Omostrich (Omarchy + ostrich) is a local Nostr signing and posting tool for the Omarchy desktop. One bar chip lets you unlock a locally-encrypted key, compose and post notes, and act as a NIP-46 remote signer for other Nostr apps — all without any app but this one ever touching your private key.
It is not a Nostr client. There's no timeline, no feed, no replies, no DMs to read. It composes, it posts, it signs on request. That's it.
One icon (id tim.omostrich), an ostrich mark that dims when the
signer daemon is unreachable and shows a small red badge with a count
when something needs your approval.
Unlock. Paste your nsec once to set up; after that, just a passphrase. Pick how long the unlock lasts before you type it — 5 minutes, 30 minutes, 1 hour, 12 hours, or 24 hours — right on the unlock form. Whatever you pick becomes the default next time. You can also change it later from Settings without re-locking.
Compose. A text field that wraps and grows with your note (up to a sane cap, then it scrolls). Enter posts. Shift+Enter adds a newline. Shows a live character count and flags a "long note" past 700 characters — a soft nudge, not a hard limit; relays accept longer notes fine. Posts as a kind-1 text note to every relay you've configured, and tells you exactly which relays it actually reached and which it didn't — not just a bare success/fail.
Attach an image. Optional, needs a Blossom server URL configured
first (see Settings). Three ways in: type/paste a file path, paste an
image straight from the clipboard, or use the real file picker
(zenity, filtered to png/jpg/webp/gif). The daemon uploads the blob
(BUD-01/NIP-98 signed auth), shows you the preview and resulting URL,
and appends it to the note on post — with an imeta tag if the hash,
mime type, and size are all known.
Identity. Once unlocked, the chip shows who you are — your NIP-05
if you've set one, else your kind-0 display name, else your name, else
a truncated npub as a last resort. Fetched from the relay pool after
unlock and cached locally so it doesn't have to wait on a fresh
network round-trip every time. A circular profile picture shows up too,
if your kind-0 has one. The panel never loads that URL itself: the
daemon downloads it (after unlock only, 10 s timeout, at most 1 MB,
PNG/JPEG/GIF/WebP by both Content-Type and file signature, at most
2048×2048 by the file header), strips it down to what drawing it
needs (no text, ICC, EXIF or XMP metadata, first frame only; animated
WebP is skipped) and keeps that as avatar.<ext> in
~/.local/share/omostrich/; the panel loads that file with a small
sourceSize. A picture that fails any check is skipped (logged) and
the placeholder shows instead.
Settings (collapsed by default — one line, click to expand):
- Relays — the comma-separated list you post to.
- Blossom — one media server URL. Empty = attachments are off, and the composer says so plainly rather than pretending an upload happened.
- Alerts — an on/off switch for desktop notifications. While
unlocked and on, Omostrich watches for mentions, DMs, and zaps
addressed to you and fires a toast (via
notify-send, picked up by Omarchy's own notification service) — content is never decrypted for the toast, DMs just say "encrypted message." No history replay, no inbox, no feed — the toast is the whole feature. NIP-17 gift wraps (kind 1059) are backdated by up to two days on purpose, so they're watched from two days back: the ones relays already hold when you unlock are only marked seen, and a new one toasts once when it arrives. A DM that came in while you were locked gets no toast. Ifnotify-sendcan't run, the daemon logs that once and carries on without toasts; it never takes the signer down. - Auto-lock — same 5m/30m/1h/12h/24h picker as the unlock form, so you can change how long you stay unlocked without having to lock and re-unlock to do it.
NIP-46 remote signing (also collapsed, but auto-expands the moment
something's actually pending, so you never miss an approval). Copy your
bunker:// connection string straight from here to paste into any
NIP-46-aware client — Amber, nsec.app, a web app, whatever supports
Nostr Connect. Every incoming request from an app you haven't already
trusted shows up here with the method being requested, and:
- for
sign_event, the event's kind and a content preview (no preview for kinds 4, 13 and 1059, whose content is ciphertext; a kind 4 shows who it's to instead); - for
nip04_encrypt/nip44_encrypt, who it's to (a shortened npub) and a preview of the text being encrypted; - for
nip04_decrypt/nip44_decrypt, who it's from.
The preview is held in memory for the prompt only; it's never written
to the log. Three choices per request: Approve once, Always
allow, or Deny. Always allow is scoped to that exact method for
that exact app, and for sign_event to that exact kind: the button
says "Always allow sign kind 4 (DM)", and that grant doesn't let the app
post a note (kind 1) or change your profile (kind 0) without asking.
A request to sign a kind 14 (a NIP-17 rumor, which must stay unsigned)
or a kind 1059 (a gift wrap, which must be signed by a one-time key,
not yours) shows a warning, and can only be approved once, never
always. When several are waiting (an app
like Amethyst signs a burst of events right after login), Approve
all / Deny all settle every request on screen at once, each one
just this once; one that arrives after you click still waits for its
own answer. Unanswered requests auto-deny
after two minutes; a signer should never sit there waiting on you
indefinitely. Authorized apps and what they've been granted are listed
and revocable. get_public_key is answered without a prompt only for an
app you've approved a connect from (since the daemon started) or that
is under Authorized apps; anyone else asking gets a prompt.
Upgrading from 0.3.2 or earlier: a saved sign_event grant used to
mean every kind. The first time the daemon starts after the upgrade, it
removes those grants (the app and its other grants stay, and the log
says which apps) because there's no way to tell which kind you meant
when you approved it. The app asks again, once per kind, and Always
allow then saves sign_event:<kind>, for example sign_event:4.
Show the bunker as a QR code. QR next to Copy shows the same
bunker:// string as a QR code (black on white, drawn by qrencode)
for a phone app to scan, such as Amethyst's login with a bunker. It's
the easier direction for a phone: its camera has autofocus, and the
bunker string is short. The string is public (your pubkey and NIP-46
relays, no secret), and so is the PNG it's drawn to,
~/.local/state/omarchy/omostrich/bunker-qr.png. The app's connect
request then shows up for approval like any other.
Connect an app by QR code (nostrconnect://). Some apps start the
connection themselves: Amethyst's "login with a remote signer" shows a
nostrconnect:// QR code instead of asking for a bunker:// string.
Under Remote signing, Scan QR opens your laptop camera with a
mirrored preview window (bin/scan-qr.mjs: ffmpeg for the camera,
ffplay for the preview, zbar for decoding) and closes it as soon as it
reads a nostrconnect:// code, when you close the preview or press
Cancel scan, or after two minutes. Hold the phone so the code fills
the box drawn on the preview, about 25-35 cm away, and keep it still.
Amethyst's code is dense and laptop webcams are fixed-focus, so each
frame is enlarged 2x and sharpened before decoding; frames go straight
to the decoder and are never written to disk. Paste code does
the same with a nostrconnect:// string you copied. The daemon checks
the code (hex pubkey, ws:///wss:// relays only, at most 8, a secret
required) and answers the app on the relays it named, which completes
its login. Scanning is your approval of that connection only: the app's
requests after it (sign_event and the rest) show up for approval like
any other. The app is added to Authorized apps under the name it gave,
with its relays, so the daemon keeps listening on them after every
unlock; Revoke removes it. The vault must be unlocked to connect.
About. The button at the top right of the dropdown shows the
About page: version, license (readable in place), Nine Point Labs, the
support address, help@ninepointlabs.com,
with Email, Copy and Copy version details buttons, and exactly
what Omostrich does with your key and your data. Back or Esc
returns; closing the dropdown resets it. It reads only the plugin's own
files, so it works locked or with the daemon stopped.
Hit SUPER + N from anywhere and a small compose-only overlay appears,
already keyboard-focused — no click required. Same wrap/grow field,
same Enter-to-post/Shift+Enter-for-newline, Esc dismisses. If the
vault's locked or unreachable, it says so plainly and points you at the
bar icon instead of pretending to be a working composer. This is the
exact same daemon, same publish command, same everything as the bar
dropdown — just a faster way in when you don't want to click through
the bar first.
The daemon (daemon/daemon.mjs) is the only process that ever holds
your decrypted private key, and only in memory, and only while
unlocked. It:
- Imports an nsec once and encrypts it at rest with AES-256-GCM under a key derived from your passphrase via scrypt (N=2^17 — deliberately expensive, to raise the cost of an offline guess against a stolen vault file).
- Talks over a Unix domain socket (
~/.local/state/omarchy/omostrich/control.sock, mode 0600 in a 0700 directory) — the trust boundary is "anything running as this user can already reach it," same as any local IPC socket.bin/ctl.mjs(installed asomostrich-ctl) is a thin CLI wrapper over it: one JSON command in, one JSON response out. - When Tailscale is up (a
tailscale0interface with an IPv4 address at daemon start), also listens on TCP port 4737 on that Tailscale address, with the same protocol and commands, for remote clients such as a phone. The first line a TCP client sends must be{"auth": "<token>"}, with the token from~/.local/state/omarchy/omostrich/tcp-token(created on first start, mode 0600). Until it authenticates nothing is run; a wrong token, a command, or more than 4 KB first closes the connection, and so do 10 seconds of silence. After that the client can do anything the local socket can, including signing and posting while unlocked and replacing the vault, so treat the token like a password. Delete the file and restart the daemon (systemctl --user restart omostrich) to issue a new one. - Publishes a note to every relay you've configured by resolving each one explicitly and publishing to it directly — not just whichever subset happened to already be connected at that instant, which was a real bug caught and fixed (a partial-connect race could silently skip a relay that hadn't finished its handshake yet).
- Runs NIP-46 remote signing on its own separate relay list
(
nip46Relays), not your posting relays. This matters: a NIP-46 client's very first connection is signed by a fresh, disposable keypair it just generated — not your identity — and an invite-only posting relay that already trusts you will happily reject that unknown throwaway key. Splitting the lists is what makes a real third-party NIP-46 client's first connection actually work. - Never puts a secret on a command line.
omostrich-ctlreads its JSON payload from stdin, not argv — an nsec or passphrase living in argv would be visible to any other process running as you (ps aux,/proc/<pid>/cmdline) for the life of that process. This is verified, not assumed — it was proven by actually inspecting a running process's argv while it held a payload over stdin. - Keeps an audit log (
~/.local/state/omarchy/omostrich/daemon.log) of every signature it produces and every note it publishes — event kind, event id, a truncated content hash, content length. Never the key, never the passphrase, never full content for the log's own sake. Silent internal signing does not reset the auto-lock timer on its own — only real, user-initiated activity (a publish, an approved NIP-46 request) does, so nothing can quietly poll for signatures while also keeping the vault open indefinitely. - Auto-locks after your configured idle window, always. If it isn't
unlocked,
publishrefuses outright with a clear "locked" error — it does not queue, does not retry, does not attempt to unlock itself. - Runs as a systemd
--userservice with real sandboxing:LimitCORE=0(this machine can produce crash dumps; without this a crash while unlocked would leave your raw key sitting in a core file on disk, completely outside the vault's own encryption),NoNewPrivileges=true,ProtectSystem=strict,ProtectHome=read-only,PrivateTmp=true, and explicitReadWritePaths=limited to exactly the two directories it actually needs.
The plugin (manifest.json, Panel.qml, ComposeOverlay.qml,
LockIcon.qml at the repo root) is Quickshell QML for the Omarchy
shell. It never imports a key, never decrypts anything, never touches
the socket protocol directly — every single action shells out to
omostrich-ctl and reads back a JSON response. If you never trust
anything else in this repo, that's the one property worth verifying
yourself: grep the QML for nsec and you'll find it only ever appears
as something typed into a field and handed straight to a subprocess's
stdin, never stored, never logged, never held longer than it takes to
write it to that pipe.
- The QML never holds your key. Not in memory, not on disk, not briefly, not ever. The daemon is the only thing that decrypts it, and only into its own process memory, and only while unlocked.
- Locked means locked. Every signing path checks the unlock state first and refuses with a plain error if it isn't unlocked — there's no back door, no cached signature, no "just this once."
- The trust boundary is your own user account, plus your tailnet.
The control socket has no separate authentication beyond filesystem
permissions (0600, owned by you) — the assumption, standard for local
Unix-socket IPC, is that anything already running as you is already
inside your trust boundary. The TCP listener on the Tailscale address
extends that boundary to devices on your tailnet that hold the token
in
tcp-token; a tailnet device without it gets nothing run. This is why other agents (Hermes, Claude Code, etc.) can post on your explicit request by calling the same CLI you do, and also why the daemon logs every signature it produces — so "who signed what, and when" has a real answer even though the socket itself doesn't ask "who are you." - The shell never loads a remote image. Every
Imagein the QML loads a local file with a boundedsourceSize. The profile picture is fetched by the daemon with a byte cap, timeout and type/dimension check, and cached with its metadata stripped, since Qt decodes PNG text chunks beforesourceSizeapplies (daemon/avatar.mjs); an attachment preview shows the local file you uploaded, never the Blossom URL. The Blossom upload's response is capped at 64 KB. - NIP-46 approvals are per-app, per-method (per-kind for
sign_event), and never blind. Approving one thing doesn't silently approve everything, and you can always see what you're signing, and who you're encrypting to or decrypting from, before you say yes. - Known gap: an app that never connected can still raise prompts.
The bunker string carries no secret, so anyone who knows your pubkey
and NIP-46 relays can queue approval requests (they're auto-denied
after two minutes). Requiring
connectfirst would lock out apps you approved once before a daemon restart, which don't sendconnectagain. Deny anything you don't recognize.
The bunker QR needs the qrencode package; Scan QR needs ffmpeg
(which includes ffplay) and zbar. Install any that are missing with
your package manager; everything else works without them.
Requires Node (any install — system package, mise, nvm, fnm,
volta, whatever puts node on your PATH) and
Omarchy.
git clone https://github.com/ninepointlabs/omostrich ~/Projects/omostrich
cd ~/Projects/omostrich
npm ci # installs exactly what's pinned in package-lock.json
./install-daemon.sh # installs + starts the systemd --user service, symlinks the CLI
./install-plugin.sh # deploys the bar widget + Super+N overlay, adds it to your barinstall-daemon.sh symlinks omostrich.service into
~/.config/systemd/user/, enables and starts it, and symlinks
bin/ctl.mjs to ~/.local/bin/omostrich-ctl (make sure ~/.local/bin
is on your PATH).
install-plugin.sh copies every *.qml file, *.js, *.png,
manifest.json, and LICENSE from the repo root into
~/.config/omarchy/plugins/omostrich/ — including LockIcon.qml,
which is easy to forget if you ever copy plugin files by hand instead
of running the script. Panel.qml's bar-icon iconComponent
instantiates LockIcon directly; without LockIcon.qml present in
the deployed directory, Quickshell fails to draw the chip at all and
you get a blank slot on the bar where the icon should be — this
happened for real on this machine right after an earlier repo rename,
from a hand-copy that only grabbed Panel.qml + ComposeOverlay.qml
ostrich.png+manifest.jsonand missedLockIcon.qml. Don't hand-copy a subset of files — always runinstall-plugin.sh, which globs*.qmland picks up every QML file in the directory (currentlyPanel.qml,ComposeOverlay.qml, andLockIcon.qml) with nothing to remember or leave out. It also runsomarchy bar put tim.omostrichand restarts the Omarchy shell so the change is live.
Re-run install-plugin.sh after any plugin edit — Quickshell doesn't
reliably pick up a changed QML file without a shell restart, which
this script always does at the end.
Add the SUPER + N overlay bind to your own
~/.config/hypr/bindings.lua:
o.bind("SUPER + N", "Omostrich compose", "omarchy-shell shell toggle tim.omostrich")Then open the bar icon, paste in an nsec and a passphrase, and you're set up.
If you ever want to move the bar icon, remove it, or re-add it:
omarchy bar move tim.omostrich --section left # move it
omarchy bar put tim.omostrich --before omarchy.tray
omarchy plugin disable tim.omostrich # take it off the bar
omarchy plugin enable tim.omostrich # put it backTo uninstall:
omarchy plugin disable tim.omostrich
rm -rf ~/.config/omarchy/plugins/omostrich
systemctl --user disable --now omostrich.service
rm ~/.config/systemd/user/omostrich.service ~/.local/bin/omostrich-ctl| What | Where |
|---|---|
| Repo | ~/Projects/omostrich |
| Plugin manifest + QML | repo root (manifest.json, Panel.qml, ComposeOverlay.qml, LockIcon.qml, AboutView.qml, About.js, ostrich.png) |
| Daemon source | daemon/ (nostrconnect.mjs parses scanned QR codes) |
| Encrypted vault + config + profile cache + profile picture copy | ~/.local/share/omostrich/ |
| Control socket, daemon log, notification dedup state | ~/.local/state/omarchy/omostrich/ |
These paths are always under your home directory; XDG_DATA_HOME and
XDG_STATE_HOME are ignored. The panel, the About page and the systemd
unit's ReadWritePaths use the same fixed paths, so the daemon moving
on its own would split them up. To run a second, isolated instance (for
testing), give it a different HOME; setting the XDG variables alone
still points at your real vault.
| CLI | omostrich-ctl (symlinked from bin/ctl.mjs) |
| MCP server (stdio) | bin/mcp.mjs — run directly, no symlink needed |
| QR scanner | bin/scan-qr.mjs — run by the panel's Scan QR |
| systemd unit | omostrich.service |
| Plugin (deployed) | ~/.config/omarchy/plugins/omostrich/ |
| Plugin id | tim.omostrich |
systemctl --user status omostrich.service
tail -f ~/.local/state/omarchy/omostrich/daemon.log
omostrich-ctl statusTwo ways for an agent (Hermes, Claude Code/CLI, Grok, etc.) to post a
note on your explicit ask — MCP if your setup supports connecting one,
omostrich-ctl otherwise. Neither surface ever holds, logs, or accepts
an nsec or passphrase; both are thin clients of the same daemon.
Policy for any agent using either is in SKILL.md —
explicit-ask-only, no auto-posting, never attempt to unlock.
bin/mcp.mjs is a stdio MCP server exposing exactly two tools:
status—{ locked, vaultExists, npub, relays }. No secrets.publish— takes{ content: string }, returns{ eventId, publishedTo, failed }on success or an error string (e.g."locked") withisError: trueon failure. Same daemonpublishcommand the bar plugin uses — signs a kind-1 note and broadcasts to every configured relay.
Nothing else is exposed over MCP. Unlocking, importing a key, changing relays/settings, and approving/denying NIP-46 requests all stay bar-icon only — deliberately not reachable from any agent.
bin/mcp.mjs ships only in the git clone — it is NOT part of the
deployed bar plugin. install-plugin.sh copies *.qml, *.png, and
manifest.json from the repo root into
~/.config/omarchy/plugins/omostrich/ — that directory does not (and
should never) contain mcp.mjs. Installing the bar chip does not
register MCP with anything. Running ./install-daemon.sh (daemon)
and ./install-plugin.sh (bar icon) gets you a working chip and a
running daemon, but Hermes and Claude Code have no idea Omostrich
exists until you separately run the mcp add command below, pointing
at the repo path directly (~/Projects/omostrich/bin/mcp.mjs) — not
the deployed plugin directory, which never has this file. So the full
sequence for a new machine is: clone, npm ci, ./install-daemon.sh
(daemon), ./install-plugin.sh (bar icon) if you want the chip too,
then mcp add per-agent as its own separate step.
bin/mcp.mjs talks to the exact same Unix control socket
omostrich-ctl does — same daemon, same trust boundary, no separate
auth. Unlock the vault from the ostrich bar icon first; a publish
call against a locked vault returns the daemon's own "locked" error
verbatim, same as omostrich-ctl would. No nsec or passphrase ever
crosses the MCP protocol — connecting MCP doesn't change what the
daemon will let you do or see; it's just a second way to call the same
two operations (status, publish) that were already possible from
the CLI.
Connect it in Hermes:
hermes mcp add omostrich --command node --args ~/Projects/omostrich/bin/mcp.mjsThis prompts interactively — Enable all 2 tools? [Y/n/select] — and
if stdin isn't a real terminal (running it from a script, a non-TTY
shell, piped from another tool) that prompt gets nothing and the whole
add is cancelled with nothing saved, silently, no error. Confirmed
on this machine: running the bare command above with stdin closed
prints "Connected! Found 2 tool(s)" and then just "Cancelled." —
hermes mcp list afterward shows nothing configured. Answer the
prompt explicitly in that case:
printf 'Y\n' | hermes mcp add omostrich --command node --args ~/Projects/omostrich/bin/mcp.mjsThat saves it (confirmed: hermes mcp list then shows omostrich with
2/2 tools enabled) and prints its own reminder — start a new Hermes
session before the tools show up. An already-running session doesn't
pick up a newly-added MCP server on its own.
Connect it in Claude Code / Claude CLI:
claude mcp add omostrich -- node ~/Projects/omostrich/bin/mcp.mjsNo API key, no network endpoint, no daemon restart needed to connect either — the MCP process is a separate stdio client of the same running control socket, launched fresh by each agent's own MCP runtime.
omostrich-ctl status
omostrich-ctl publish '{"content":"your note text"}'Response is one line of JSON on stdout:
{"ok":true,"data":{"event":{...},"publishedTo":["wss://..."],"failed":[]}}or, if the vault is locked:
{"ok":false,"error":"locked"}or if the daemon isn't running at all:
{"ok":false,"error":"daemon_not_running"}Rules every calling agent must follow — these are not enforced by the
daemon, they're the deal for using either path at all (also in
SKILL.md):
- Only call
publishwhen the user has explicitly asked that agent, in that moment to post that text. No auto-posting, no scheduled/cron posting, no posting as a side effect of some other task. - If the response is locked, tell the user the vault is locked and to unlock it from the Omostrich bar icon. Do not attempt to unlock it yourself — no agent has (or should ever construct) the passphrase.
- Never print, log, or otherwise surface an nsec. No agent should ever hold one; both surfaces exist specifically so none of them ever need to.
- Not a Nostr client. No timeline, no feed, no reading UI, no DMs, no replies, no quotes, no zaps sent from here.
- Not a general-purpose key manager for other protocols — Nostr only.
- Not a multi-account tool — one vault, one identity, by design.
This project's product name is Omostrich (Omarchy + ostrich) as of
2026-09-03. Before that it was simply called "Nostr" — display strings
changed first (2026-09-02), then the full rename followed: plugin id
tim.nostr-signer -> tim.omostrich, repo directory
~/Projects/omarchy-nostr-signer -> ~/Projects/omostrich, GitHub repo
ninepointlabs/omarchy-nostr-signer -> ninepointlabs/omostrich (renamed
via gh repo rename, git history preserved), CLI omarchy-nostr-signer-ctl
-> omostrich-ctl, systemd unit omarchy-nostr-signer.service ->
omostrich.service, data dir ~/.local/share/omarchy-nostr-signer ->
~/.local/share/omostrich, state dir
~/.local/state/omarchy/nostr-signer -> ~/.local/state/omarchy/omostrich.
The existing vault/config/profile-cache were migrated to the new data
dir, not recreated — same key, same identity, no re-onboarding needed.
Before that, this was two separate icons: tim.nostr-signer (lock/unlock,
approvals, relays) and tim.nostr-compose (a standalone composer that
called into the signer's control socket). Feedback after the composer's
first real use was that a second icon just for typing a note was one chip
too many for what it actually does, so compose lives at the top of this
same dropdown (and, separately, in the SUPER + N overlay).
install-plugin.sh cleans up both that old tim.nostr-compose icon and
the pre-rename tim.nostr-signer install dir if either is still present
from an earlier version.
As of 2026-09-02, manifest.json and the plugin QML moved from a
plugin/ subdirectory to the repository root, and the daemon's source
moved from src/ to daemon/ — the Omarchy plugin marketplace requires
manifest.json at the repository root, which this repo's original
layout (daemon and plugin as sibling directories) didn't satisfy.
Built against the same Omarchy shell plugin contract as
omarchy-hermes-chat — PluginRegistry.qml for the manifest schema, and
the qs.Ui base components (Panel, KeyboardPanel, BarIconButton,
Button, TextField, PanelHero, PanelSectionHeader,
PanelSeparator) those sibling plugins already use. Compose uses Qt
Quick Controls TextArea (qs.Ui has no multi-line field; same primitive
as hey-calendar's journal editor) with qs.Ui.TextField chrome. The
pending-count badge on the bar icon reuses the same small-badge-dot
pattern as TailscaleIcon.qml's warning indicator
(/usr/share/omarchy/shell/plugins/panels/tailscale/TailscaleIcon.qml) —
BorderSurface circle anchored to a corner, deliberately not a novel
pattern. omarchy plugin validate <folder> (ships with Omarchy) checks
the manifest against the current schema if something breaks later.
npm ci
npm test # node --test tests/Checks the About page's facts against the real manifest.json,
LICENSE and first commit, and that its privacy statement names every
relay and host in the code (and NDK's own default relays), and that
the panel's Images load only local files. tests/avatar.test.mjs runs
the profile-picture downloader against a local HTTP server (size cap,
timeout, type and dimension checks, metadata stripping for each
format). tests/approvals.test.mjs covers per-kind sign_event
grants and their migration, what encrypt/decrypt prompts show (with
throwaway keys generated in the test), the kind 14 / 1059 warning, and
get_public_key for unknown apps. tests/notify-watch.test.mjs
checks that a notify-send that can't run doesn't crash anything, and
the gift-wrap window. They read source files only; no
vault, daemon, socket or relay is touched.
MIT — see LICENSE.