I build command-line tooling for other engineers: security scanners, DevOps utilities, and infrastructure diagnostics. Eight are public and listed below.
They are the same shape on purpose. Small enough to read in one sitting, zero runtime dependencies, tested in CI against real systems rather than mocks, and explicit about what they do not do. A tool that overstates its coverage is worse than no tool, because you stop looking.
I also make things for people who never open a terminal: a video downloader with a native Mac app and a Firefox add-on, and a browser extension that turns long YouTube videos into courses. Those two are listed after the tools.
Alongside those I build private production systems: an async trading engine, a sub-second video platform, and a multi-tenant SaaS. Summarised further down.
Reach me at mo7sen.cto@gmail.com.
01 · sentinel-auditLinux server hardening auditor in pure Bash. Thirty-four read-only checks across SSH, filesystem, accounts, network exposure, Docker, and leaked credentials. Reads the effective SSH config through
|
02 · deploy-forgeZero-downtime deploys for servers you reach over SSH. The release swap is an atomic
|
03 · wp-sentinelDetection-only WordPress scanner. Finds the misconfigurations that actually get
sites compromised: readable It observes and never exploits, which is what makes it safe to point at production. Scans a real WordPress container in CI.
|
04 · tls-sentryCertificate monitoring that tells you what is wrong, not that something is. An expired certificate fails verification, so most checkers can only report
"connection failed". This one handshakes again without verification to read the
certificate and comes back with
|
05 · cronscopeSee inside a crontab before it surprises you. Predicts real run times and
catches the mistakes that cause silent incidents: Renders the week as a heatmap. Cross-checked against
|
06 · groundcheckA deterministic groundedness linter for retrieval answers. Flags fabricated numbers, unsupported claims, and broken citations. A model asked to judge another model scores differently on each run, so it cannot gate a pipeline. This returns byte-identical output for identical input, proven across seeds, which is what makes it usable in CI. Try it in your browser.
|
07 · logsiftFinds the one request that failed inside a log stream nobody can read by eye. Lines that are structurally the same are collapsed into a single pattern with a count, and the rarest patterns are shown first, because the line that appears once in ten thousand is usually the one you are looking for. Reads stdin, so it composes with
|
|
08 · plumblineChecks that a README is true. Every command, path, make target, package script, module and environment variable a document asserts is pulled out and checked, and the run exits non-zero when one fails. A local model reads the prose to find claims but never decides whether one holds; the checks do. It started with a real bug: a deploy guide of mine whose every command had pointed at the wrong path since the scripts moved.
|
|
The same rules apply to all of them, and they are the reason the list is short.
Nothing is added to your machine that is not already there: no runtime dependencies, so a tool runs on a box you have just been handed. Scanners observe and never exploit, which is what makes them safe to point at something you cannot afford to break. Output is deterministic, so it can gate a pipeline instead of merely informing one. Everything is tested in CI against a real system rather than a mock of one.
And every README has a section on what the tool gets wrong, where the heuristic breaks down, and what it cannot see. That section is the one most projects leave out, and it is the one that decides whether you can trust the rest.
09 · vidkeepVideo downloads for offline viewing, built on yt-dlp. A file is fetched over several bounded range requests at once, picked to fit the disk you have, and resumed range by range after an interruption. In its own benchmark four connections ran four to six times faster than one. Ships as a command, a native SwiftUI app for macOS, and a Firefox add-on that hands downloads to it over native messaging, tested together in a real Firefox in CI.
|
10 · coursemarkStudy long YouTube videos like a course. Chapters become a checklist that ticks itself only after real playback, so seeking past a chapter or an ad playing does not count, with notes pinned to the moment and a Markdown export. Manifest V3 for Chrome and Firefox, English and Arabic, no server. Tested end to end in Chromium against a local copy of the watch page, never against YouTube.
|
|
Gold Engine Async SMC trading engine for XAUUSD. Event-driven
|
Zeem Dashcam video to the browser in under a second.
|
Watheeq Pro Enterprise multi-tenant SaaS under its own organisation, split across service repositories with a staged rollout.
|
masar Locating wide-area network degradation from traffic a fleet already sends, with no access to the routers in between.
|
For sale or licence: a WAN optimizer for DC ⇄ DR replication. In lab tests the next full VM backup crossed a 100 Mbit link in 8 seconds instead of 89. Product overview.
How the Gold Engine is put together
The split that makes it work is where the expensive analysis happens. Structural work runs only when a candle closes, which leaves constant work per tick for sweep detection and the mitigation state machine, so tick latency does not drift as history grows.
Numba compiles the hot paths, with a pure-Python fallback that produces the same
results when compilation is unavailable. Feeds sit behind a TickFeed interface,
so the same engine runs against a live feed or a recorded one. The risk layer
sits in front of the router, and the router defaults to paper, so no code path
reaches a funded account by accident.


