New fleet.rs module in houndd: enrollment with token validation,
periodic 30s heartbeat POST carrying agent status, metrics, and
buffered events. Identity persisted to ~/.local/share/hound/fleet.json.
Settings: fleet_url + fleet_token fields (both optional, default off).
Engine trait: verdict_cache_stats() for fleet telemetry.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
A security product whose agent runs as root had no way to receive a private
report. Adds an RFC 9116 security.txt (served at /.well-known/security.txt)
and a SECURITY.md that GitHub surfaces as the repo's Security tab: a contact,
a private-first disclosure ask, the in-scope surfaces, and what we commit to
(signed-before-parse, degrade-safe, no telemetry). Closes the last open
trust gap from the review.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Option C pricing on the page: the Free card now names the recent
malicious-package feed it ships with (not just heuristics), and the Pro card
reframes the feed as the value that is actually paid for — the full 235k
corpus, refreshed daily, new threats within hours of disclosure — plus the
incremental delivery. JSON-LD Free offer updated to match. Design untouched.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two changes to how definitions are distributed, sharing one mechanism.
Incremental updates. The old nightly rebuilt the full per-ecosystem pack
every day, so its hash changed and every client re-downloaded 44 MB of npm
daily. Now the server publishes an immutable baseline plus small daily delta
packs (crates/hound-defs/examples/build-delta.rs); the client — which
already fetches only packs whose sha256 it lacks — pulls the baseline once
and then kilobytes a day. When deltas pile up the builder folds them into a
fresh baseline and drops the old files from the index; the client prunes
whatever the index stops listing, so both the server dir and every client's
defs dir stay bounded. Verified end to end: day-2 fetched only the delta
(baseline untouched), day-3 rebaseline pruned the superseded packs.
Free community tier. Free now gets a recent subset of the public OSV feed
(build-community.rs, ~5,000 newest indicators) so a Free install detects
current threats out of the box — not just the heuristics. Pro is the full
235k corpus, daily/near-real-time freshness, and the curated threat pack.
The client always fetches the community channel and gates the full feed +
threat pack on the licence; a lapse prunes both back to exactly what a fresh
Free install has — community pack + built-in rules — while leaving any custom
.yar the user placed themselves untouched. reload_rules() lets a new threat
pack go live without a daemon restart.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Joe's note: the visual identity is right, the words were not. This keeps the
design exactly as shipped and rewrites the verbiage to be sharper and more
specific — the hero lede, the stat labels, the "what it catches" framing, the
supply-chain / execution / model-file cards, the agents section, and the
pricing line.
Also reconciles the repo with what is actually live: the current hero-video
design, the new pricing ($59.99/yr Pro, $19/seat/mo Fleet with a 3-seat
minimum), softwareVersion 0.1.12 in the structured data, and the referenced
static assets (hero.mp4, favicon, og-image, privacy/terms, icons). Source
links now point at the public GitHub mirror.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Make the product buyable and the open-source claim true.
Licence system, end to end. license.rs was well-designed dead code; wire
it up: an Ed25519-signed token (same key and verify-before-parse discipline
as definition packs), `hound license install`, houndd loads and verifies at
boot, and the execution gate and full supply-chain feed now gate on
Capability checks. Verification failing always degrades to Free, never to a
locked-out security tool; an expired licence downgrades with the reason
shown. Adds tools/issue-license.py.
Hound Linux threat pack. 34 curated YARA rules — miners, IoT/DDoS bots,
backdoors, rootkits, ransomware, webshells, droppers, reverse shells —
shipped through a new signed rules-pack channel (.rpack) alongside the
definitions feed. Every rule is ELF- or size-anchored and keyed on
family strings, never syscalls; the builder refuses to sign a pack that
matches a system binary (the goodware gate caught two bad rules), and a
regression test proves every rule fires on a sample and stays quiet on a
document about malware.
Action signature verification. The composite action claimed Ed25519
verification "against the same signed manifest the desktop agent uses" but
only compared a same-host sha256. It now fetches latest.json, verifies the
Ed25519 signature over the canonical release statement against the pinned
release key, and installs the checksum from the verified manifest.
Licence resolved to Apache-2.0: Cargo.toml, a real LICENSE file, README.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Three tray bugs today, each caused by fixing the previous one without
looking at what else the fix touched. Worth writing down as one thing
rather than three.
**Two dogs, again.** 0.1.4 taught the app to notice its own package
being replaced and reopen. 0.1.5 added the same behaviour to `hound
update`, which is the better mechanism because it works no matter which
version was running — and I did not remove the first. Both fire during
an update: the running app re-execs under a new pid while apt is
working, so the updater kills a pid that has already moved and starts a
second instance beside it.
The in-app restart now waits eight seconds before acting. If the
updater is doing its job this process is terminated during the pause
and never restarts itself; what survives the wait is the only case the
updater cannot cover, which is somebody running `apt upgrade` directly.
Two mechanisms, one arbiter.
**The launcher stopped working.** The single-instance lock added in
0.1.5 to prevent the duplicate icon made a second launch exit quietly,
so clicking the desktop shortcut while Hound was open did nothing at
all — no window, no error, no feedback. It now hands the request to the
running instance, which shows, unminimises and focuses. The same
channel the right-click scan already used: a launch with paths means
"scan these", a launch without means "show yourself".
Also in this release, and the reason the version is what it says it is:
the workspace version did not get bumped. Adding the hound-watch crate
pushed `version =` from line 6 to line 7 of Cargo.toml, and the bump
was a line-numbered sed. The three GUI files went to 0.1.10 and the
workspace stayed at 0.1.9, so the package built as 0.1.9 while the
manifest would have advertised 0.1.10 — every installed agent would
have offered an update forever, installed it, and still seen one
available.
`the_declared_versions_all_agree` caught it before the build finished.
tools/bump-version.sh now does this by matching content and section
rather than line numbers, asserts every substitution actually happened,
and refuses to write invalid JSON.
Adds crates/hound-watch: the registry firehose. npm publishes a CouchDB
_changes stream and PyPI an RSS feed of every new project, both public,
so this needs no user data at all — which is the answer to how a
product with no telemetry grows its detection. Each new package is
handed to the detectors Hound already ships, so there is one definition
of "this install script is hostile" and a rule that fires in CI fires
identically on the firehose. Verified against the live registries: 240
releases in one poll, 60 triaged, parsers tested against captured
responses rather than the network.
476 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Seven detections, all under the free tier. The line Joe and I settled
on is scope rather than capability: an individual protecting their own
machine gets every check at full depth, and what is sold is the same
protection made continuous, enforced, and shared across a team.
**Containers.** The Docker socket is the one that matters. Mounting
/var/run/docker.sock into a container is not access to Docker, it is
root on the host — anything that can talk to that socket can start a
privileged container with the host filesystem mounted — and the
explanation says exactly that. Also privileged: true, host networking,
bind mounts of /, /etc, ~/.ssh and ~/.aws, secrets baked into image
layers (with the part people learn too late: docker history keeps them
after a later instruction deletes them), curl piped into a shell during
a build nobody watches, ADD from a URL, running as root, and COPY . .
with no .dockerignore shipping the .env and the whole .git directory
into a published image.
**Disabled certificate checking**, across nine ecosystems. Reported as
a warning rather than a critical because it is often deliberate, and
skipped entirely in test files — turning verification off in a fixture
is normal, and flagging it there is how a check gets switched off
wholesale. The explanation leads with the trap: the connection still
looks encrypted.
**Prompt injection inside dependencies.** Hound already read the
project's own CLAUDE.md; an assistant working in a repository reads far
more than that, including the README of every package it touches. All
of that is attacker-controlled text, and publishing a package whose
README addresses the assistant instead of the reader costs nothing to
try. The finding names the package, including scoped ones, and resolves
nested dependencies to the innermost package — the one that actually
shipped the file.
An injected instruction file inside a dependency used to produce two
findings, one from each check. It reports once now, as the dependency
finding: "the package awesome-agents ships an injection" is something a
person can act on, and "there is an injection in this file" is not.
**CI beyond GitHub** — GitLab, Jenkins, CircleCI, Bitbucket, Azure,
Drone. The GitLab equivalent of pull_request_target requires both merge
request pipelines and a protected-variable reference before it reports;
merge request pipelines alone are how everybody uses GitLab, and
flagging them would fire on nearly every project.
Also adds crates/hound-api/src/license.rs: an entitlement check so
nothing ships untiered by accident. It does not try to stop anybody —
the binary is Apache-2.0 and the check can be deleted — and there is no
phone-home, so a machine with no network still knows what it bought and
we never learn where our software runs. An expired licence falls back
to Free rather than failing closed: somebody whose card lapsed must not
end up with less protection than a stranger who installed Hound this
morning, and there is a test for it. Every message about an absent
capability has to name what still works, which is also tested.
466 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every existing test asserted the history scan reports nothing — on this
repository, which has never committed a credential. That is the wrong
direction to test alone: a detector exercised only against clean input
is indistinguishable from a function that returns an empty vector, and
this one has already shipped four false-negative-shaped bugs today.
The new test builds a repository from scratch by writing zlib-compressed
git objects directly — no subprocess, for the same reason the scanner
uses none — commits a key, and leaves no working-tree copy at all. Only
the history has it, which is the situation the whole module exists for.
It asserts the scan finds it, names the file, says revoke, and does not
carry the credential's value into the finding. The fixture key is
assembled at runtime so this source file does not itself contain a
credential-shaped string, which is the trap that made the detector
report its own definitions earlier today.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
**The action.** `hound hygiene` and `hound supply-chain` only ever saw
repositories somebody had already cloned onto a machine with Hound
installed. action/ runs them on every push and pull request: it
installs the published .deb, verifies it against the same signed
checksum the desktop agent uses, and annotates findings on the lines of
the files they concern so a reviewer sees them in the diff rather than
in a log nobody opens.
report.py will not print a credential it found — GitHub masks only
values registered as secrets, so anything else in an annotation is
readable by everyone who can see the run and stays in the API
afterwards. And it will not report a partial scan as clean: a history
walk that hits its limit says so, because "no findings" and "no
findings in the part we looked at" mean different things to somebody
deciding whether to merge.
**The background.** A radial gradient was set on `html, body` — both,
each 100% tall — so it painted twice and the seam between the two
layers appeared as a band across the middle of the page when scrolled.
It was also a hardcoded near-black the light theme had no way to
override. Three more like it: the active tab, button hover, and the log
panel. The ground is a token now, and every colour in the stylesheet
comes from one, so no rule can put one theme's text on the other's
background.
**The palette.** The light theme now uses houndav.com's values exactly
— #5A58C8 buttons, #147A3D, #9A6100, #C22222 — so the app and the site
are recognisably the same product rather than two guesses at it.
Then measured rather than assumed, and found two failures Joe had not
mentioned: "faint" text was 2.90:1 in dark and 3.37:1 in light, and the
dark button hover was 4.41:1. All three now clear 4.5:1, and all eight
text pairs pass WCAG AA in both themes.
**And four more places still naming ClamAV**, which has not been the
engine for a long time: the auto-update caption said the daemon runs
freshclam, the update log said the same, an error suggested `apt
install clamav`, and a permissions hint pointed at /var/lib/clamav. The
engine swap replaced the code and left the copy describing software
this product no longer runs.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Malware scanning asks whether a file is hostile. The check that
actually loses people their accounts is a different one: what has
already been exposed, and what is about to be? crates/hound-supply/src/
hygiene.rs answers it, and runs as part of every project sweep.
- A secret file tracked by git. The emergency case: it is in the
history, in every clone, and in every fork. The advice says rotate
BEFORE `git rm --cached`, because removing a pushed secret does not
un-share it, and a test asserts that ordering.
- Credentials hardcoded in source, recognised by issuer format —
AWS, GitHub, Anthropic, OpenAI, Stripe, Slack, GitLab, npm, PyPI,
Google, and the PEM private-key headers.
- Secret files readable by every account on the machine.
- Secret files with nothing in .gitignore covering them: the near
miss that the next `git add -A` turns into the emergency above.
- GitHub Actions: pull_request_target with a checkout of the pull
request (a stranger's code, your secrets, your write token), a
secret echoed into the build log, a downloaded script piped into a
shell, and third-party actions on a moving tag.
Two rules govern all of it. **Findings are actionable**: no entropy
heuristics, because "high entropy string" is a coin flip a human then
has to adjudicate, and people stop reading after the second false
alarm. Every detector recognises a documented credential format or
reports a structural fact that is true or false. **Nothing secret is
copied into a finding** — a report naming the key it found has moved
the key into a log, a CI artefact, or an assistant's context window,
which is the thing being prevented. There is a test for that.
Tracked-file status comes from parsing .git/index rather than running
git: the sweep is pointed at repositories precisely because they are
not trusted, and starting a subprocess inside one is what a hostile
repository wants.
Against a deliberately bad test repository: four critical, five
warnings, and correctly silent on .env.example and actions/checkout@v4
— flagging those is how a scanner gets ignored.
Also in this release:
- Right-click "Scan for Threats with Hound" in Nemo, Caja and
Dolphin, whose menu entries are system files. GNOME Files and
Thunar keep theirs per-user, so `hound context-menu install`
handles those. The icon is symbolic, so the file manager recolours
it to the menu's own theme instead of dropping a violet dog into a
monochrome menu.
- A right-click while the app is already open hands the request to
the running instance rather than refusing. A menu item that
silently does nothing because the app happens to be open is
indefensible.
- Two fixes for the duplicate tray icon. The updater slept a fixed
600ms after SIGTERM and then started the replacement; if the old
process outlived that, the panel kept its item and the result was
two dogs, the older of which could not be clicked or closed because
nothing was behind it. It now waits for the process to actually
leave /proc, escalating to SIGKILL after five seconds. And the app
itself now holds an advisory lock for its lifetime, so a second
instance cannot exist — the kernel releases the lock however the
process dies, so a stale one is not a state that can happen.
402 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every serious bug found in desktop testing had one shape: Hound
reporting success it had not achieved. A build script that said "built"
without building. A window showing "Protected" while its front-end had
failed to load. A rootkit check calling 988 processes hidden when it
was the one that had been blinded. A settings write refused while the
switch stayed where the user put it. A visible error is something a
person can act on; a false green is not.
`hound selfcheck` asks the question directly — what can this
installation not currently do? — and the daemon prints anything wrong
at startup rather than waiting for it to be inferred from behaviour.
The states are ok, degraded, and blind; the last is the one that
matters, because blind means a detector is running and cannot see. It
exits non-zero when blind, so it can be wired into monitoring.
It earned its place within a minute of existing, by reporting the vault
as /root/.local/share/hound/quarantine. The daemon runs as root, root
has a home directory, and the XDG rules therefore sent the system vault
into root's dotfiles — while the installer created and hardened
/var/lib/hound/vault, which sat empty, and the desktop app read the
user's own vault. Three vaults, none agreeing, which is exactly why the
Quarantine tab reported "vault is empty" beside two quarantined files.
Root now uses the system vault; an unprivileged daemon keeps its own,
since a developer running houndd by hand must not need /var/lib.
Also: `hound update` restarts the desktop app itself after installing.
The app can notice its own package being replaced, but only from the
version that learned how — updating from an older one leaves the stale
process showing the old front-end, which is indistinguishable from an
update that did nothing. The updater matches processes on the
executable rather than a command line anyone could imitate, and
relaunches each as its own owner with the session environment it was
already using: DISPLAY, Wayland socket and bus address are taken from
the running process, because guessing them breaks on Wayland or a
second seat.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three reports, one shape: the app told the user something had happened
when it had not.
**Appearance was stored in the wrong place.** theme and tray_icon_style
lived in the daemon's settings, which need root to write. So toggling
to a monochrome tray produced a rejected write, no visible change, and
a switch still sitting where it had been put — the screen disagreeing
with reality. Appearance is a per-user preference: it belongs to the
person, not the machine, and putting it in the daemon would also have
made one user's choice everyone's on a shared box. It is client-side
now, applies the instant the control moves, and never prompts for a
password.
**There was a Save button.** A settings screen with one lets you walk
away with your changes discarded, and here it also hid the failure
above. It is gone. Every control writes on change, batched behind a
short delay so toggling three switches is one request and one
authentication prompt rather than three, and typed fields wait for a
pause rather than firing per keystroke. A write the daemon refuses now
reloads the real settings, so a control never keeps a value that was
not accepted.
**The app did not notice being replaced.** After an update the running
process is still the old binary showing the old front-end, which is
indistinguishable from an update that did nothing — the Appearance
panel was in the installed package the whole time and could not be
found without quitting and relaunching. The daemon restarts on upgrade,
so its reported version is the authority on what is installed; when it
stops matching this process's own, the app says so and reopens. One
attempt only, so a version that never matches cannot become a restart
loop.
Also: the tray repaints immediately on an appearance change rather than
waiting for the next poll, and the stored appearance is applied before
first paint so the window does not flash the wrong theme on the way in.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every pack in the published feed was dated 2026.08.21 because that is
when I last ran the ingest by hand. The client-side update machinery
shipped in 0.1.2 works well against a source that never moves, which
means it would have kept every installation confidently green on data
that aged a day for every day that passed — and the staleness warnings
added in the same release would have started telling users the truth
about a problem we caused.
tools/refresh-definitions.sh runs the steps that already existed
(ingest-osv.py, build-pack, index.json) with the properties a published
feed needs:
- Never publishes an empty feed. If every pack fails to build it
exits non-zero and leaves the previous one live. An agent that
installed an empty feed would report a clean machine with no
indicators loaded, which is worse than one keeping yesterday's.
- Packs are written before index.json, and each lands via rename.
The index is what tells an agent a pack exists, so writing it first
would advertise files that are not there yet, and a rename means a
fetch mid-run never sees a half-written pack.
- Old packs stay on disk; the index advertises only the newest per
ecosystem. An agent that has been offline for a while still has a
URL that resolves.
- One ecosystem failing does not cost the others.
The first dry run built a pack called stage-bfsqqa-2026.08.21.pack:
the staging directory was inside the work directory, and the loop
treats every directory in there as an ecosystem. Staging now lives
outside it, and directory names are filtered as well.
The timer runs at 05:20 UTC with Persistent=true, so a builder that was
off does a catch-up rather than silently skipping a day — that being
the exact failure that produces a stale feed nobody notices. The
service runs as the publishing user rather than root: it needs the
signing key and write access to one directory, and a build pipeline
running as root to write a web directory is a bigger target than the
thing it protects.
Verified: the timer is enabled, a real run republished all six packs
and 235,577 indicators, and a client installed from the result.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A home scan reported two EICAR threats whose paths were Hound's own
vault. SKIP_PREFIXES held the system vault at /var/lib/hound/vault but
not the per-user one under $XDG_DATA_HOME/hound/quarantine, which is a
different absolute path for every user and so cannot be a literal. The
check now matches a `hound` component followed by `quarantine` or
`vault` anywhere in the path, with a test that ordinary paths merely
mentioning either word are still scanned. A scanner that detects its
own evidence locker reports threats that no longer exist anywhere they
can hurt anyone.
Appearance settings, the third pair of fields declared from the start
and wired to nothing (after close_to_tray and auto_update_signatures):
- Theme: follow system | light | dark, previewing live rather than
only on save, since a control that does nothing until you press
another control feels broken.
- Monochromatic tray icon: one glyph instead of the colour ladder.
The tooltip still names the state, so only the colour is dropped.
The light theme had to be built — the stylesheet was dark-only. It is
not an inversion: the state colours are darkened until they hold their
contrast on white (the dark theme's green is 2.2:1 there, unreadable as
text) and the neutrals keep a slight violet bias so they read as chosen
rather than as a default grey. One hardcoded near-black on the log
panel would have been near-invisible in light mode; it is a token now.
Which monochrome tone to draw depends on the panel, and no portable way
exists to ask a panel what colour it is — so the webview resolves the
theme (including "follow system", which only prefers-color-scheme can
answer) and tells the tray. One place decides which theme is showing.
The tray now announces a release once per version rather than once per
poll, and its menu entry reads "Update available — install Hound X…".
Also removes ~120 lines: the GUI had its own download, hash-check and
staging implementation for updates. `hound update` does exactly that,
as root, and is the path with tests behind it, so the app runs it under
pkexec instead. That left `hound stage-update` with no caller, and dead
privileged code paths are liabilities, so it is gone.
Verified end to end: 0.1.2 discovered 0.1.3 and installed it
unattended. 380 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Definitions and the application were two separate acts with two
different mechanisms, and only one of them had a command. `hound
update` now does both:
sudo hound update definitions, then offer the app
sudo hound update -y both, unattended
hound update --check what is available, install nothing
sudo hound update --definitions-only
Definitions run first and independently: a release check that fails —
laptop offline, host down — is a warning, not a failed command. The
definitions half is what matters between releases and it either worked
or it did not, regardless of what the manifest server did.
The app half prompts unless given --yes, because definitions are
verified before they are parsed while replacing the running binary is a
larger step. Non-interactive callers never block on a read: they are
told to pass --yes and exit cleanly.
`release.check` is classified read-only, so any member of the `hound`
group can ask whether they are current — it changes nothing and reveals
nothing the website does not. Installing still requires root and says
so.
Verified end to end on a real machine: a 0.1.1 install discovered
0.1.2, verified the signed manifest, checked the SHA-256, staged it
root-owned and installed it through apt, unattended.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two mechanisms Joe asked for, neither of which existed.
**Definitions now update themselves.** `auto_update_signatures` shipped
from the start and nothing ever read it — definitions only moved when
somebody typed `hound update`, which for a security product means most
installations were running whatever they were installed with. There is
now a scheduler thread: definitions hourly, release check daily, and a
90-second startup delay so a fleet that reboots together does not
arrive at the CDN in one wave. It calls the same install path as
`hound update` rather than a second implementation that could drift.
**Release notifications, signed.** dl.houndav.com/latest.json carries
an Ed25519 signature from the same key as the definition packs, for the
same reason: whoever serves that host must not be able to invent a
version and point our users at a binary of their choosing. A manifest
that does not verify is discarded, and no manifest means "nothing newer
known" — never "update available". The safe default is silence.
The signature covers a canonical byte form defined identically in
release.rs and tools/publish-release.py. Signing a re-serialisation of
a parsed struct is the classic way to verify one thing and act on
another, so a test runs a real manifest produced by the Python tool
through the Rust verifier against the production key. If those drift,
manifests verify nowhere and the only symptom is that nobody ever hears
about an update.
Version comparison is numeric, not lexical, because "0.9.0" > "0.10.0"
as strings and that is precisely the pair where it would first be
noticed. Downgrades are never offered — a signed-but-old manifest must
not become a way to reintroduce a fixed vulnerability.
The app does not replace its own binary. A root daemon that can rewrite
itself is the mechanism a supply-chain attacker most wants. Installing
goes through apt with the user present and authenticating: the package
is downloaded, checked against the manifest's SHA-256, staged into a
root-owned 0700 directory so nothing can substitute it between the
check and the read, and handed over. Two guards beyond the signature —
the URL must be on our own download host, because a signature proves
publisher intent and not that the publisher got the URL right.
Tray, per Joe's design: a new amber "attention" state ranked below a
threat and an in-flight scan, above protected. Deeper amber than
"scanning" so the two are distinguishable — scanning lasts seconds and
the user started it, attention persists. The tooltip carries the reason
rather than only that there is one, and "Install Hound X…" appears
enabled only when there is something to install; a permanently greyed
item teaches people the menu is decorative.
Definition staleness is a first-class signal, not just update
availability: amber at 7 days, and at 30 the wording stops pretending —
"this machine is not currently protected against anything found since
then". An antivirus showing green on month-old definitions is lying in
the same way a dead front-end showing "Protected" was.
Also: the four files that declare a version now have a test asserting
they agree. Drift there means a release looks older than what is
installed and the update silently never offers, or offers forever.
377 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A clean laptop reported 988 critical rootkit findings; this server
3786, PID 1 among them. Every one was false, and the cause was our own
systemd hardening.
ProtectProc=invisible hides processes the daemon does not own from its
view of /proc, while kill(pid, 0) keeps answering truthfully because it
is a syscall and not a filesystem lookup. The hidden-process check
compares exactly those two sources, so with that setting every process
on the machine looked concealed. Enumerating processes is this daemon's
job, so it needs the default view.
Removing the setting is not enough on its own — hidepid= on the /proc
mount produces the same blindness and we do not control that. So the
detector now recognises when it cannot see:
- PID 1 is the control. It always exists and nothing hides init; a
rootkit that did would break the machine it is living on. If PID 1
answers kill(1, 0) but is absent from the listing, we are blind and
say so as info rather than crying rootkit.
- A plausibility ceiling of 32. Hiding a handful of processes is the
entire point of a rootkit; hundreds means a broken observer. An
antivirus that reports a critical rootkit finding on every clean
machine teaches people to ignore the one time it is real.
Also in this change, from testing on a real desktop:
- Closing the window hides it to the tray instead of exiting, with a
one-time notification so it does not read as a crash. Quit lives
only in the tray menu and confirms first. The settings already had
close_to_tray and confirm_quit fields wired to nothing; they are
honoured now rather than hardcoded.
- The tray menu and Scan Home sent the literal string "~". A shell
would have expanded it, nothing here did, so the daemon was asked
to scan a directory of that name. It failed silently until the
per-peer readability check made it audible.
- Administrative actions elevate through polkit instead of telling
people to open a terminal. The app tries unprivileged first and
only on a privilege refusal runs `pkexec hound admin-rpc`, which
forwards one request as root. auth_admin_keep, because prompting on
every settings toggle trains people to authenticate without reading
the prompt. This grants what `sudo hound` already grants to people
who could already run sudo — a transport, not a new privilege.
- `hound settings exec-gate on|off` now exists. The install script,
the AppImage banner, the rpm spec, the AUR install file and
llms.txt all told users to run `hound settings set exec_gate true`.
There was no `set` subcommand and no way to enable the execution
gate from the CLI at all: the flagship paid feature was unreachable
and the first thing a new user was told to type returned an error.
A test now asserts every documented command parses.
- `settings show` displays the exec gate state, and no longer prints
its own header twice.
- The CLI help still described ClamAV, which has not been the engine
for some time. So did the socket permission error, which now
explains the `hound` group and the log-out-and-back-in it needs.
368 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The start-menu entry ran `hound` with Terminal=true — the CLI, which
printed help and exited. No GUI binary had ever been built or packaged.
Four separate faults were stacked behind that report:
1. build-deb.sh now builds and ships hound-gui, and writes a .desktop
entry only when that binary exists. A launcher for software that is
not there is worse than no launcher.
2. Tray icons were loaded from a relative "icons/" path, which resolves
only from the build tree. Installed to /usr/bin the setup hook failed
and Tauri panicked before a window appeared. They are include_bytes!
now — four ~1 KB PNGs that can no longer be missing.
3. The front-end never ran at all. app.js opened with a bare module
specifier ("@tauri-apps/api/core") and there is no bundler, so the
webview could not resolve it and the script silently failed to parse.
The window rendered its static HTML forever, which looks exactly like
a daemon that never answered. withGlobalTauri + window.__TAURI__.
4. build-deb.sh ran the Tauri build as `>/dev/null 2>&1 || true`, so a
config error scrolled past unseen and the package shipped the
PREVIOUS binary. Two fixes appeared to do nothing. That step is no
longer silenced or tolerant of failure, and the build fails outright
on a bare import in gui/dist/*.js.
Guards, because each of these failed quietly: index.html flips to an
interface-error message if app.js never sets a boot flag within 5s. An
antivirus showing "Protected - your system looks healthy" while its own
front-end is dead is the worst failure mode there is.
Then the window came up and could not reach the daemon: the socket was
0700 root:root. Widening it needed more than a chmod, because
quarantine.restore writes files back out as root — handing that to a
desktop group would hand out root. So the daemon now checks SO_PEERCRED
per method (crates/houndd/src/peer.rs):
- group `hound`: status, settings.get, events, quarantine.list,
rootkit.scan, persistence.scan
- scan/supply.sweep: only paths the caller could read itself, decided
by forking a child, dropping to the peer's uid, gid and
supplementary groups, and asking access(2) — which honours ACLs and
mount options, unlike anything reconstructed from mode bits
- everything that writes: root, or the uid the daemon runs as
Unclassified methods fall into Admin, so a new mutating method fails
closed rather than becoming public by omission. The end-to-end socket
test caught that "root only" broke every developer run; the owner
clause collapses to "root" under the packaged root daemon and is
verified to do so.
CAP_SETUID/CAP_SETGID join the gate capability set for the readability
check. There was a test asserting CAP_SETUID must never be retained —
it is updated with the reasoning rather than deleted. The daemon
already holds CAP_DAC_OVERRIDE and CAP_DAC_READ_SEARCH, so becoming
another user widens nothing that matters. The unit gains Group=hound so
the socket can be chgrp'd without CAP_CHOWN; it stays uid 0.
Also: the footer claimed "engine: ClamAV via Unix socket". It reports
what the daemon actually loaded, which has been yara-x since the engine
was replaced. Every error path in the front-end goes through explain(),
so a privilege refusal reads as "run it from a terminal: sudo hound …"
rather than "daemon error -32000".
358 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four things Joe asked for, one of which needed a decision he did not
make.
WHITE BUTTON LABELS. "White on both themes" could not be done by reusing
--brand: white on the dark theme's #9896E0 is 2.70:1, which fails AA
outright and would have looked washed out rather than crisp. Buttons now
have their own background — #5A58C8 light, #5A58D6 dark, the same
periwinkle the desktop app already uses for its primary button — at
5.52:1 and 5.73:1. The label is white in both, which is what makes the
control read as the same object when the theme changes. Contrast was
computed before choosing the colour rather than after somebody
complained.
PRICING. Cards are equal-height flex columns so the buttons line up
regardless of how many features each lists. The featured tier is marked
by a rule along its top edge and a badge rather than a different fill —
it is the same object, just the one being pointed at. Differentiating
features are bold, each tier carries a one-line footnote answering the
question the price raises ("Annual only", "Over 20 seats? We will quote
you"), and the feature text now says what the thing does rather than
naming it: "a malicious binary is refused before it runs, in 2 ms".
COPY BUTTONS are a clipboard icon that becomes a green tick, with
aria-label toggling to "Copied" so it is not a visual-only signal.
GEO AND SEO — the part with real leverage:
- JSON-LD SoftwareApplication with 11 features and all three offers,
and a FAQPage with nine questions.
- A VISIBLE FAQ backing that schema. Schema describing content that is
not on the page is both invalid and dishonest, so the nine questions
are really there, in <details> elements, and they are the questions
somebody actually asks before installing a root daemon.
- llms.txt — a structured summary written for a model deciding whether
to recommend Hound, including a section of honest limitations.
Linux only, no RPM package yet, gate needs root, small rule pack,
full-download definitions. A recommendation that omits those is
worse than none, and a model that repeats them is a model that got
the answer right.
- robots.txt naming GPTBot, ClaudeBot, PerplexityBot, Google-Extended,
Applebot-Extended, CCBot and others explicitly. Hound exists partly
to be used BY coding agents; an agent reading this site should not
have to guess whether it is welcome. Several of those crawlers treat
silence as refusal, and silence is not the same as consent.
- Canonical, Open Graph, Twitter card, theme-color, sitemap.
Every number in the structured data and in llms.txt is measured or comes
from a published feed. No ratings, no awards, no "trusted by" — a model
that quotes a fabricated number does more damage than one that quotes
nothing, because it is confident.
All four assets verified live: HTML 200, robots 200, sitemap 200,
llms.txt 200, and both JSON-LD blocks parse from the served page.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Joe opened the page on a phone and found what I could not: the install
commands pushed the layout sideways, and the header did the same.
ROOT CAUSE, and it was one thing behind both: grid and flex children
default to min-width:auto, so a box holding a long unbreakable string —
a curl command, a terminal block — refuses to shrink below that string
and pushes the whole PAGE sideways instead of scrolling inside itself.
`overflow-x: auto` on the inner element does nothing while its parent
will not shrink. min-width:0 on the children of every grid is the fix,
and it is why the page was fine at 1080px and broken at 390.
The copy blocks also had the button inside the scrolling region, so on a
narrow screen the one control anybody actually uses scrolled out of
reach. The command now scrolls in its own box, the button sits outside
it, and below 720px the command wraps instead of scrolling — the whole
thing visible at once, break-all because a URL has no useful break
points.
HEADER: hiding the nav links on small screens was never a fix, because
the theme switch and the call to action still competed for a 360px bar.
Below 760px the bar now holds the mark and one button, and everything
else moves into a panel: aria-expanded on the control, Escape closes it,
following a link closes it — a panel left open over the thing you just
asked to see is its own bug — and a resize past the breakpoint closes it
so rotating to landscape cannot strand a phone panel over a desktop
layout.
One regression I introduced and caught before it shipped: I had written
`#menu { display: contents }` to make the panel transparent on desktop.
An ID beats a class, so that quietly overrode nav.links's own flex
layout and margin-left:auto and unstuck the desktop navigation from the
right-hand side. The panel styles now live inside the breakpoint and
nowhere else.
I still cannot see this page — the browser here renders with CSS
disabled — so this was reasoned from the cause Joe's report pointed at
rather than from looking. Another pair of eyes on a phone would be worth
having.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Phase 7's first half. A single self-contained page — no build step, no
framework, no dependencies — deployed to /srv/houndav/www and serving.
Downloads published to dl.houndav.com, and both install commands on the
page were run end to end from those public URLs before this commit.
The positioning is the one the product actually earns. It does not lead
with "antivirus", because at $89.95 that reads expensive next to ESET at
$40 and cheap next to JetBrains at $169 — and because the claim would be
wrong. It leads with the fact that turned out to be the whole thesis:
97% of npm's OSV records are malicious packages, not vulnerabilities.
Alongside 235,577 indicators, 2 ms to refuse a binary at execve, and
0.8 ms added to a normal program starting. Every number on that page is
measured, and every one of them appears in a commit message here.
Themes: light is the bare :root, dark is layered twice — once for the
un-stamped "follow the system" state via prefers-color-scheme, once for
an explicit [data-theme="dark"] — so all three of auto/light/dark
resolve as complete sets rather than half a palette. The toggle persists
to localStorage inside try/catch, because a private window throws on
access and a theme switch must not take the page down with it.
The sticky header now sets an opaque background BEFORE the color-mix and
backdrop-filter enhancement, behind @supports. A sticky header that
falls back to transparent puts scrolling content underneath the
navigation, which is worse than having no blur at all.
WHAT I COULD NOT VERIFY, and it matters: the browser available to me
renders with CSS disabled — document.styleSheets.length is 0 with the
<style> element present, every computed style comes back unset and every
font resolves to Times. So I have confirmed the bytes are served
correctly, the CSS parses (balanced braces, no unterminated comments, no
undefined custom properties, no missing semicolons), the HTML tags
balance, the viewport meta is present and there are eight responsive
breakpoints — but I have NOT SEEN THIS PAGE. Somebody with eyes should
look before it is advertised anywhere.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The machinery has been real for a while and the content was 19 crates.io
records. Now it is the whole malicious-packages feed:
npm 220,324
PyPI 11,702
RubyGems 3,512
Go 19
crates.io 19
Packagist 1
────────────────────
235,577 indicators, 45 MB across six signed packs
The striking number is npm: 220,323 of its 227,149 OSV records are
malicious packages rather than vulnerability advisories. NINETY-SEVEN
PERCENT. That is the whole thesis in one statistic — the dominant
security fact about the npm ecosystem is not that libraries have bugs,
it is that the registry is full of things that exist only to be malware.
It is also why the malicious-versus-vulnerability classifier had to come
first: without it this pack would have been 227,149 indicators, and
7,000 of them would have been ordinary libraries.
Verified end to end from a genuinely empty definitions directory:
before: defs: no verified packs were found
update: 6 pack(s) installed, 0 already current
loaded 235577 indicators from 6 pack(s) [2026.08.21]
elapsed: 1.4 seconds for 45 MB, verified and loaded
RSS: 214 MB (unit cap is 1G)
And it detects. A lockfile with 500 ordinary packages and three real
malicious ones — ineldua, dian-kue20-riris, restart-rocket-jasmine-jwt,
picked at random from the feed rather than chosen to work — produced
three criticals and no false positives, in 11 milliseconds. The cuckoo
filter is doing exactly what it was built for: 500 clean lookups never
touch the map.
tools/ingest-osv.py does the coarse filter so build-pack does not walk a
quarter of a million files it will discard. It mirrors the Rust
classifier deliberately, and the Rust side still applies its own — a
record slipping through the Python changes nothing.
`hound status` now reports definitions, and reports them when there are
none, with the reason. Silence would let somebody believe they were
protected by a feed they never received.
Known and not yet solved: every agent downloads 43 MB of npm pack on
first update, and the whole pack again whenever it changes. Incremental
updates and a CDN in front of defs.houndav.com are both wanted. The
subdomain split exists precisely so the second one is a DNS change.
362 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Closes Phase 3's delivery half. `hound update` now asks the definitions
host what exists, downloads what this machine lacks, verifies it, and
installs it. Verified end to end against the live host over TLS:
starting from an empty directory:
defs: no verified packs were found
hound update:
definitions: 1 pack(s) installed, 0 already current
crates-io-2026.08.21.pack — 19 indicators, version 2026.08.21
loaded 19 indicators from 1 pack(s) [2026.08.21]
and immediately afterwards a lockfile naming rustdecimal is flagged,
while tokio beside it is not.
running it again:
definitions: definitions are up to date (1 pack(s))
a pack with one byte flipped in its signed payload, advertised in the
index with a correct hash and a version of 9999.1.1:
definitions: 0 pack(s) installed, 1 up to date, 1 REJECTED
tampered-test.pack: the definitions pack is not signed by Hound and
was discarded
Three refusals are the design:
* THE INDEX IS A HINT, NEVER AN AUTHORITY. It says which packs exist and
what they hash to, and both are unverified — anyone who can serve the
index can lie about either. Only the Ed25519 signature decides whether
a pack is real. A tampered index can waste bandwidth and nothing else,
which is exactly what the test above demonstrates: correct hash,
higher version, still refused.
* NOTHING UNVERIFIED REACHES THE DEFINITIONS DIRECTORY. Downloaded to a
temp file, verified there, then moved with a rename inside the same
directory so it is atomic. The daemon cannot observe a half-written
pack, and a crash mid-download leaves a stray temp file rather than a
loadable one.
* A FILENAME FROM A REMOTE INDEX IS UNTRUSTED INPUT. The updater runs as
root, so an entry of ../../../etc/cron.d/evil.pack would be remote code
execution. Only a plain basename ending in .pack, with no separators,
no dot-dot and no leading dot, is accepted. Tested against eight
hostile shapes.
Sizes and timeouts are bounded — a pack is a list of package names, so a
server offering a hundred gigabytes is broken or hostile and the
difference does not matter to a full disk. The read is bounded
independently of Content-Length, which is also just something the server
said.
Definitions are fetched BEFORE rules are reloaded, and a failure to fetch
does not stop the reload. No network, a mirror down, a pack that will not
verify — none of those are a reason to skip the half that still works.
HOUNDD_DEFS_URL points the client at a mirror, which matters for the
air-gapped deployments that are a real part of the Fleet story.
build-pack now writes index.json beside the pack, so publishing is one
command.
358 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four tools over MCP stdio, so a coding assistant can ask Hound about a
project at the moment it matters rather than after:
hound_check_project the full supply-chain sweep
hound_check_package is this package known bad, before installing it
hound_check_file one file: a model, a lockfile, a manifest
hound_check_mcp_config audit the servers your assistant already trusts
READ-ONLY, permanently. There is no hound_quarantine, no hound_delete,
no way to change a setting. An agent can be persuaded by the very
repository it is inspecting — that is the threat this product exists to
detect — so a destructive MCP tool would hand the attacker exactly the
capability they were reaching for. A test asserts every tool name starts
with hound_check_ and contains none of quarantine/delete/remove/restore/
settings/set/update/install/write/exec, so the property cannot erode.
The output is written to be read twice: by the model that called the
tool, and by the human reading that model's summary. That rules out rule
identifiers and jargon — a test greps the output for both — and it rules
out ambiguity about severity. A model reading "1 warning" may well decide
to proceed, so a critical finding leads with an explicit recommendation
and names the consequence: this runs code during installation, before
any of the project's own code runs.
Clean results say what they did NOT check. "Nothing wrong" that reads as
a blanket endorsement is worse than no answer, so a clean project notes
it is not a source review, a clean name check notes it did not read the
package's contents, and a clean MCP audit still points out that every
server listed runs with your permissions and is called without asking.
The server passes its own audit, which was the point:
{ "mcpServers": { "hound": { "command": "/usr/bin/hound-mcp" } } }
No npx, so nothing is fetched at launch. No env, so no secret is handed
over. No path argument, so it is granted no directory. Hound's MCP audit
flags all three in other people's configs; a security tool that failed
its own check would have answered the only question that mattered.
Verified against the installed binary.
Protocol notes, since both are easy to get wrong and fatal:
- a notification carries no id and must never be answered; MCP sends
notifications/initialized straight after the handshake, so replying
corrupts the stream on the first exchange
- an id of 0 is still an id, and treating it as absent silently drops
the first call from any client that counts from zero
- a tool failure is a RESULT with isError, not a JSON-RPC error: the
agent should see "I could not read that path" as an answer it can act
on, not a transport fault that looks like a broken server
- nothing but protocol messages ever goes to stdout; diagnostics go to
stderr, because one stray println corrupts the session
Shipped in the .deb, and the postinstall now prints the config entry.
334 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Quarantine deletes a file from where its owner put it. Until now every
detection did that, so every false positive was destructive rather than
merely wrong — which on this machine cost an 8.5 MB compiler cache and a
4.3 MB session transcript, the latter's history permanently.
Each rule now declares what Hound may do:
action = "quarantine" move it to the vault
action = "alert" report it, leave it alone
**The default is alert**, and so is an unrecognised value, and so is a
detection name the engine does not know. One misspelt "quarantne" must
not turn an advisory rule into a destructive one across every machine
that updates.
Quarantine has to be earned by an ANCHOR, not by the author's
confidence:
EICAR-Test-Signature quarantine exact 68-byte payload, size-bounded
Linux.Coinminer.XMRig quarantine ELF magic
Linux.Rootkit.Preload quarantine ELF magic
Linux.Webshell.PHP-Eval ALERT content-only — PHP has no file
magic, so it can still match a
security write-up, a log or an AI
transcript quoting a webshell
A test asserts that property directly: any rule declaring quarantine
must contain a file-type check or an exact size bound. A future rule
cannot quietly claim the destructive action without one.
Both the execution gate and the inotify fallback consult it, kept in
step deliberately — a fallback more destructive than the primary path is
a trap for whoever ends up running unprivileged.
Verified live on the gated filesystem: a webshell written to disk is
reported and left in place; an ELF miner written beside it is
quarantined. Event text changed to match — "threat detected in X —
reported, not moved" rather than implying something happened.
One process note. The first attempt at this edit silently did nothing:
the replacement did not match because of indentation, the tooling
reported success, and the webshell was still moved. Second time I made
the edit assert its anchor before applying. That is the third silent
no-op edit in this session and the pattern is now obvious enough to
stop assuming an edit landed.
303 tests pass. Gate off.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Stage 3 passed on throughput — 27,339 events, zero rescues, Caddy
unmoved — and then the soak found what the load test could not.
Two more false positives, both the same shape as the ones before:
* EICAR matched an 8.5 MB rustc incremental-compilation cache, because
the test source being compiled contains the literal. Hound moved it to
quarantine mid-build and rustc panicked. The standard defines the
EICAR file as exactly that 68-byte string, optionally padded to 128,
so the rule now says filesize <= 128.
* The webshell rule matched a 4.3 MB AI session transcript, because the
conversation had been discussing webshells and therefore contained
"<?php", the eval pattern and "$_POST". The transcript was moved to
the vault and its history lost. A webshell is a PHP file: small, and
opening with a PHP tag. Now filesize < 1MB and $php in (0..4096).
The interesting part is why the second one happened at all. After the
first, I added a test asserting that a large file containing rule
strings is not a threat — and hand-listed the strings. I listed the
miner's and the rootkit's and forgot "<?php". The test passed and the
transcript was quarantined anyway.
So the test now extracts every string literal from the rule pack itself
and builds the haystack from those. A rule added tomorrow is covered
without anybody remembering to cover it. It also asserts the extractor
actually found the strings, because a parser that silently returns
nothing would make the whole thing vacuous.
Both fixes have a paired test that the detection still works: a real
68-byte EICAR file is caught, padded to 128 it is caught, and a real
webshell is caught.
Worth recording, because it is not a bug: six houndd tests failed while
the gate was armed. Hound quarantined the EICAR fixtures the test suite
had just written — correct behaviour, colliding with a suite that
creates real malware samples. Running the antivirus's own tests on a
gated machine needs thought; the tests are not wrong and neither is the
gate.
The definitions chain now works end to end: pack built from OSV, signed
with the release key, published to /srv/houndav/defs, installed, and
verified on load against the public half compiled into the agent —
"defs: 19 indicators from 1 pack(s) [2026.08.21]".
The public key is in the source on purpose. The agent is open source and
anybody should be able to check that the definitions they received are
the ones we published.
300 tests pass. Gate is off pending these fixes being soaked.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Armed the gate on / on the live server. Aborted after about twenty
seconds. The box was never at risk — Caddy stayed sub-millisecond and
load never rose — but the gate blocked reads of an AI agent's session
transcript, reporting it as Linux.Coinminer.XMRig.
It was not wrong about the bytes. That transcript contains
"stratum+tcp://", "donate-level" and "xmrig" because the miner rule was
being written in that session. The rule matched a document ABOUT
malware.
Three bugs, none of which the tmpfs stage could have shown:
1. The miner rule had no file-type condition, so any text mentioning
mining tripped it: threat-intelligence reports, security blog posts,
support tickets, an antivirus's own logs. It now requires ELF magic,
as the rootkit rule always did. Two regression tests: a transcript
discussing the rule is clean, and an ELF carrying the same strings
still matches — the fix must not cost the detection it exists for.
2. The gate requested FAN_OPEN_PERM, so it held every OPEN, not every
execve. A matching file could not be read by anything. That is a
different product from the one advertised, and on a multi-tenant box
it is a denial of service against the operator rather than a defence.
Read events are no longer requested at all. FAN_OPEN_EXEC_PERM and
FAN_CLOSE_WRITE cover the threat: execution is refused before it
happens, and anything malicious written to disk is quarantined when
the write completes. An interpreted script is caught as it lands
rather than as it is read — the same protection, one step earlier.
`serve` also guards deny-on-exec explicitly, so re-requesting read
events later cannot silently restore the old behaviour.
3. Hound did not exclude its own state. /var/lib/hound and /run/hound
are now always excluded; the vault holds live malware by definition.
Henry asked whether the single watchdog rescue was queue pressure or
scan time. It was scan time: the gate inherited the on-demand 100 MB
limit and tried to read and match a multi-megabyte transcript inline
while holding a process. A gate's budget is a deadline, not a size, so
it now caps at 32 MB — anything larger is allowed through unread rather
than turned into a rescue, which is a process released unscanned and
worse than never having looked.
Dropping read events made everything faster, because most opens on a
running machine are reads:
latency +1.38 -> +0.79 ms per exec
throughput 2,680 -> 4,178 execs/sec (58% of ungated, was 36%)
events 1,179 in five seconds on an idle tmpfs -> 1
Re-verified on the tmpfs: an ELF miner is quarantined before it can even
be made executable, a document naming every one of its strings is
readable, and a clean binary runs.
297 tests pass. The gate stays off; stage 3 gets attempted again with
these fixes and fresh numbers.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Armed the execution gate on the live server for the first time. It
reported itself armed on a dedicated tmpfs, and then let EICAR execute.
Counters: 0 allowed, 0 blocked. Not one event was ever delivered.
Cause: systemd gives the service a PRIVATE MOUNT NAMESPACE. Several
perfectly ordinary hardening options force one — ProtectProc,
ProtectKernelTunables, ProtectControlGroups — and none of them mention
it. FAN_MARK_MOUNT marks a vfsmount, and a private namespace holds its
own vfsmount for the same filesystem. So the daemon marked its copy,
every other process on the machine used the host's copy, and the gate
protected nothing while claiming to be armed.
That is the worst way for a security feature to fail: silently, with a
reassuring status line. Nothing in the unit tests could have caught it —
they run in the host namespace, where the mount mark works.
Fixed by always using FAN_MARK_FILESYSTEM, which marks the SUPERBLOCK.
A superblock is shared across namespaces, so events arrive from
everywhere, and scoping still works because a superblock is exactly one
filesystem: marking a dedicated mount covers that mount and nothing
else. mark_mount is kept for the smoke-test example, which runs outside
systemd, with a doc comment about when it lies to you.
Two more that only appeared once the gate was actually armed:
* SystemCallFilter=@system-service kills the daemon with SIGSYS the
moment the gate is switched on. fanotify_init and fanotify_mark live
in @privileged, which @system-service deliberately excludes. Granted
individually rather than by adding @privileged, which would also admit
setuid, chroot, bpf and kexec_load. Invisible until armed — the
service starts fine with the gate off.
* The capability reduction reported "60 capabilities could not be
dropped" while the end state was perfectly correct. systemd's
CapabilityBoundingSet had already done the work, and the service does
not hold CAP_SETPCAP afterwards, so every redundant drop failed EPERM.
It now checks what is actually present, attempts only that, and judges
by the end state rather than by return codes.
Also removed AmbientCapabilities from the unit. Ambient capabilities are
inherited by children, the daemon shells out to freshclam/rpm/pacman on
some paths, and a root process already receives the bounding set as
permitted — so it bought nothing except a way for CAP_SYS_ADMIN to leak
into a subprocess.
Performance, measured on the live server rather than guessed at:
+2.70 ms/exec as first written
+1.47 ms/exec after the reader blocked on poll() instead of sleeping
a millisecond between empty reads — that sleep sat on
the critical path of every execve
+1.38 ms/exec after answering cache hits in the reader thread, with
no channel handoff or worker wakeup
2,680 execs/sec sustained through the gate, 16-way parallel, with
ZERO watchdog rescues — the queue never fell behind. Ungated is 7,455.
Caddy stayed at sub-millisecond throughout and load did not rise.
Joe and Henry are right that the exec-heavy paths on this box — Docker
overlays, agent workspaces, PM2 — are the performance bar rather than an
exclusion list. Protecting agent workspaces from injected payloads is
the product. 2,680/sec with no backlog is roughly ten times what this
machine generates, so the bar looks clearable; stage 2 will say for sure.
295 tests pass, and the three-phase gate smoke test still passes
including the fail-open case.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Installed the .deb on the live server. The daemon did not start, and
everything below is what that one command surfaced.
1. MemoryDenyWriteExecute=yes stopped the service dead.
yara-x compiles rules to WebAssembly and JITs them, so it needs pages
that go writable then executable. With W^X enforced the daemon aborts
at startup: "unable to make memory executable". The unit had passed
systemd-analyze verify, which checks syntax and cannot know this.
Now off, with the reasoning in the unit rather than in a commit
nobody will read: a hardening directive that stops the service is
worse than the exposure it prevents, because the machine ends up with
no antivirus at all. What compensates is spelled out beside it.
2. Hound detected itself.
The goodware gate reported /usr/bin/houndd as Linux.Coinminer.XMRig
and Linux.Rootkit.Preload. Correctly: the built-in pack matches on
"stratum+tcp://", "xmrig", "RTLD_NEXT" and "ld.so.preload", and the
pack was embedded verbatim, so the daemon's own binary contained all
of them.
Not cosmetic. With the execution gate armed, Hound would have refused
to execute itself or quarantined its own binary — a scanner that eats
its own daemon the moment protection is switched on.
The pack is now XOR-masked at build time (build.rs) and unmasked at
startup. Not a secret — the rules are open source — the only job is
keeping the literal bytes out of the executable. Two regression tests:
the embedded blob carries no plaintext rule strings, and a built
daemon binary in target/ carries none either.
3. The postinst copied the built-in pack into /var/lib/hound/rules,
where the daemon compiled it a second time and logged a duplicate
declaration on every start. That directory is for ADDITIONAL packs;
the built-ins live in the binary. Removed from deb, rpm and AUR.
4. The daemon and the CLI disagreed about the socket. systemd gives the
service /run/hound; the CLI looked in $XDG_RUNTIME_DIR and reported
the daemon unreachable — technically true, entirely unhelpful.
default_socket_path() now prefers /run/hound when it exists, the unit
states HOUNDD_SOCK explicitly, and a permission error on the socket
says "try: sudo hound" instead of "Permission denied".
Also: Recommends: clamav-daemon was wrong and apt duly installed clamd,
which took 970 MB of RSS on the live server. clamd is an optional
arm's-length engine, so it is a Suggests. I stopped and disabled the
copy my install pulled in.
Verified on the server after fixing: service active, status reports the
engine and the gate, EICAR caught, rootkit scan clean, and the goodware
gate passes across 3,955 system binaries including the now-installed
houndd.
The gate remains OFF. Turning it on for the host that serves Caddy is a
separate decision.
295 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Lockfile parsing for npm (all three lockfile versions), yarn, cargo,
poetry, requirements.txt, go.sum, Gemfile.lock and composer.lock, wired
through the indicator index so a sweep checks real dependencies against
real definitions. Signed packs load in the daemon; the sweep gets the
index; `hound supply-chain` cites the OSV record it matched.
A lockfile is the right thing to read: it names every transitive
dependency at an exact version in one small file, and it lists what WILL
be installed rather than what already is — which matters when the
payload runs during installation.
Every parser is hand-written rather than pulling in a TOML and a YAML
crate. Two fields from each format, and a scanner parsing hostile input
should have as little parsing surface as it can.
The important part of this commit is a false positive it fixes.
Building a pack from the whole crates.io OSV export and sweeping a
project produced TWO criticals: rustdecimal, correctly, and **tokio
1.38.0**, which is not malware and never has been. The export is 1,524
GHSA and 1,206 RUSTSEC vulnerability advisories against 19 malicious-
package records, and the parser treated all of them as malware.
GHSA-2grh-hm3w-w7hv describes a tokio race condition fixed in 1.8.1;
Hound reported a version released years later as malicious.
Two independent bugs, either of which alone is fatal:
* Vulnerability advisories were ingested at all. A malicious package
should not exist; a vulnerable one is a legitimate library with a bug
and most of its versions are fine. Records must now PROVE they are
malicious-package reports — a MAL- id, the malicious-packages-origins
marker, or GHSA's "Malicious code in" wording — and anything
unrecognised is dropped.
* Unrecognised version ranges fell back to "all versions", which is the
opposite of safe. That is what turned a range of 1.8.0-to-1.8.1 into
a verdict on every tokio ever published.
Rebuilt against the same input, the pack now holds 19 indicators rather
than 3,614, rustdecimal is still caught and cites MAL-2022-1 rather than
a GHSA advisory, and tokio and serde are clean. The real tokio advisory
is now a regression fixture, because anything that flags tokio is a
product nobody trusts twice.
Also: definitions loading fails CLOSED on authenticity and OPEN on
everything else. No trusted key means no definitions and a message
saying so, because an operator who believes they are protected and is
not is worse off than one who knows. A pack that fails verification is
skipped and the rest still load. No packs at all is a working daemon —
install scripts, prompt injection, pickles and MCP audits need no feed.
There is deliberately no placeholder signing key compiled in. A fake key
that looks real is how a development shortcut becomes a shipped
vulnerability; an empty trust store is noisy in the way that gets fixed
before release. HOUNDD_DEFS_KEY supplies one for development.
294 tests pass across the workspace.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The previous commit went out with a failing test. It passed on a rerun,
which is worse than failing — a flaky test in a security product either
gets ignored or gets deleted, and both are how a real regression ships.
The offender was rootkit's own thread test. It read /proc/self/task and
took the /proc snapshot at different moments, so a thread started by
another test between the two reads looked like a thread that answered
kill() but was missing from the listing. That is precisely the
start/exit race the hidden-process check exists to avoid, reintroduced
in the test written to prove the check avoids it.
Each thread now reports its own tid via gettid and then parks, so all
eight are demonstrably alive across the whole measurement window. Five
consecutive full runs, 124/124 each.
Also isolated persistence's read-only-scan test behind the env lock:
baseline_path() reads XDG_DATA_HOME and the quarantine tests reassign
it, so two scans either side of that disagreed about first_run. It now
takes the lock, points at its own directory, and additionally asserts
the thing the test was named for — that a read-only scan writes no
baseline file, and that --accept does.
256 tests pass across the workspace.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Phase 3's foundation. Three jobs that share a data model:
osv parse the ossf/malicious-packages feed (Apache-2.0, ~226k
records, daily) into indicators
index answer "is this package known bad?" fast enough to ask it
thousands of times per project sweep
pack sign a definitions pack, and verify one before loading it
Validated against the real feed rather than fixtures: the whole
crates.io OSV export, 2,749 records, parsed with zero failures — 3,885
indicators across ten ecosystems, 19 of them MAL-. `rustdecimal` (the
real typosquat of rust_decimal) resolves in crates.io and stays clean
in npm and PyPI, which is the ecosystem isolation working.
Notes on the three:
* A malicious-package record is not a vulnerability record. It almost
always carries introduced:"0" with no fix, meaning EVERY version is
malicious — the package exists only to be malware, so there is no safe
version to upgrade to. Conflating that with a version-bounded
vulnerability either misses real hits or condemns safe versions of
legitimate packages, so the two are modelled separately.
* The index is a cuckoo filter in front of a map. Cuckoo rather than
bloom specifically because a definitions feed needs DELETION: OSV
withdraws records — it once withdrew 157 malware reports after a
false-positive incident — and a filter you cannot remove from means a
withdrawn record costs a probe forever or forces a rebuild. 226k
indicators fit in under 4 MB; the crates.io set is 8 KB.
The property that must never break is no false negatives, and it has
its own test. A false positive costs a hash lookup; a false negative
is malware reported as clean. That is also why a fingerprint hashing
to zero is nudged to one — zero marks an empty slot, so without the
nudge one key in 65,536 would silently vanish.
* Signing is not about entitlement; the subscription gates the server.
It answers "is this pack really from us?", because someone who can
substitute a definitions file can add an entry for /usr/bin/sudo and
have Hound quarantine it on every machine that updates — a supply
chain attack delivered through the security product. Verification
happens on the raw bytes BEFORE anything parses them, so a hostile
pack never reaches the parser at all.
256 tests pass across the workspace.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Eight commits taking Hound from a compiling skeleton to something that
detects and blocks. Everything below was measured on real machines rather
than reasoned about.
Phase 0 yara-x in process, replacing a clamscan fork.
6.5s -> 4ms per file. 400 files cold in 9ms.
Phase 1 fanotify execution gate with a fail-open watchdog.
EICAR refused at execve in 2ms; a stalled scanner still
releases the process. Daemon holds 4 capabilities of 41.
Phase 2 .deb and AppImage built and tested; rpm spec and PKGBUILD
written. Hardened systemd unit. App icon, optically sized.
Phase 4 hound-supply: pickle RCE, prompt injection, install scripts,
typosquat/slopsquat, MCP audit. `hound supply-chain <path>`.
Phase 5 Rootkit checks rewritten from FP generators into questions
with factual answers, plus the persistence ledger.
214 tests, up from 29. The goodware gate — every rule scanned against
every binary in /usr/bin, /bin and /usr/sbin — has already earned its
keep twice, and cross-distro verification passed on Ubuntu 26.04.
exec_gate ships OFF in every packaging format. It needs CAP_SYS_ADMIN
and covers the whole root filesystem; turning it on is the operator's
decision, and flipping the default waits for Phase 2 soak testing.
Completes Phase 5. Half of a Linux compromise is not a file on disk, it
is a line added to a startup file — a curl in a shell profile, a systemd
unit with a dull name, one extra key in authorized_keys. The payload is
often unremarkable; what makes it an incident is that it survives a
reboot and nobody reads those files from one year to the next.
So this is not a scanner but an inventory with a memory. It records
systemd units (system and per-user), cron in all its locations,
autostart entries, shell profiles, authorized_keys and ld.so.preload,
then reports what CHANGED.
Three decisions, all of which are the difference between a report people
read and one they turn off:
* Content is hashed, not stat'd. An mtime can be set backwards with one
touch, and someone editing a startup file is exactly the person who
would. Verified: a backdated edit is still caught.
* A first run reports no changes and says so. Everything would be a
change, and a first-run report full of alarms is one nobody reads.
What a first run can honestly say is how many entries no package
claims — 97 of 1,026 on this machine — because that is true
regardless of history.
* Writing the baseline is an explicit act (`--accept`, or
update_baseline on the wire). A plain check must never quietly record
whatever is currently installed as normal; that is how a compromise
becomes the new baseline.
Package ownership decides what is ordinary: a unit that arrived with a
package is the system working, the same unit unowned is somebody's
decision. Reuses the merged-/usr-aware index from the rootkit rewrite,
with a test asserting most units resolve to a package — if that ratio
collapses, ownership lookup has broken and the whole report is noise.
Exercised end to end against this machine: baseline of 1,026 items,
a planted user unit caught as ADDED, an in-place edit with a backdated
mtime caught as CHANGED, and its deletion caught as REMOVED. The test
artifact was removed afterwards.
Cross-distro verification of the rootkit rewrite is now MET. Henry ran
the suite on the Ubuntu box (26.04, glibc 2.43): 20/20, including both
unowned_setuid_does_not_fire_on_a_healthy_system and
no_false_positives_on_system_binaries.
214 tests pass across the workspace.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Phase 5's correctness half. The previous implementation could not ship:
its two main checks were structurally wrong rather than badly tuned.
hidden processes was "any /proc/<pid> whose comm we cannot read",
which fires on every process that exits between the
listing and the read. A race, not a signal.
setuid anomalies compared against a hardcoded allowlist of binary
names, written on one distribution.
Replaced with questions that have factual answers:
A process is hidden when the kernel agrees it exists and /proc does
not list it. kill(pid, 0) answers the first half for the whole PID
space — ESRCH means gone, EPERM means it exists and belongs to
somebody else, which is the case that matters since a rootkit's
process will not be ours. The sweep is bracketed by two listings and
candidates are re-verified, so a process that merely started or
exited during the scan cannot be mistaken for a hidden one.
A setuid binary is suspicious when no installed package claims it.
The package manager already knows what belongs on the system.
Two bugs found by testing against this machine rather than reasoning
about it, both of which would have made the feature useless in the
field:
* /proc lists thread-group leaders; kill() accepts any THREAD id. A
process with twenty threads therefore has nineteen ids that answer
kill and appear in no /proc listing. Comparing against the pid set
alone reported dozens of criticals on a completely healthy laptop.
The honest set is the union of leaders and their /proc/<tgid>/task
entries.
* Merged-/usr breaks package ownership in BOTH directions. /bin is a
symlink to usr/bin, so every binary has two names, and dpkg's own
index is inconsistent about which it records: sudo.list says
/usr/bin/sudo while fuse3.list says /bin/fusermount3 and cifs-utils
says /sbin/mount.cifs. String comparison reported the entire setuid
set as unowned. Both spellings now go into the index, candidates are
deduplicated by resolved path, and lookups try both.
Also: ld.so.preload is now checked (it is empty on a healthy system and
is the classic userland rootkit), the writable-directory check no
longer counts sticky-bit directories, and the hidden-file check uses
symlink_metadata so an ordinary dangling symlink is not an incident.
Verified on this machine, privileged and not: clean, 0 findings, 2.8s
including the full 4.2-million-pid sweep. The exit criterion asks for
five machines across three distributions and only one was available
here, so treat the cross-distro half as unmet.
The regression tests are the point: a scan run while processes churn
continuously must produce no criticals, and a process with eight live
threads must not produce eight findings.
200 tests pass across the workspace.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Phase 4's detection core, as a standalone crate. This is the part with no
competitor on Linux, and deliberately the part with no Linux in it —
file parsing and logic only, no fanotify, no /proc, no eBPF — so the
macOS and Windows port is weeks rather than a second product.
Five detectors, 88 tests:
pickle GLOBAL/STACK_GLOBAL walk over .pt/.ckpt/.pkl/.joblib.
torch.load runs a stack machine; a model file is a
program and downloading weights is a code-execution
decision.
injection Instructions aimed at a coding agent in CLAUDE.md,
AGENTS.md, .cursorrules, copilot-instructions.
installscript preinstall/postinstall hooks that curl|sh, decode and
run, reach for credentials, or install persistence.
typosquat Damerau-Levenshtein against popular names, plus
slopsquat detection: new + near-zero downloads + one
edit from something popular is the signature of a name
a model invented and somebody then registered.
mcp Servers fetched unpinned at launch, handed secrets, or
pointed at $HOME or credential paths.
Wired through `supply.sweep` on the socket and `hound supply-chain
<path>`, which exits 1 on a critical so it drops into CI.
Three things worth recording:
* Scoring is by independent category, not by keyword count. One
suspicious phrase is a phrase; two categories at once is an attack.
A file that only says "ignore previous instructions about formatting"
is a warning, not a critical.
* Proximity matters more than presence. The first version flagged an
entirely ordinary conventions file, because it mentioned ".env" in
one paragraph and "prefer small commits" in another. A credential and
a movement verb now have to appear within a sentence of each other.
The test that caught it is kept as the regression.
* Pickle call detection has to come out of the opcode walk, not a byte
search. REDUCE, INST and OBJ are the ASCII letters R, i and o, which
also occur inside every string the stream carries — searching raw
bytes finds the o in "os" and reports a call that never happens,
turning every warning into a critical.
Every finding carries a plain-language explanation and a next step, and
there is a test asserting explanations do not leak rule identifiers or
jargon. The audience includes people who cannot triage a YARA match and
should never be shown one.
Verified against a demo project holding a squatted @vue plugin with a
curl|sh postinstall, a poisoned CLAUDE.md in a vendored repo, an
unpinned MCP server holding a GitHub token, and a pickle calling
os.system — four criticals and one warning, while the legitimate
CLAUDE.md, the real express manifest and the properly-scoped MCP server
beside them stayed clean.
189 tests pass across the workspace.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Phase 2. "Distro-agnostic" was a claim with no packaging behind it.
.deb and AppImage are both built and tested here; the rpm spec and
PKGBUILD are written but not yet built, since neither rpmbuild nor
makepkg exists on this machine.
The AppImage is deliberately not a packaged daemon. An AppImage is
unprivileged by design — no install, no root, no systemd — so the
execution gate cannot exist in one, and pretending otherwise would be
worse than saying so. What it is instead is the try-before-you-install
build: on-demand scanning, quarantine under the user's own data dir,
rootkit and supply-chain checks, all with nothing left behind. Asking
it for the gate prints why it cannot and how to get it. Verified with
an isolated HOME: status, scan, EICAR caught, gate refusal.
Two packaging bugs caught by looking rather than assuming:
* mktemp -d creates 0700 and dpkg applies the staging root's mode to
"/". Installing that package would have chmodded the root directory
to 0700 and broken the machine.
* AppRun wrote its log before creating the directory, and built a
socket path that can exceed sun_path (108 bytes) when
XDG_RUNTIME_DIR is long. Both fixed; the socket falls back rather
than failing with an error nobody can act on.
The systemd unit is hardened as far as this daemon can be. Notably
ProtectSystem=strict is WRONG here and is left off on purpose: it
remounts everything read-only, and quarantine has to remove a threat
from wherever it landed. ProtectHome is off for the same reason. The
CapabilityBoundingSet mirrors what caps.rs drops to, so the machine is
protected even if the binary is replaced by one that does not reduce
itself.
App icon: the mark in white on a periwinkle tile, per Joe. Small sizes
are not the same artwork scaled down — the mark is line-weight, so at
16px a 62% inset leaves about a pixel and a half of stroke and the head
turns to mush. Each size is authored with its own inset and corner
radius, and the ground goes flat below 32px because a gradient across
16 pixels is just noise. The tray ladder is untouched: those glyphs stay
transparent and re-tint per state so they can sit on any panel colour.
Package installs are NOT enabled by default beyond the daemon itself —
exec_gate stays off until the operator turns it on, in every packaging
format.
99 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Completes Phase 1. The gate now asks for FAN_CLOSE_WRITE alongside the
permission events, so a threat written to disk is quarantined and a
threat being executed is refused — one mechanism, one mark, no
watch-descriptor ceiling and no blind spots outside a configured list.
Verified live. The nicest evidence is an error message:
$ chmod +x /tmp/hound-live/malware.sh
chmod: cannot access '/tmp/hound-live/malware.sh': No such file or directory
Hound had already quarantined it. `hound quarantine list` shows the
entry, the clean binary beside it still runs, and CapPrm/CapEff/CapBnd
read 000000000020000e.
Three bugs, each of which looked like working code:
* A file descriptor number is not an identity. The kernel allocates an
fd per event and recycles the number the moment we close it, so one
write arrives as FAN_OPEN_PERM on fd 6 and then FAN_CLOSE_WRITE on fd
6 again. Idempotency keyed on the fd treated the second as a duplicate
of the first and dropped it — detection ran, matched EICAR, and threw
the result away. Events now carry a monotonic seq that is never reused.
* rename(2) fails EXDEV across filesystems, and for quarantine that is
the common case rather than the exotic one: the vault is under
/var/lib while threats land on /home, in a tmpfs, on a USB stick or
in a container overlay. Quarantine now falls back to copy-then-unlink,
unlinking only once the copy is safely down, and seals the stored file
at 0600 with every execute bit cleared.
* The capability set was too small to do the job. CAP_DAC_READ_SEARCH
lets us read a threat but not unlink it, so quarantine failed EACCES
as root. The set is now four capabilities — SYS_ADMIN, DAC_READ_SEARCH,
DAC_OVERRIDE, FOWNER. DAC_OVERRIDE is close to "write anywhere" and
that is worth being honest about; an antivirus that quarantines cannot
avoid it, because the threat is by definition in a directory somebody
else owns. What the reduction still buys is what it excludes, and
there is a test asserting SYS_MODULE, SYS_BOOT, SYS_PTRACE, NET_ADMIN,
NET_RAW, AUDIT_CONTROL and SETUID never creep back in. Narrowing
further means a separate privileged helper for quarantine.
realtime.rs is now documented as the unprivileged fallback and does not
start when the gate is armed — running both would scan everything twice
and quarantine the same file from two threads.
99 tests pass. HOUNDD_GATE_DEBUG=1 dumps every event and decision.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Wires the execution gate into DaemonState::boot, reports it on the wire
and in `hound status`, and reduces the daemon to the two capabilities it
actually needs.
Verified live, gate marked on a scratch tmpfs rather than /:
clean binary ran
EICAR binary execve -> EPERM, "Operation not permitted"
hound status Exec gate: armed on /tmp/hound-live, 2 allowed, 1 blocked
CapPrm/CapEff/CapBnd 0000000000200004 (CAP_SYS_ADMIN | CAP_DAC_READ_SEARCH)
Three ordering bugs found by checking rather than assuming, all of which
returned success while doing nothing:
* Capabilities are per-thread. Dropping them after spawning the reader
and workers reduced only the main thread and left four workers holding
full root — the exact opposite of the intent. The drop now happens
after fanotify_init and the marks, but before any thread exists, so
workers inherit the reduced set.
* PR_CAPBSET_DROP needs CAP_SETPCAP in the effective set, and capset had
already thrown it away. Every bounding-set drop failed EPERM, silently,
leaving a full CapBnd behind a log line claiming otherwise. Bounding
set is now drained first, while the authority to do it still exists.
* Because both of the above looked like successes, drop_to_gate_minimum
now reads CapEff and CapBnd back from /proc/self/status and errors if
they are not what was asked for. A privilege reduction that cannot be
observed has not happened.
Also: `hound status` grew an Exec gate line. timed_out above zero is the
number worth alarming on — it means the watchdog is releasing processes
unscanned and the gate has quietly degraded to advisory.
Gate arming and every failure path now log to stderr, so the journal
records a security-relevant state change instead of only the in-memory
event ring.
86 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
FAN_OPEN_EXEC_PERM hands us the open and waits for an answer, so a
binary can be refused before it runs. inotify could only report what
had already happened.
Verified end to end as root against a dedicated tmpfs (examples/
gate-smoke.rs, three phases):
benign binary ran 7.2 ms
EICAR binary blocked 1.8 ms never executed
scanner stalled 5 s ran 1.6 s watchdog rescued 3 events
The third phase is the one that matters. A gate that can hold a process
forever is a machine-wedging bug wearing a feature's clothes, so the
watchdog answers ALLOW for anything unanswered past DEADLINE and counts
it. A missed detection is a bad day; a frozen machine ends the product.
Two things this cost, both worth recording:
* Scanning by re-opening the path deadlocks the daemon against itself.
The open() lands on the watched mount and queues a permission event
behind the one we are currently answering, and we cannot answer that
one until we finish this one. Allowing our own pid does not help —
the thread never gets back to the queue to apply the rule. The gate
reads through the descriptor the kernel already handed it, with
pread so the gated process still sees its own file offset. This is
what hung the first smoke run.
* The watchdog can only rescue events it has been told about, and it
learns of them when the queue is drained. Scanning on the draining
thread makes every event behind a slow scan invisible to the
deadline. Reader and workers are therefore separate threads: the
reader never blocks on a scan, so every event is registered within
microseconds of arriving.
Also:
- ScanEngine::scan_bytes — the seam the gate needs, since it must never
scan by path. Engines that cannot do it return None and simply are
not usable behind the gate.
- Settings gain exec_gate and exec_gate_paths, defaulting to OFF. It
needs CAP_SYS_ADMIN and a root-filesystem mark holds every process on
the box; that is not a default to ship before Phase 2 soak testing.
- ABI constants are defined locally rather than taken from libc, so a
version bump cannot quietly change what we ask the kernel for.
78 tests pass, up from 57.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The old engine shelled out to clamscan for every scan, and clamscan
reloads a 169 MB signature database on every invocation. Measured on a
68-byte EICAR file: 6.5 seconds and ~1.5 GB RSS — paid once per file,
and realtime.rs called it once per inotify event.
Replaces it with HoundEngine: yara-x compiled once at daemon start,
held in memory, one scanner reused across a whole walk, plus a verdict
cache keyed on (dev, ino, mtime, size) so an unchanged file that has
been seen before never reaches the matcher.
Measured after, same machine, same EICAR file:
single file 6.5 s -> 4 ms
400 files cold -- -> 9 ms
400 files warm -- -> 5 ms
Also here:
- rules.rs: hot-swappable rule store. Built-in pack is embedded so a
fresh install detects something before it has ever reached the
network; on-disk packs load from $HOUNDD_RULES_DIR, /var/lib/hound
or the XDG data dir. Reload swaps an Arc, so in-flight scans are
never torn out from under.
- cache.rs: bounded FIFO verdict cache. Any of the four key fields
changing means rescan, so edits, truncates and replace-by-rename all
correctly miss.
- The goodware gate: every rule is scanned against all of /usr/bin,
/bin and /usr/sbin in CI, and a single hit fails the build. It has
already earned its keep — it caught a reverse-shell rule that matched
/usr/bin/sudo, which is now removed rather than tuned. A rule that
quarantines sudo is worse than no rule at all.
- ScanEngine is Send + Sync and selection stays per-call, so
HOUNDD_ENGINE=clamav still reaches the legacy path for comparison.
- ScanResult.skipped reports files passed over for size instead of
quietly counting them as clean.
- Settings gain theme (auto/light/dark), tray_icon_style (color/mono),
close_to_tray and confirm_quit, normalised daemon-side because
clients are not trusted to send a theme we can render.
57 tests pass, up from 29.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ClamAV is a temporary dependency. Everything ClamAV-specific moves
behind a 4-method trait in crates/houndd/src/engine.rs:
trait ScanEngine { name; probe; scan; update }
struct ClamAvEngine // today
const ENGINE // one-line flip when the native engine lands
- parse_clamscan() is now a pure fn with unit tests (OK/FOUND/INCOMPLETE,
dedup, malformed lines)
- main.rs drops all clamscan/clamd/freshclam knowledge; Status reports
engine='clamav' via the trait
- README documents the seam and the wire contract that stays stable
GUI (gui/, Tauri 2, standalone workspace, vanilla JS premium-dark shell):
- system-tray sentinel on the 4-state dog-head ladder (green/amber/red/gray)
with tooltips; clicks open the window / scan / update
- window: protection hero, path + recursive scan, progress bar, results
table, signature update log; state mirrored to the tray
- compiles clean (cargo build, 0 warnings) on Mint 22.3 + webkit2gtk-4.1
All 11 workspace tests pass, incl. the E2E EICAR scan over the real
Unix-socket daemon (clamscan finds the planted EICAR).
- hound-api: DbFile (file + updated_at), Status.db, UpdateResult, Client::update()
- houndd: probe() extracts signature freshness from /var/lib/clamav mtimes;
update() tries 'sudo freshclam', falls back to plain 'freshclam' and
reports the honest reason on failure; output capped to 2KB tail
- hound CLI: 'hound update' with --json parity and human output
- rename Status.clamav_present -> engine_present + add Status.engine
(prep for the engine seam; wire stays engine-agnostic)
Live-verified: status shows DB file + last-modified RFC3339; update
returns ok:false with the real log for a plain user (exit 1).
dev
