Integration review findings: the scripts they run ssh'd to whatever 'frame' means
in ~/.ssh/config. They now take FRAME_ALIAS and FRAME_SSH_OPTS from the server's route.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The Mac view's tunnel is its own ssh, so it now takes the headset's route (and its
pinned identity, which also checks the USB-C address), and closes when the app
switches headset. The MCP adapter's private server (FRAME_PRIVATE_SSH=1) skips the
one-server lock and can't add, remove or switch headsets.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Stop always calls systemctl and treats exit 5 (unit already collected)
as done, so there is no is-active/stop race. Cleanup never masks the
copy error, the play ssh timeout covers the remote worst case, and
tests cover stop exit codes and systemd-run stderr reporting.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Stop is a no-op when the collected player unit is already gone
(raw systemctl stop exits 5 on the Frame; verified 2026-09-29).
- Surface systemd-run stderr when the player can't start.
- Keep the copy error if the cleanup ssh also fails; reject upload
names that the play path can never accept.
- Allow 60 s for play (ffprobe 30 s + systemd-run 15 s remote).
- Docs: four-hour cap is unconditional; no delete action yet; fix a
garbled timing sentence.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
HTTPServer.server_bind calls socket.getfqdn, which stalled past the MCP
backend's 10-second startup window on GitHub's macOS runners.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Two servers each connected, reconnected and edited the headsets on their own,
and several review findings were ways one could move the other's install to a
different headset. A lock file in the data folder now refuses a second server
with a plain message; FRAME_CONTROL_DATA_DIR still gives a separate one.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
ssh's %C hashes only address, user and port, so two headsets reached at one
address shared a ControlMaster and one's commands could run on the other: the
ControlPath now names the headset. Another Frame Control server choosing a
different headset no longer moves this one's commands mid-install.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Every change now takes a lock file shared across processes and starts from
what's on disk; reads pick up a newer file.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- While the connector is taking a queued reconnect off its list, the old
connection no longer counts as live, so no install starts on it.
- Importing a block without a Port takes the port ssh would really use
(ssh -F <config> -G), e.g. one a later Host * sets.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- An upload's answer arriving after a switch opens nothing; the APK
alternatives dialog installs on the headset the APK was checked for.
- A handshake that goes silent (e.g. a jump host's forward hanging) moves on
to the next address.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Behind a jump host, a forward it couldn't open moves on to the next address;
only a refused key stops (judged by ssh's words, not the step).
- Add a headset suggests an alias no Host in ~/.ssh/config already uses.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A jump host's own "Authenticated to" line no longer counts as the headset's,
and a master that logs in but doesn't start lets the next address be tried.
- Devices tab changes refresh through the ordered list load, so a late answer
can't undo a newer selection.
- A probe's time out starts after the name lookup: macOS can take 5 s to look
up a .local name (found on the real Frame once its USB link went away).
- An attempt's ending is published from a method, not a return in finally.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A switch the server refuses no longer drops answers the page is waiting for
(an install's job id): only an actual change of headset does.
- Test now goes through a jump host when the alias uses one.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A set-up headset whose alias goes through a jump host (ProxyJump or
ProxyCommand in ~/.ssh/config) is reached through it, address by address,
still pinned per headset.
- A refused switch puts the header's switcher back on the headset in use.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A command that fails after a switch doesn't make the connector drop the new
headset's connection.
- On first import, the app keeps using the `frame` headset even when Set Up
Connection put another block above it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Terminals, power and a reconnect's probes use the route commands have now
(a pinned bare-alias destination, a login change still deferred).
- Saving port 22 keeps an explicit Port line where there was one.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A bare alias's route is pinned to where ~/.ssh/config sent it when it was
routed (HostName, Port, User), so editing that file can't move an install.
- Renaming during an install is allowed: only a real user or port change waits.
- The Devices tab follows a network change even while the headset is offline.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Retry now (while connected) and Forget identity wait for running installs.
- Terminal windows get the headset's address by name, so a link-local IPv6
zone never has to pass through Windows' console.
- Renaming the headset in use shows at once in the header.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Every command to a set-up headset checks its pinned key
(StrictHostKeyChecking=yes, whatever ~/.ssh/config says); only the
connector's first handshake may save one.
- A reconnect during an install keeps the whole route it started with, also
when a bare alias is set up meanwhile.
- Removing the headset FRAME_ALIAS named doesn't bring it back as a bare alias.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A reconnect while an install runs keeps the login it started with; a new
one from ~/.ssh/config applies after.
- Frame > Open SSH goes through the server, so it uses the same headset and
address as the app and refuses when there's none.
- A bare frame alias in use when a headset is set up stays selectable.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A headset set up while a bare alias is in use doesn't take over by itself;
a login change from ~/.ssh/config waits for running installs.
- Saving a headset writes only the login fields that changed, and only if the
block still holds the old ones.
- SSH, SFTP, power and remote desktop open with the same headset and address
as every other command, and refuse when there's no address.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Removing or moving the active headset's address waits for running installs,
like switching.
- A volume change still waiting to be sent goes to the headset whose slider
it was, and a switch cancels it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A title waiting its turn to be read stays with the headset it was dropped
on, and is dropped if the app switches meanwhile.
- Probes still finishing from an earlier attempt can't overwrite the rows of
a newer one.
- The FRAME_ALIAS the server started with stays on the list after switching
away, so it can be picked again.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A batch of dropped files stays with the headset it was dropped on, and
stops if the app switches.
- Removing or moving the address in use reroutes at once; a headset with no
addresses reaches nothing rather than whatever ~/.ssh/config says.
- A late answer to an older device-list request is ignored.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Removing every headset leaves none in use (commands fail at once) instead of
falling back to the `frame` alias.
- ssh goes to the IP that answered for IPv6 too, with a link-local address's
interface (verified: frame.local over fe80::…%en9 on the real Frame).
- Find results only show in the panel of the headset they were for.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- The headset a change is meant for is checked and the work counted in one
step, so a switch can't slip in between (uploads too).
- A sideloaded title read on one headset can't be installed on another; open
confirmations close on a switch.
- The only headset can't be removed while its ssh alias stays behind.
- Answers about the previous headset are dropped without touching panels; the
catalogue's Installed tags are rebuilt for the new headset.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A switch publishes the new headset at once, so the page clears the old one's
panels and the lists behind them (games, store, Android apps, screenshots).
- The page names the headset its changes are for (X-Frame-Device); the server
refuses one meant for a headset it has switched away from (409).
- A rejected address edit changes nothing.
- Test now goes to the IPv4 address that answered, like the connection.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Any unexpected error while launching clears the launch and reports it, so
the pad can always be started again; the temporary stderr file is made
inside the handled path and a failure reading it is tolerated.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A switch clears every headset-specific list and its buttons at once.
- SSH goes to the IPv4 address that answered the probe, not the name again.
- A rejected headset edit changes nothing.
- The SteamOS/Lepton builds recorded in reports are read again per headset.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
discard() swallows OSError as well as Failure, so a launch that fails and
can't remove its copy still reports the error and can be started again.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The agent holds its copy from its first status line on and tidies it up
however it ends. Before that (a launch error, or stopped before the agent
ran) the server removes the copy itself. discard() only ever removes
incoming copies, never the iPhone bundle's own. The retry race test waits
for both contenders' decisions instead of sleeping.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Attempts carry a generation: one overtaken by a switch, a removal or a login
change routes nothing back to the old headset and can't report connected.
- Removing the active headset or changing its user/port reroutes at once,
before anything that can fail.
- One known_hosts file per headset (~/.ssh/frame-control-hosts/<id>):
forgetting one headset's key can't drop another's, whoever else writes.
- learn() checks, under the config lock, that the block is still what the
attempt started from before writing to it.
- A switch stops live video and drops captures from the previous headset.
- A probe shares its time between the addresses a name resolves to.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- A start turned off while copying removes the copy instead of starting an
agent that would be killed before it could tidy up.
- A local read error while copying removes the partial copy too.
- The retry race test holds the new start until the retry has decided, so
it fails every time without the fix (checked 3/3), not by luck.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Review round 12: a failed USB-C tunnel now retries the normal path; an
existing HostKeyAlias wins; --host with --usb is rejected; the Steam
desktop-streaming claim is now 'untested' (Valve documents the desktop
showing when a game loses focus); the Show/cleanup overlap test blocks
for real (it fails without the lock). Verified live: with the cable out,
the route is the normal path.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Switching headsets is serialized with the start of any install; background
work counts as running from before its thread starts. use() reroutes every
command at once and makes ensure() wait for the new headset.
- A new user or port reroutes commands even if the attempt then fails.
- ~/.ssh/config edits take a lock file shared with Set Up Connection
(frame_connect.py and connect.sh, which now also writes atomically).
- A finished attempt no longer writes its older settings over a change Set
Up Connection made meanwhile.
- Pin edits are locked and swapped atomically.
- A bare alias behind ProxyJump/ProxyCommand is left to ssh to reach.
- The page drops answers about the previous headset after a switch; the
header switcher takes clicks in the macOS title bar.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Each start keeps its copy (holding a lock on it) until it ends, however
it ends, then removes it; a restart can still unpack it.
- The server removes a partial copy when copying fails.
- Other copies are removed only when unlocked and over an hour old.
- Tests: a start racing the need-packages retry (one agent, not two), a
restart that must unpack its copy, a held copy surviving the sweep.
Both regression tests fail without their fix.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Plugged into the Mac, the Frame is a USB network device ("Steam Frame",
usb0 at ~0.9 ms). The tunnel uses it when the Frame's usb0 answers,
with the usual host key; otherwise the normal path. Interleaved runs:
content p50 7 vs 10 ms, click to drawn 17 vs 27 ms, scroll p95 23 vs
31-37 ms. FRAME_MACVIEW_USB=0 turns it off; the card says "over USB-C".
- Bench: --usb, and the route is recorded per run.
- Docs: Steam's own streaming (no SteamVR host on macOS; Remote Play pairs
but streams games, not windows; test blocked); the Frame's USB network;
the 2026-09-28 health-check boot-loop recurrence.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- The need-packages retry only runs if no start() has taken over, so two
agents can't end up running for one session.
- Each copy goes to its own incoming folder; the agent removes its own once
connected (and any a cancelled start left over an hour ago), so a stopped
start can't delete files another is copying or still needs.
- A missing package folder means need-packages, not an error.
- Tests keep fake agents running, so they check ready and which agent owns
the session, plus a bounded retry and the tidy rule.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- No switching headsets (or changing the active one's user/port, or removing
it) while installs run: they read the ssh settings step by step.
- Switching reroutes every command to the new headset at once, even if it
never answers.
- ~/.ssh/config edits are serialized, use unique temp files, and back off if
another program wrote the file meanwhile.
- stop() ends a handshake in progress and joins the connector.
- Pinned keys are written unhashed (HashKnownHosts=no); hashed ones are still
found and forgotten via ssh-keygen.
- Set Up Connection changing a headset's user or port updates the registry.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- iPhone build fails if the bundle can't be made, instead of shipping a
stale archive with an empty version.
- A different build that another device is using right now is left running
and replaced once nobody is, rather than stopped under them.
- If what's installed changed after the server looked, the agent asks for
the packages (need-packages) and the server copies them and starts again.
- The installed-build check sends bytes, so Windows' CRLF can't break it.
- Starting again while a stopped start is still copying launches anew;
concurrent copies use their own temporary names.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Review round 11: the cleanup's check and pkill now hold a lock that Show
takes to count itself in, so a new viewer can't start between them.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Frame Control can now manage more than one Steam Frame, and reach each at any
of several addresses (LAN IPs per network, its .local name, Tailscale). A
connector in the server tries them all at once, picks the best one that
answers, follows ssh -v through each stage (network, finding, SSH, identity,
login) and streams that to the page. The header shows it live; a new Devices
tab (key 5) manages headsets, addresses and network names.
- ui/frame_devices.py: registry in devices.json, imported from the managed
~/.ssh/config blocks; per-headset host key pinning; config block updates.
- ui/frame_network.py: gateway IP+MAC fingerprint, Wi-Fi name, Tailscale.
- ui/frame_link.py: the connector, Test now, Tailscale/mDNS discovery, API.
- server.py: ensure_master delegates to the connector; /api/connection,
/api/connection/events (SSE), /api/devices.
- Electron: headset switcher and Devices item in the Frame menu.
- frame_connect.py --alias; FRAME_CONTROL_DATA_DIR / FRAME_CONTROL_SSH_DIR
keep tests off real data.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Review round 10: a Show replacing its own stream could have its new viewer
ended by the cleanup; a viewer reset left the delayed relay pending.
Verified: the relay delivers what's queued, then closes the agent side.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Chromium on the Frame outlived its last viewer window (verified: 11
processes left after a run). Once nothing is shown, Stop ends it, unless
Show was pressed again meanwhile; its profile is Frame Control's own.
- Relay --delay: if the agent side fails, close the viewer side too
(review round 9).
- Docs: the final scroll run captured 57 fps; don't blame ScreenCaptureKit.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The keyboard and trackpad no longer fetch KDE Connect from Valve's package
repository on the Frame. The desktop apps and the iPhone app's Frame bundle
carry Valve's arm64 build of kdeconnect 24.02.2-1 and the five libraries it
links (kcontacts, kpeople, modemmanager-qt, pulseaudio-qt, libfakekey),
pinned by SHA-256 in frame/kdeconnect/packages.json and downloaded at build
time from the kdeconnect-frame-24.02.2-1 release, which also holds Valve's
complete source package for each.
On first use the computer copies them over its SSH connection (the iPhone
bundle already has them on the Frame); the agent checks each SHA-256,
unpacks them and stamps which build it is, so later starts copy nothing.
No internet on the Frame, 3.6 MB instead of 8 MB, 18 MB unpacked instead of
82 MB (ModemManager and friends were packaging-only dependencies).
GPL/LGPL compliance: frame/kdeconnect/NOTICE.md names each exact version,
licence and source; per-project licence texts in frame/kdeconnect/LICENSES;
THIRD_PARTY_NOTICES.md; an About and licences dialog in the app.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Third review follow-up (PackageParser.buildClassName). Confirmed the
targetActivity resource id 0x01010202 in Open Saber Plus's manifest.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Brings in #4, #5, #8 (fbl100's verified Remmina/VNC mirror), APK
alternatives and the website. docs/streaming.md: Mac in the headset stays
the recommendation (now verified on the Frame); Remmina keeps #8's verified
evidence as the whole-screen fallback. docs/mac-in-headset.md cites #8's
lag finding.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Review round 8 (GPT-6 Astra xhigh): --delay paused upstream reads, so busy
streams saw 30-60 ms instead of 30. Verified locally: 60-62 ms round trip
with 30 ms each way under load. No saved result used --delay.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Per-frame timing on the Mac's clock (capture, encode, network, decode,
draw), viewer clock sync and reports, input echo, /stats and a HUD.
- scripts/macview-bench.py: repeatable runs on the real Frame, a shaping
relay (no sudo), interleaved A/B between agent settings; results in
bench/results/.
- Adaptive controller: ack-based send gate with jitter-aware slack, AIMD
bitrate that knows when a stream is app-limited, fps then size tiers.
On a 50->3->50 Mbit/s step, scroll p95 went from 4.7 s to 72 ms; no cost
on a clean link.
- Separate mode: real AppKit event loop (HiDPI and NSScreen now work),
cropped capture for fixed-size windows, windows kept on their display,
graceful quit restores windows; stop/start races fixed.
- Encoder timeline clamp (no oversized frame after a pause).
- Frame Control shows each live stream's fps, delay, bitrate and tier.
Reviewed by GPT-6 Astra xhigh (read-only), 7 rounds; findings fixed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Second review follow-up: with several activities (e.g. a splash activity ahead
of the game), the alias fallback now prefers the real activity named by the
alias's android:targetActivity. Reads targetActivity by resource id, updates
the error text and docs/vr-apks.md.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Review follow-up. inspect() now returns 'repairable' and the filters it can
patch: the real activity's VR MAIN filter, or, when LAUNCHER/VR sits only on an
<activity-alias>, any real MAIN filter with a category to copy. install() and
patch() repair on 'repairable' instead of 'vr_activity', so a flat Godot 4
export is fixed and a VR category only on the alias no longer aborts install.
Tests cover both shapes, an already-launchable activity with an alias, and an
end-to-end patch.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
On a maintainer's Mac the compatibility-database key is in the Keychain, so a
test that reached install reporting published fake reports. Every test module
now imports tests/sandbox.py first, which points app data at a throwaway
directory (new FRAME_CONTROL_DATA_DIR), turns telemetry off and sends the
database nowhere.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Lepton's apk-info-extractor ignores <activity-alias>, and Godot 4 exports put
LAUNCHER only on an alias (com.godot.game.GodotAppLauncher). Open Saber Plus
0.7.67 installed but Lepton exited with 'APP_ACTIVITY is empty'. Count only
real activities as launchable and patch the activity's VR intent filter.
Verified on the Frame: Open Saber Plus launches and renders.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- Analytics go to the maintainer's PostHog US project 343535, tagged
$lib = frame-control. Every event carries $ip 0.0.0.0, since PostHog
stores the sender's address otherwise (checked live), including events
queued by earlier versions.
- Report a problem sends a private problem_report event to PostHog instead
of a public GitHub issue, with its own random id so a contact address
can't be linked to analytics. The dialog asks how to reach the person and
shows a reference. Maintainers read reports on the PostHog dashboard or
with `python3 ui/frame_report.py inbox`.
- Community sync pages by timestamp in UTC: PostHog refuses OFFSET for
personal API keys.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
patch() turns corrupt-manifest struct/index errors into FrameError; a missing
layer build says so; the signing key falls back to a rename where hard links
aren't supported and explains how to recover from a bad cached key; the layer
nulls an instance it can't destroy and logs xrLocateSpaces once.
Not changed: the 1.1 Meta profile names match xr.xml's promoted names
(meta/touch_pro_controller, meta/touch_plus_controller), and grip_surface is
palm_ext renamed, so the rewrite stays (now commented).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
