A WizardConnect pairing could land on an address the user had never seen.
Pairing took the selected wallet when it qualified and otherwise the first
pairable one in list order, so with the selected wallet ineligible (a WIF
import has no xpub) it silently fell through to whichever wallet happened to
be first. The approval named that wallet by label only, which does not help
when the label is one Aegis generated.
Three changes, one idea: the wallet a purpose uses should be something you
said, not something that fell out of list order.
- Roles. A wallet can be nominated for payments and/or for WizardConnect,
set from its manage modal and badged on its row. A role that points at a
removed wallet reads back as null instead of being trusted, so a stale
pointer can never quietly redirect a payment.
- The pairing approval asks. With more than one candidate it offers a
dropdown of them, each labelled with its own short address; with only one
it shows that wallet's full address. The rows are static, so naming a
wallet above a select the user can change would contradict itself — hence
one or the other, never both. The panel's own Connect pane now follows the
same precedence, because two rules for "which wallet" is how a pairing
surprises someone.
- Send gets a From row listing the wallets on this coin and network, with
balances. It switches the panel selection rather than carrying a separate
source: planSend and send resolve the wallet host-side from that, and a
second notion of "current" would let the form and the approval disagree.
Payments also becomes the opening selection when nothing has been picked yet,
which is what nominating it is for.
No Theseus release needed — approvalModal has supported a select row all
along, and the pick comes back as "allow+wallet=<id>", validated against the
options offered.
Ariadne 0.1.13 exposed /api/status and per-source enable flags in
policy.json. Theseus's Plug-ins > Ariadne's Thread sub-page now wires those
into a full UI, no daemon restart, no UAC.
Added to the plugins-ariadne sub-page (after Status, before Remove):
Collision policy -- radio group (BCNR-first / ICANN-first) writes
C:\ProgramData\Ariadne\policy.json.policy; hot-reloaded
by the daemon within 5 s.
Sources -- 3-column grid, one row per source (snapshotHttps,
electrumWss, perQueryLookup, diskCache, localApi):
enable checkbox + last-state summary
(last success / last error / hit-miss counters /
disk-cache size+mtime). Toggle writes
policy.json.sources.<name>.enabled and re-polls after
the 5-s hot-reload tick so the state text catches up.
Status report -- <pre> JSON dump of GET http://127.0.0.1/api/status
with Copy report + Refresh report buttons. This is
the paste-me-into-support artefact for any diagnosis.
IPC wiring:
main.js
ariadne-get-status -> GET http://127.0.0.1/api/status ({ok, status|error})
ariadne-get-policy -> read C:\ProgramData\Ariadne\policy.json (or {})
ariadne-set-policy -> merge {policy}, write back (validates enum)
ariadne-set-source -> merge {sources.<name>.enabled}, write back
(validates against the known 5 names)
settings-preload.js
ariadneGetStatus, ariadneGetPolicy, ariadneSetPolicy, ariadneSetSource
All four handlers write policy.json as the local user; no UAC. Works because
install.ps1 grants BUILTIN\Users Modify on the file (0.1.7+).
Sub-page auto-refreshes state every time it opens (listens on the existing
'section' custom event dispatched by showSection).
Not building/shipping Theseus here -- this rides the next Theseus release.
Panel gracefully handles: daemon down (shows 'Daemon unreachable' with a
pointer to the Status toggle), localApi disabled (daemon returns 503, panel
shows the error), missing policy.json (all sources default to true).
Privacy › Location is three modes now: Show real, Hide, Manual. Manual reveals a
50-country dropdown whose pick becomes the coordinates navigator.geolocation
returns to pages — country-capital granularity, no regions or free-form cities.
Old profiles on the retired "Spoof (region)" auto-migrate to Manual + the
region's representative country on first open, so nothing breaks.
VPN row in Privacy stops opening the wrong add-on: the sidebar now no-ops on a
specific panelId that isn't registered (used to silently substitute panels[0],
which surfaced Aegis whenever the VPN add-on was disabled), and the row hides
itself when vpn:main isn't in the sidebar panel list.
Language picker (globe chip menu + Settings › General › Website language) drops
the BCP-47 tag from every visible label — the tag surfaces only as the 2-letter
chip in the URL bar once picked. "English" is the UK original; the US variant
row is retired (same 2-letter chip, ~same text). Ukrainian dropped from the
quick list too. "Automatic" reads as the OS language name (Intl.DisplayNames)
instead of a raw en-US style tag.
A globe chip next to the URL-bar star shows the language sites see you in
(Accept-Language + navigator.language) — "AUTO" while following the OS locale,
the two-letter code once you pin one. Click opens a 23-language menu; the same
setting has a friendly row at the top of Settings › General. Both write to the
existing languageMode/languageValue and stay in sync with the Anti-fingerprinting
Language row through a settings-update broadcast (settings.html and chrome.html
both react live).
Settings › Plug-ins is now two compact rows — one per plug-in — with the on/off
toggle on the right and the update controls beside it. Clicking a plug-in's title
opens its own sub-page (plugins/ariadne, plugins/aegis) with the full description
and the Uninstall button, so the main list stays scannable and dangerous actions
stop travelling with the everyday ones. The Ariadne toggle and its sub-page
mirror the same scheduled-task state.
Went looking for a performance fix and found a correctness bug underneath it.
getTx() slims each transaction on the way into the cache, and the slim shape
kept only { value, scriptHex } — dropping vout.tokenData. That field is where
the server reports CashTokens. It is NOT in scriptPubKey.hex, which Fulcrum
returns with the token prefix already stripped. So the classify pass fetched
one transaction per UTXO, looked for a prefix that was never there, and
concluded "no token" every single time.
Measured against a real chipnet wallet (130 UTXOs, 91 of them token-bearing):
the old pass cost 54 requests for that address and found 0 tokens; sampling
12 of those transactions, exactly 0 had a scriptPubKey starting with the
token prefix. On the faucet-fed address with 28,289 UTXOs it was ~28k
requests, still finding nothing — which is what made the wallet look hung.
So HD BCH wallets have never shown CashTokens. 0.15.0 fixed the imported
adapter, which reads token_data off listunspent, and I took the HD path's
silence for an empty wallet.
Now: when the server negotiated protocol >= 1.5, listunspent carries
token_data and a UTXO WITHOUT it is definitively not token-bearing, so the
whole set is classified from the one call we already make — 1 request instead
of 28,289. Below 1.5 the fallback fetches parents as before but reads
vout.tokenData (present even at 1.4) and only decodes a prefix as a last
resort, so it is correct now too.
Both routes are reconciled onto one shape. They speak different dialects:
the decoder yields a numeric capability (0/1/2) labelled
immutable/mutable/minting, while Electrum sends a string and calls 0 "none".
Left alone, an identical UTXO would have described itself differently
depending on which server answered. The decoder's vocabulary wins, and the
Certificates pane treats immutable as the quiet default so only capabilities
that change what the holder can do get a tag.
txCache is versioned and dropped once: entries written by the old shape carry
no token information, and an absent field cannot be told apart from "no
token", so a warm cache on a pre-1.5 server would have reported a token
wallet as empty.
Verified against the live wallet: one request, 91 token UTXOs, 46 categories,
17 assets, 51 certificates — matching what the panel reports — with
capability labels normalised (5 immutable, 28 mutable, 18 minting).
Assets was one list holding two different kinds of thing. A fungible balance
answers "how much do I have"; a non-fungible one answers "which ones do I
hold", and they want different rows — an amount against a symbol, versus a
commitment and a capability. On a real chipnet wallet that meant 46 rows
where 32 of them were only ever going to say "1 NFT".
Split by what each category actually holds, so a category carrying both
appears in both — which is honest, because it really does hold both. Assets
counts categories; Certificates counts individual certificates, since "3"
should mean three things you hold rather than three groups.
Certificate rows carry the commitment, because it is the only thing that
distinguishes two certificates of the same category, and a MINTING or MUTABLE
tag, because "can still issue others" versus "is fixed" is worth seeing
without opening anything. A `none` capability is left unlabelled rather than
adding noise to every row.
"Certificates", not "NFTs": these are membership, licence and record tokens,
and NFT carries collectible-market baggage that misdescribes them. Ids and
data keys stay `nft` — that is the CashTokens protocol field, so it is
protocol name inside and product name on screen.
Two things caught while building it. `draw({})` was overwriting the "no
assets" empty state with an empty string, which is now the common case since
purely-non-fungible categories no longer appear there. And certificate rows
printed the category twice, as the title and again on the right — the right
column now only carries it when the row has a real name to lead with.
Metadata is looked up for every category, not just the fungible ones, so
naming a token also names its certificates.
The Receive tab carried two stacked toggle rows: Receive/Consolidate above
Address & QR/Assets. That is a lot of chrome for a 380px panel, and the two
rows were not the same kind of thing — Address and Assets are views of the
wallet, while consolidating is an action on its UTXO set. Consolidate now
sits with the other address actions, beside "Next unused address", with the
sibling count on the button.
The old toggle was also the only way out of the consolidate view, so that
view gains its own "← Back to address". Entering still resets the inline host
so the preview is costed fresh, which is what the toggle did.
The Send tab keeps its Send/Consolidate toggle: there, consolidating really
is an alternative way to send, so a toggle is the right shape.
Removed the two now-dead [data-rcv-mode] blocks rather than leaving selectors
that match nothing.
Verified by real visibility rather than the hidden attribute — a child of a
hidden parent keeps hidden=false, which made a first check look like both
panes were showing at once. rcvNormal and the address pane go away while the
consolidate pane shows, and Back restores them.
The public repos carried no license, so nobody could legally copy or build
on the code, and the whitepaper's "free software" had nothing behind it.
Theseus and its companions take the Mozilla Public License 2.0, the
file-level copyleft Firefox and Brave use, which is compatible with every
component they bundle. The resolver and gateway libraries take Apache-2.0
so that other implementations of the registry can reuse them without
copyleft in the way. The protocol documents and the whitepaper are CC BY 4.0.
A third-party notices file lists what the browser ships and fetches, with
the source offer the GPL sing-box binary the VPN add-on downloads requires;
the matching source archive is now published beside the binaries. The
names and marks are reserved in TRADEMARKS.md, separate from the code
license, so a fork must ship under its own name. Settings › General says
the license and links the three files; the whitepaper says the same.
Assets listed but every row read as a hex string. Three separate reasons.
**Both default registries were dead.** Checked 2026-09-29:
raw.githubusercontent.com/cashonize/registry/main/bcmr.json returns 404, and
bcmr.salemkode.com does not resolve. So no token resolved a name on any
chain, mainnet included — and the panel's `.catch(() => {})` meant the
failure was completely silent, indistinguishable from a token nobody has
registered. Replaced with OpenTokenRegistry, which answers and carries
chipnet identities. A dead registry is now reported instead of swallowed:
the reply says whether any registry answered, and the hint says so.
**The tokens in question publish no metadata at all.** Not a wallet problem
and not fixable by any registry list: their genesis transactions carry no
OP_RETURN whatsoever, so there is no BCMR authchain to follow and no
registry entry to find. Their names exist only inside the app that minted
them. So a token can now be named locally, per category, stored under
bcmr/local/<cat> and taking precedence over any registry — with a YOURS tag
so a self-assigned name is never mistaken for a published one. Clearing both
fields removes it and lets a registry entry show through again. Optional
decimals, because a raw fungible amount with no scale is its own kind of
wrong: 15000 became "150 GMX" once told there were two.
**The unnamed row printed the category twice**, once as the name fallback and
once as the sub-line. Unnamed rows now show the category as the identity and
"unnamed token · N UTXOs" beneath it.
The header comment also promised a bundled static registry fallback for
well-known tokens. There is no such directory and no load path for one; the
comment is gone rather than left to mislead.
Verified against the real module: local names beat registry entries, a
category with only a local name resolves instead of returning null, clearing
restores the registry value, out-of-range decimals are dropped, and the new
default registry resolves a real chipnet identity (OTRC, 6 decimals). In the
panel: naming a token repaints it with the tag and rescales the amount, and a
non-numeric decimals entry is refused with the modal left open.
Two changes to the same idea: one way to do each thing, and the same shape
for every coin.
**Assets is no longer buried under the QR.** The Receive body stacked
address → QR → assets in one column, so in a 380px panel the token list sat
permanently below the fold behind 200px of QR — and a coin holding no tokens
simply lost a section, which made every coin look structurally different.
Address & QR and Assets are now peers behind a persistent chip row. The count
rides on the chip, and a wallet with nothing says "No assets held by this
wallet" instead of the section vanishing. That mattered more than it sounds:
a chipnet wallet with 46 CashToken categories was indistinguishable from an
empty one until 0.15.0, because the only signal was an absent card.
The choice persists across re-renders and wallet switches — comparing token
balances between wallets should not throw you back to the QR on every click.
Switching back to Address redraws the QR, since a canvas sized while hidden
comes out blank and renderReceive only repaints when the address changes.
**The coin row's ▾ network dropdown is gone.** The coin view already carries
real network chips (data-browse-net: they filter the wallet list and show a
per-network count), so the popover was a second control for one job — and the
only one that hid its options behind a click. Chips are the single network
control now and the ticker is plain text again.
Verified in the panel: chips switch panes and track their own state, the
count shows, the empty state reads correctly, the QR survives the round trip,
and no .wchev / .wcname.wswitchable remain in the DOM.
Aegis negotiated a flat protocol "1.4". Measured against Fulcrum 2.1.0 on
chipnet, for a wallet holding CashTokens:
asked "1.4" -> negotiated 1.4 -> 39 utxos, 0 with token_data
asked ["1.4","1.5.3"] -> negotiated 1.5.3 -> 130 utxos, 91 with token_data
So 1.4 did not merely omit the `token_data` field — Fulcrum left the
token-bearing outputs out of listunspent altogether. Ninety-one UTXOs were
invisible to the wallet, along with the BCH sitting in them. That is a
balance-correctness bug, not only a missing Assets card, and it applied to
the HD path too, since lib/wallet.js reads the same listunspent.
Now a [min, max] range: a modern server picks 1.5.3, an older one still
settles on 1.4, so nothing that worked before stops working. The negotiated
version is recorded on the client for diagnosis.
Second half: the imported BCH adapter had no CashToken code at all — it never
set tokenBalances and snapshot() never exposed it, so the panel's Assets card
was hidden for every WIF import however many tokens the address held. A
chipnet test wallet with 46 categories showed nothing. It now aggregates from
the server's own token_data, which costs one call for the whole set rather
than the per-UTXO transaction fetch the HD path uses, and emits the same
serialised shape the panel already reads. Verified against that wallet: 130
UTXOs, 46 categories, 17 fungible, 32 with NFTs, JSON-clean.
Found because the user said their asset "uses a different asset category" and
suggested checking with the explorer. It is ordinary CashTokens; the wallet
simply could not see them. My earlier conclusion that the empty Assets card
was correct came from probing a single address that genuinely holds no tokens
and generalising from it.
Importing a Bitcoin.com seed showed a balance of zero. Two causes, one mine.
Mine: a Copay / Bitcoin.com backup QR is "1|<words>|<network>|<ACCOUNT
path>|…", and 0.13.2 dropped that path straight into a box whose contents are
derived as a LEAF. Deriving the account node itself yields an address the
wallet has never used — for the standard BIP39 vector,
qr96x72dpwrjmg8gtmfemmdhn6u8aqdgnvn4fp2906 instead of
qqyx49mu0kkn9ftfj6hje6g2wfer34yfnq5tahq3q6. An address with no history, so:
zero. An account path now extends to /0/0 instead of being derived in place.
The deeper one: a seed import mounted on the single-address adapter, which
watches exactly one scripthash. Bitcoin.com, Electron Cash and the rest
spread funds across a whole BIP44 account, so even with the right leaf the
balance only shows if it all happens to sit on the first receive address.
A seed import is an HD wallet and now mounts as one — the same BchWallet the
vault-derived wallets use, whose WalletKeys walks receive AND change to a gap
limit of 20. That is what actually finds the money, and it brings real spend
support to seed imports as a side effect.
WIF imports are unchanged: one key is one address, nothing to scan.
Existing seed imports are picked up without re-importing. accountPath is
stored in either shape — older imports kept the full leaf, the QR carries the
account — and the mount trims both to the last hardened element before
handing it to WalletKeys. The stale display address stored at import time is
irrelevant, since the panel reads the address off the adapter's snapshot.
Mounting now reads the signer up front to decide which adapter to use. A
locked vault still mounts watch-only from the stored address rather than
failing, and the WC manager gets its own copy of the root because it keeps a
live reference for the per-URI relay-identity HKDF.
Auto-detect already worked — the backend returns
detectedLanguage:{language,confidence} and the panel put it in the
status line. But that only appeared after a translation had already
run, in small text, away from the control that raised the question. You
could not tell what "Auto-detect" had decided before committing to it.
The first row of the source select now says "Auto-detect · German", and
it says so while you are still typing. Detection runs on its own via
LibreTranslate's /detect, debounced 700 ms and gated at 12 characters,
because a detector given two words is guessing and firing per keystroke
would pound a public mirror for nothing. It chains the same mirror
fallback as translation, so a dead primary does not make detection look
broken while translating still works.
Confidence below 60 renders as "German?" rather than silently asserting
a coin-flip. The Google backend has no detect-only route, so there
doDetect returns null instead of burning a request, and the label is
filled from the translation response — which every backend returns
anyway, so a skipped or failed detect is never worse than before.
Detection retires when a source is named explicitly, comes back on
returning to Auto-detect, and is wiped by clear. A right-click
selection schedules one too, since that text arrives with no keystroke.
Verified against the real mirror (de/fr/ja at 100/100/90%) and in the
harness: short text fires nothing, long text fires once, four rapid
edits debounce to one call, and every transition above lands.
Also adds the xray removal script used to take the old engine off all
three exits now that they run sing-box.
Nothing handled HTML fullscreen. Electron put the window in fullscreen for a
page (a video player) with the toolbar still on top, and when the page left
fullscreen while its tab was hidden, or the tab was switched away from or
closed, the window stayed fullscreen: no title-bar buttons, the taskbar
covered, and no key to get out. Tabs now report entering and leaving
fullscreen; the toolbar and sidebar make way for the page; switching or
closing the tab ends it and tells the page; F11 toggles a fullscreen with
the toolbar kept and doubles as the way out. A page's own exit is left to
Electron, which has already taken the window out by the time it tells us;
exiting again during that transition brought the window back maximized.
It was a native <select>, which can only show text, so each option carried
an emoji in front of the name; after the engine icons moved into the build
that was the one place still showing emojis. The control is now drawn by
Settings with the same icons as the rows below, grouped like the toolbar
picker, with arrow-key and Escape handling. Choosing a default there also
repaints the toolbar at once: the generic setting write never told it.
A real export read "1|buffalo body coyote poem …" and was rejected: the
classifier only accepted a bare phrase, so a version marker in front of the
words was enough to make a valid seed look like junk.
Wallets wrap the phrase in their own envelope — a version number, sometimes
a derivation path, pipe- or comma-separated, sometimes JSON. Rather than
guess which wallet produced it, the payload is now split on every run of
non-letters (spaces survive, since they separate the words) and any
BIP39-shaped run in the pieces is taken as the phrase. A derivation path
found anywhere in the original string is carried over too.
Being tolerant of the wrapper does not loosen what counts as a phrase: the
pieces must still be 12/15/18/21/24 words of 3-8 lowercase letters, so a URL
or an address breaks into single-word pieces and matches nothing. Verified
that the phishing-URL and address cases still fill no field.
The path only lands in the box when the box is empty. That field is
prefilled with the coin's default and may have been edited, and silently
changing which addresses get derived is what lost funds look like; if a path
is already there and differs, the message names both and leaves the choice
to the user. The unrecognised-payload preview also grew to 90 characters,
since 48 truncated the evidence needed to tell what a rejected QR actually
contained.
Every engine icon was an <img> pointing at Google's favicon service, fetched
again each time the picker, the toolbar or Settings rendered. Offline the
whole list collapsed to the emoji fallbacks, and each open told Google
which engines the user has configured. The catalog's icons now live in
engine-icons/<id>.png inside the app; a custom engine's icon is fetched
once (its own /favicon.ico first, the favicon service as fallback), cached
under the profile, and removed with the engine. Settings no longer falls
through to DuckDuckGo's icon service either. Phind ships no icon: its site
serves none through the bot wall.
The gateway checks a name's host rules before it decides what to serve, so a
blocked or redirected subdomain behaves the same whatever record the name
carries. Theseus only inherited that for names it proxies through the
gateway's /bns/ mount. A name with both s3 and ip — the shape that caused
the 2026-08-13 subdomain bug — would have had its blocked subdomain answer
anyway, because Theseus talks straight to the IP.
It now asks the gateway for the host's verified rule before taking either of
the paths it serves itself, and only for those paths, so an ordinary
subdomain navigation gains no round trip. Verification stays in one place:
the client reads a decision, it does not re-derive one.
The spec catches up with what is implemented — it still described v1 and
called hosts a future idea.
chipnet.imaginary.cash's web explorer is returning 502, so every "Explorer"
button on a chipnet wallet led to a Bad Gateway. Its Electrum endpoint on
:50004 is a separate service and is fine — balances, history and the send
path were never affected, which is worth stating because a dead explorer
link looks like a dead wallet.
The explorer now resolves transactions as well as addresses, so it can back
both links rather than half of them. One search box: 64 hex characters is
treated as a transaction id, anything else goes to the cashaddr decoder so
it can complain precisely. A transaction shows confirmations, total out,
size and timestamp, its inputs as links to the parent transactions, and its
outputs as links to the receiving addresses — in-page, without a reload.
Inputs deliberately show no amount: Electrum does not resolve it, and
fetching every parent transaction to display one number is not worth the
round-trips.
Aegis's chipnet explorerTx/explorerAddr now point there. Same chain data,
read over the same Electrum connection the wallet already trusts, and a page
we can fix ourselves the next time one breaks. Mainnet is untouched.
Import accepts a picture of a QR code. The file is decoded in the panel by
the vendored jsQR, drawn to a canvas: no upload, no network, no camera.
Image, not camera, on purpose. Theseus sets blockCamera + hideMediaDevices
by default and hides device labels from fingerprinting; a camera scanner
would fail silently for everyone until they turned that off globally, and a
wallet should not be the reason a privacy browser gives up the camera. A
photo or screenshot of the code needs no permission at all.
What comes back is classified, never trusted. A QR is opaque to the person
holding it, and "scan this to restore your wallet" is a working phish, so
the decode only chooses which field to fill:
BIP39-shaped (12/15/18/21/24 lowercase words) -> mnemonic field
WIF or 32-byte hex -> private key field
anything else -> nothing is filled; the
decoded text is shown so
the user can see it was a
URL, an address, or junk
Nothing auto-submits. The user reads what landed in the box and presses
Import, and the host handler still does the real validation.
jsQR 1.4.0 is vendored at lib/jsqr.js under Apache-2.0 with its LICENSE
beside it, unmodified and unminified — code that touches seed phrases should
be auditable in the shipped add-on, not an opaque blob. It is 57 KB gzipped.
Verified by round-tripping through the shipped path: Aegis's own encoder
builds the QR, it is rasterised to a real PNG File, and decodeQrFile() reads
it back byte-exact; an image with no code returns null rather than throwing;
and driving the actual file input fills the mnemonic for a seed, fills the
key field for a WIF, and leaves every field untouched for a phishing URL.
DNS over HTTPS through Chromium's secure DNS (app.configureHostResolver),
under Privacy › Network: Default protection (encrypted via the chosen
provider, plain if that fails — the default), Increased protection
(always the provider, never plain) or Off, with Quad9, Cloudflare,
Mullvad, AdGuard or a custom resolver URL. Any DoH mode also turns on
Chromium's built-in resolver, as Chrome does. Silent Mode names never
touch DNS, and Tor resolves remotely through the SOCKS proxy, so
neither path goes around it.
Global Privacy Control, on by default, under Tracking protection: the
Sec-GPC header on every request (added in the one request-header hook
beside the client hints) and navigator.globalPrivacyControl in pages.
Navigation base, the way Firefox does about:preferences#privacy: the
address bar follows the Settings page (theseus://settings/privacy) and
the hash mirrors it, so every page has a link; a page can have
sub-pages (theseus://settings/privacy/exceptions) with a breadcrumb and
a back arrow; open-settings and theseus:// links accept the two-level
slug.
Privacy now reads top-down: a "Theseus is on guard" card (Shield and
its running total, cookie pop-ups answered, Tor state, version), then
Tracking protection with the Shield and Cookie Pop-ups cards moved here
from Performance and a Manage exceptions sub-page listing the sites
each add-on was told to leave alone (remove to protect again), then
Device access, Anti-fingerprinting, Network (Tor switch and the VPN
panel) and Browsing data. Performance is about resources again.
Every client hard-coded the chipnet beacons, address prefix and electrum
servers on its own: resolver, registrar config, wallets, gateway, indexer,
mirror scripts, the browser bundle and the mobile Java. A mainnet launch would
have meant finding all of them and hoping none was missed.
The table now lives in resolver-web.js, the one file every client already
shares, so it stays a single-file drop-in. BNS_NETWORK selects the record;
unset means chipnet, so nothing changes today: the live index resolves the
same 60 names and 20 TLDs, the 67 offline tests pass, and the dashboard,
market and studio load the same values through BNS.NETWORK.
The mainnet record carries the verified public servers, the prefix and its
own Sia bucket, but its beacons, start height and operator address are
deliberately null: requireBeacons() refuses to scan until they are pinned in
the order ROADMAP-MAINNET.md §6 requires. Bns.java reads a generated
BnsNetwork.java so the phone cannot drift from the desktop clients.
NETWORK-CONFIG.md records what reads the table and what a launch still pins.
Both selects grow a "Frequently used" optgroup above "All languages",
holding up to ten entries ranked by how often each language has actually
been translated to or from.
Counted on a successful translation, not on a dropdown change: picking
your way down the list looking for something would otherwise rank every
language you skimmed past as highly as the ones you work in. A
detected source counts too — with Auto-detect on you never pick that
language explicitly, but it is one you read.
Ties break on the language's display name so the order is stable rather
than dependent on object key order. Counts live in the existing uiState,
so they persist through the saveUi/loadState path already there, and the
group only appears once there is something to put in it.
Verified in the harness: one translation records exactly one use for the
target and one for the detected source; the cap holds at ten with
fifteen tracked; and a pre-seeded profile comes back with its group,
target and zoom restored, Auto-detect still first in the source select.
Three bugs, all found by driving the add-on in a real Theseus and
watching the egress IP rather than reasoning about it.
1. The generated config used pre-1.11 schema. `sniff` on an inbound and
the `block` outbound type were deprecated in sing-box 1.11 and
REMOVED in 1.13, so 1.14.1 refused the whole file and exited 1.
Routing is now a bare `final`; rule `action` semantics changed in
1.12 and the explicit inbound→outbound rule was never needed.
2. xray-core 26.3.27's REALITY would not complete a handshake with a
sing-box client — and, after ruling out keys (three derivations, a
fresh pair used verbatim), shortIds (explicit and empty), clock skew,
dest reachability, TLS 1.3/X25519 on the dest, and xtls-rprx-vision,
not with a correctly configured xray client either. sing-box against
sing-box works first try. The exits now run sing-box, which is what
the add-on already ships to every client, so there is no longer a
cross-implementation surface at all. Migration script included; it
keeps the port, the SNI and the existing uuid pool and only changes
the Reality keypair.
Worth recording separately: xray's REALITY inbound field is `dest`,
not sing-box's `target`. That was wrong too, independently.
3. leaseEndpoint cached the full vless URL. The Reality key and short id
live inside that URL, so re-keying an exit left every client failing
against a stale copy for the whole 24h lease. It now caches only the
uuid and rebuilds the URL from the current catalogue entry, so a
re-key takes effect as soon as the catalogue refreshes.
Verified in Theseus over CDP: baseline 80.187.100.105, tunnel up
81.31.210.65 (the sm-1 exit), off restores the baseline, and sm-3 is
correctly refused to a free-tier caller.
Found by driving the add-on in a real Theseus rather than reasoning
about it. turnOn() failed with "Silent Mode · 1 is not yet configured
(ready)" — resolveEndpoint required the catalogue entry to carry a
vless URL, but the gateway catalogue deliberately ships none, because a
vless URL is the credential and that endpoint is public. The add-on
predates the key-issuer and was never taught to ask for a lease.
It now POSTs to /api/vpn/session for any ready exit that has no URL of
its own, and caches the lease under its serverId until a minute before
expiry. Baked-in and subscription entries still use their own URL and
never hit the network.
autoHideQuietDock ran only when the sidebar state was pushed; on a
fresh boot the chrome pulls it, so Shield and Cookie Pop-ups showed
in the dock until something re-emitted. Run it on the pull path too.
(cherry picked from commit 4f489303f3591444980f0f456b1672fcbf3dae7a)
autoHideQuietDock ran only when the sidebar state was pushed; on a
fresh boot the chrome pulls it, so Shield and Cookie Pop-ups showed
in the dock until something re-emitted. Run it on the pull path too.
Shield and Cookie Pop-ups are settings more than tools, so their
switches, the cookie mode, the counters and "Update rules" now sit in
Settings › Performance under a Protections heading, driven through the
add-ons' own message handlers (Settings-only IPC). Each card opens the
add-on's panel for the per-site details, and each panel links back to
Settings. The two add-ons start hidden from the toolbar's extension
row (manifest dock:"hidden", honoured once so a user who shows them
keeps them); "Show hidden" on the row brings them back.
theseus://settings and theseus://settings/<section> are now addresses,
so any page or note can link to a Settings page.
Also: a Settings or add-on tab that the user navigates elsewhere stops
counting as that tab, otherwise "open Settings" kept focusing a tab
that no longer showed Settings.
The 380px panel is the wrong shape for anything that needs room. This opens
the wallet as a full Theseus tab, with the sidebar's own furniture — identity,
network, balance, section nav, wallet list — laid out as a left rail and the
tab body given to the selected section.
It is the SAME panel.html, loaded with ?surface=web. No second wallet, no
second copy of 4,600 lines to drift apart. That works because an add-on's own
tab is handed a window.silentmode with the same invoke/on surface as the
sidebar, and addon-msg dispatches it as from:"panel" with the add-on identity
derived from the file:// sender — so every existing handler, including the
panel-only ones, works there untouched. The whole change is a CSS grid behind
one attribute plus a chip to open it.
Details that needed care:
- The QR is drag-sized against a 380px panel and the size is remembered.
Given a 720px column it filled the page, so it is capped on this surface
only; the stored sidebar preference is left exactly as the user set it.
- #drop (the coin picker sheet) is fixed-position and sized for the panel;
it is pinned to the rail instead of covering the window.
- Content columns are capped at 720px so forms and lists keep the measure
the sidebar already tuned, rather than stretching across a monitor.
- Under 900px the grid falls back to the stacked layout, so a narrow window
degrades to what the sidebar already does.
- The "open full screen" chip hides itself on the full-screen surface, so it
cannot open a second copy of itself.
Everything is scoped to [data-surface="web"], so the sidebar is byte-for-byte
unchanged. Verified both surfaces, the narrow fallback (by exercising the
real media rule) and zero horizontal overflow.
The aegis.x/app URL still needs a main.js route in Theseus; this ships the
destination over the add-on channel first.
Install-this-app, remove-app, the extension install flow (install,
already installed, installed, not found, failed) and the add-on restart
question were still bare OS message boxes titled "theseus-navigator"
after page dialogs moved to the sheet. askSheet() is a drop-in for
dialog.showMessageBox with the same options and result: title as the
headline, detail under it, the caller's buttons with the default first,
an optional checkbox, and the app's or extension's icon when there is
one. Tone follows the box type (error, warning) or the wording. The
native box stays as the fallback when the browser window is not there,
and for the two synchronous cases (beforeunload, app windows).
Each chip was a fixed 150 px, so the newest ones ran out of the row
and out of sight. Chips now take the full width while there is room
and shrink together as more are saved, down to an icon and a few
letters, before the row overflows.
A bundled add-on that answers cookie consent dialogs, rejecting all but
the essentials by default (or accepting, if the user prefers the banner
simply gone), so pages open without one. Built on DuckDuckGo's
autoconsent (MPL-2.0): its rule bundle covers hundreds of consent
managers, and a reject-button heuristic handles unknown banners in
reject mode. The library runs through the page-inject slot in every
http(s) frame's isolated world; the add-on hands each frame the user's
settings and the rules, and counts what was handled per site for the
panel, which also excludes a site with one click.
build-inject.js assembles inject.js from the library in node_modules
plus the glue, and copies the compact rules and licence into the add-on
so a rule update can ship through the add-on channel.
Host side: page-inject scripts get theseus.evalInPage for the few rules
that need the page's own JavaScript (they already reach the page via
contextBridge, so no new trust tier), and api.tabs is open to
page-inject add-ons as well as request-filter ones.
Theseus had no content blocking at all. Shield blocks requests to known
tracking and advertising hosts on every site, using EasyList and
EasyPrivacy through Ghostery's adblocker engine (the matcher those lists
are written for). The lists ship inside the add-on so blocking works
from the first launch, offline; the compiled engine is cached under the
add-on's data dir (a 22 ms load instead of a 500 ms parse), and the
lists refresh from their publishers about once a day.
The panel shows what was stopped on the current page, a one-click
allow for the site, the global switch, the running total and the rule
versions with an "Update now". Network filters only for now: a blocked
request never leaves the browser, but leftover empty ad boxes are not
hidden yet.
Host side: a "request-filter" capability. Chromium allows one
onBeforeRequest listener per session, so main owns it and consults the
add-ons' filters; a top-level navigation is never blocked, only http(s)
subresources are offered. api.tabs (active tab and a change event) lets
the panel show per-site numbers without seeing page content.
Three complaints, one cause between the first two.
A line of a PDF is rarely one run. pdf.js splits it wherever the file does
— a font change, a kerning adjustment, a colour change — so a heading can
be three spans and an invoice line ten. Replacing the span under the
cursor covered a fragment and left the rest of the line standing, which is
exactly what a replacement that looks like a copy laid over the original
is. A run is now the whole visual line: the spans that share its baseline
and sit close enough to be spacing rather than a second column, with the
spaces the geometry implies put back between them.
And that line is edited on the page. The dialog that used to hold a copy
of the words is gone: the cover goes down first, carrying the line's own
words at the page's own size and colour, and the caret opens on it. The
cover keeps its words hidden while you type, so the original never shows
through the thing covering it. Escape with nothing changed lifts the cover
again and leaves the page as it was found — no mark, no undo step.
The rotate grip was a square like the resize handles, wearing the open
hand that means drag-the-page. It is a disc with a turning arrow now, and
a cursor drawn to match, since no standard cursor means turn. The inline
style that was defeating the stylesheet is gone with it.
Ctrl and the wheel zoom, about the pointer rather than the top-left, so
the words you were reading stay where they were. A plain wheel still
scrolls.
Records what the gateway routes do, the four things that still have to
happen on the servers before the dropdown can light up (UUID pool per
inbound, the pool file, X-Forwarded-For on the vhost, deploy-from-repo),
and — deliberately at the same prominence — that expiry is bookkeeping
rather than enforcement until xray's handler API is wired per box.
Also flags the per-client byte cap as an unverified cheaper mitigation,
marked as needing a test rather than written up as though it works.
Drops the meridian ellipse so the globe is just a round field, and
spends the space that frees on the glyphs: tiles grow from 11.6 to 13.4
units and the A from font-size 10 to 12.4, with the 文 strokes
thickened to match. Checked the same way as last time — rasterized at
true 16px and 18px and magnified nearest-neighbour — against three
candidates (plain globe with tiles, letters bare on the globe, and a
single tile). Tiles won: bare glyphs on blue lose too much edge
contrast at 16px, and the two-tile arrangement is what reads as two
scripts rather than one word.
Same mark in the manifest and in the panel header.
A band starts from the space between things, so that a drag across words
still selects words to copy. On a page that is wall-to-wall text there is
no such space, and the gesture simply did nothing — which is how it failed
on a real document at 276% zoom, where every candidate starting point had
a line of text under it.
Alt-drag starts a band anywhere, text or not. The select hint says so.
Select, Text, Highlight, Draw, Sign — labelled, in that order, at the head
of the toolbar. Each one leads a kit rather than hiding it: the shapes sit
behind Draw, underline and strike-through behind Highlight, and nothing is
offered twice.
Draw arms the pen, which is what Draw means when there is one button for
it. The names drop out below 1180px, where five of them would start
pushing the zoom and colour controls off the end.
The sheet now follows the notification-card pattern: a tone icon in a
tinted circle, the message in bold under a "who says" caption, a
tinted primary action and a quiet Cancel, close in the corner. The tone
is read from the message — delete/remove/error reads as destructive
(red, and the confirm button says Yes), unsaved/required/leave as a
warning, saved/completed as success, anything else as plain info —
since a page hands over only a sentence.
Also decide "who says" from the sender's current URL rather than the
tab's prov, which lags a navigation: a tab that had just left the home
page for a site was still labelled Theseus.
Auto-translate: changing either language only called saveUi(), so the
pane below kept showing the previous language's result with nothing to
say it was stale — you had to notice and press Translate. Both selects
now re-run the translation, guarded on empty input and on
source === target (which would only echo the input back).
Text size: the panes were 13.5px, small for reading a paragraph in a
language you don't know well, which is the entire job. Base is now
15px, driven by a --tsize custom property so both panes stay matched.
Zoom: −/+ either side of a percentage in the header, 70–220% in steps
of 10, clamped with the buttons disabling at each end. Click the
percentage to reset. Ctrl/Cmd with +, - or 0 does the same from either
pane. Persists per-machine alongside the language choice.
Icon: a globe with an A tile and a 文 tile, replacing the 🌐 emoji that
was indistinguishable from every other globe in the dock. Drawn against
the 16px and 18px rasterizations rather than at a comfortable size —
the first pass used a font glyph for 文 and it turned to grey mush at
16px, so the strokes are hand-drawn paths thick enough to survive. Also
drops the hardcoded icon in registerSidebarPanel, which would otherwise
shadow the manifest's mark.
Verified in a harness with a stubbed host API: language change fires
exactly one request, the two guards fire none, zoom clamps and persists,
and the layout holds at sidebar width.
alert / confirm / prompt from a page came up as bare OS message boxes
titled "theseus-navigator" (the package name), with no hint of who was
asking and nothing of the browser's styling — the PDF Editor's "Delete
signature?" was the reported case.
The session preload replaces the page's three functions with wrappers
that hand the call to the isolated world through a DOM event, which
asks main synchronously and writes the answer back; pages see Chrome's
return values (confirm → boolean, prompt → string or null) and no new
global. Main answers from a sheet hanging under the toolbar, in the
same surface as add-on approvals, that names who is asking: the site's
host, the add-on's name for an add-on page or panel (identified by its
path under the profile's extensions directory), or Theseus for its own
pages. The sheet belongs to the tab that asked — hidden while another
tab is in front, back when its tab returns — and a closing tab or
window answers "cancel" so no renderer stays blocked. Windows without
the chrome (installed apps, plain windows) get a native box with a
proper title, and app.name now reads "Theseus Navigator" for whatever
else still shows one.
A wallet imported from a single private key cannot do WizardConnect, and no
amount of work inside Aegis changes that: the handshake ships BIP32 xpubs so
the dapp derives addresses without further round-trips, and a lone key has no
chain code to build one from. Manufacturing a parent whose child equals a
given key means inverting HMAC-SHA512, and even solved for index 0 the dapp's
next index lands elsewhere. The way out is to stop being a single-key wallet.
Promote derives a fresh wallet from the vault on the import's own network and
sweeps the key into it, after which WizardConnect works — and so does every
other thing that assumes a key tree. It reuses plan() + signAndBroadcast(),
the same pair the consolidate flow already spends through, rather than
growing a second money path.
Three things it deliberately does not do:
- The preview costs the sweep by planning a send-max to the wallet's OWN
address, so cancelling leaves nothing behind. Same inputs, same single
P2PKH output, so the fee is identical to the real sweep.
- The imported key is kept, not deleted. The sweep is unconfirmed when the
call returns and anyone holding the old address can still pay into it;
removing the key there would strand those coins. Removal stays a separate
step the user takes once the balance reads zero.
- A promote that fails in plan() takes the just-created wallet back out,
since nothing was broadcast. Past that point the wallet is kept even on
error, because a transaction may already be on the wire and its
destination has to stay visible.
BCH only: the BTC/DGB/ETH/TRX/SOL imported adapters still throw "read-only"
from plan(), and the refusal now names the chain instead of failing vaguely.
Wallet creation is lifted out of the addWallet handler into
createVaultWallet so promote builds its destination through exactly the same
purpose allocation, legacy-purpose carry-over and id numbering as the Add
flow, instead of a near-copy sitting next to a transfer.
Select does what selecting does in every editor people already know.
Double-click bare page and a caret opens there; double-click the document's
own words and they open for replacement; what is selected copies with
Ctrl+C, pastes with Ctrl+V and goes with Del. Nothing was taken away — a
drag on empty page still gathers an area, and a drag that starts on words
still selects words to copy.
Type text and Edit text were two buttons for one question the click already
answers. They are one Text tool: land on the document's own words and it
offers to replace them, land anywhere else and it starts new text. The
words light up under the cursor so which is which is visible before
clicking, not after.
The toolbar says what it is for. Select, Text and Sign are labelled and set
apart; the drawing kit and the markup kit are their own groups. Sign gets a
pen icon over a signature rather than a squiggle that could have been
anything.
Signatures take ink — black, blue, red, green — chosen while drawing and
kept with the signature, because people sign in a particular colour and it
belongs to the signature, not to whichever swatch was armed. And they turn:
a grip above the box, free rotation, Shift to snap to 15°, for the signing
line that is not square to the page.
Two faults the tests found, both invisible by eye:
The rotate grip was drawn in the right place and could not be grabbed —
the selection bar floats directly above a mark, which is exactly where the
grip sits, and it swallowed every click. The bar now stands clear of it.
Undo would not undo a first rotation. Restoring a mark with Object.assign
copies the keys the original HAD, so a property the drag introduced
survived the restore; the journal then recorded the rotated state as the
state to go back to. Restoring now forgets keys the original never had,
which fixes every future property with the same shape.
Also: building a document from pictures or joins refuses to start a second
one on top of the first, and says so rather than failing quietly.
Drag on empty page and a rubber band gathers every mark it sweeps over.
Touching counts rather than enclosing: a band you have to draw right
around a long arrow is a band you draw twice. Shift-click adds or removes
one, Ctrl+A takes the page.
What the group can then do is move, restyle, duplicate and delete, each as
a single undo step — six marks deleted is one thing the user did, so it
has to be one thing to undo. The selection bar offers only what is true of
every member: a group of shapes gets Fill, a group of stamps gets size and
weight, a mixed group gets neither, and none of them gets Edit, which
needs one mark to put a caret in.
No resize handles on a group. Stretching a mixed selection means deciding
what a stretch does to text, which scales by font size, and to a line,
which has no box at all; until there is an answer worth defending,
offering the grip would promise something this cannot keep.
Each member is outlined as well as the group, because a band that caught
one more mark than you meant is worth seeing before you press Delete.
A group lives on one page. A mark carries its page, and PDF user space
means nothing across two of them, so a band selects within the page it
was drawn on.
One trap found on the way: starting the band on a press means calling
preventDefault, which is also what gives a form field its caret — so the
band now keeps its hands off the annotation layer, and typing into a PDF
form still works.
Three things the editor made you work around.
Text was typed into a dialog and then placed, so you chose a size and a
weight for words you could not see against the page they were going on.
The click now opens a caret where you clicked, in the font, size and
colour the words will have, with the style bar over it; double-clicking a
stamp reopens it in place. Lining a CSS line box up with a PDF baseline is
measured from the font's own metrics, not guessed.
A signature lived in a single slot. There was nowhere to keep initials as
well as a name, nowhere to change the one you had, and reaching for the
tool again simply stamped the first one — which is the same fault three
times: one slot. It is a library now, with redraw, rename and delete, and
the choice is made when the tool is picked up, so placing stays one click.
An existing single signature is carried into it rather than dropped.
The mark you had just drawn could be resized by its handles and not moved
by its middle, because only the select tool let marks be hit-tested at
all. The SELECTED mark now takes a press whatever tool is armed. A press
anywhere else still draws, and an unfilled shape is still grabbed by its
outline — the same rule select has always followed.
Also: words default to dark ink rather than highlighter yellow, which was
unreadable on white and is now impossible to miss, since you watch
yourself type it.
