3) that lists them and asks for -p, the same
way several serial devices with no default already are.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
