projects ·
transmissions ·
services
getmocha.world — remote job matching that ranks postings before the crowd, then writes your application documents from your own answers: screening questionnaires, CV reads with an honest percentage, cover letters + application emails. Built fail-soft by contract — every AI path has a deterministic fallback, so the product never breaks when the model dies. Kept honest by countables: a real-server jest suite (105 tests across 18 suites), a 62-check HTTP acceptance battery against the live engine, and a browser probe that walks the auth + apply flow on every verify.
I document the work, including the negatives:
- security-toolkit — MIT, dependency-free tools for nonce-laundering XSS: a bundle scanner (
nonce_scan.py) and a rate-limited recon fetcher. - Nonce-laundering trace — traced a server value into a nonce'd inline script in an authorized commerce program, then proved the server sanitised it. write-up
- Basecamp / HEY recon — the public marketing surface ships no CSP nonce at all; the testable class lives in the authenticated app. program landscape
- Available for scoped security review — fixed-scope JS/CSP/auth reviews with evidence-backed reports. services
| repo | report | outcome |
|---|---|---|
| ppggff/vagrant-qemu | provider reported stopped while the QEMU process was alive, once the tmp-dir pidfile vanished |
fixed in 0.8.0 — runtime state moved into a project-local run dir; migrated and verified on both of my VMs |
| waydroid/waydroid | LXC mount (rw, move) denied by AppArmor blocks container start on kernel 7.x |
open — awaiting maintainer input |
| project | what it does | stack | |
|---|---|---|---|
| footorrent | A BitTorrent client in one ~3 MB Rust binary: full CLI, local web UI, prebuilt for macOS, Linux and Windows. Write-up: cyberd4ch.github.io/blog/footorrent-3-5-0-magnets-that-work/ | Rust |
★ 2 |
| txn-agent | Safety-first transaction layer for purchasing agents: deterministic policy gate, human approvals, idempotent checkout | Python |
★ 2 |
| security-toolkit | Dependency-free tools for finding nonce-laundering XSS: a bundle scanner and a safe, rate-limited recon fetcher. | Python |
|
| stride-tmr | STRIDE threat modelling with pytm — worked examples that turn attack-surface lists into ranked, testable threats. | Python |
|
| cportscan | A minimal TCP port scanner written in C — raw sockets, connect timeouts, clean CLI. Learning project. | HTML |
auto-synced from GitHub on 2026-11-17 — edit the repo description, not this table.
Footorrent 3.5.0 — rebuilding the magnet path end to end. Metadata is now fetched and verified before anything touches disk, so the window shows the real name, size and file list, then asks where to save. Along the way: a BEP 10 reply read from the wrong extension id, a handshake that rejected peers for routine chatter, and a download that re-announced to the magnet's own dead tracker list instead of using the 200-plus peers resolve had already found. Dead announce lists fall back to public trackers, peers are sampled in parallel batches, and the UI shows per-peer rates, a tracker-fallback note and a retry button.
Measured on a magnet whose own trackers were all dead: resolve in 17–42 s, 129,302,391 bytes and 11 files reported before writing, then 917 KB downloaded in the first minute across 6 peers.
→ Source · Release notes · Write-up
Practice work that runs on a real machine, not in a slide deck. Every lab carries its own README — goal, run steps, teardown, and an exercise log of what was actually executed, including the parts that failed first.
| lab | what it taught me | state |
|---|---|---|
| three-tier on Compose | nginx → Flask → Postgres, one published port, secrets in a gitignored .env |
running |
| the same app on kind | Deployments, Services, a replicated API, Postgres in-cluster; a KeyError crashloop and the secret that is a bag of keys |
green |
| observability layer | metrics-server, kube-state-metrics, Prometheus, Grafana, postgres-exporter; an alert fired on purpose with a crash pod, then resolved itself after deletion | 0 firing |
| Terraform, Docker provider | plan before apply, state as the source of truth, drift and import made visible, a destroy that leaves nothing |
exercises 1–5 |
| remote state | state migrated to S3 with SSE, versioning and use_lockfile locking; a planted lock object made a second operation refuse to start |
done |
| provider swap | a bucket and its versioning declared with hashicorp/aws — same verbs, different graph; drift introduced by hand, repaired by apply |
done |
Access for all of it is scoped: a dedicated IAM user limited to arn:aws:s3:::ftc-labs-*,
with no root access keys anywhere, and the policy proven against a denied request
instead of assumed.
The full feed lives at cyberd4ch.github.io/blog.
- A modern approach to securing kubernetes cluster — This is an update of Rob Thorne's 2021 Medium article, Adventures in Encryption: Securing Your Laptop Kubernetes Cluster, with credit to Rob for the original approach. Part I covers a local cluster, and Part II takes the same ideas to a public cluster with free Let's Encrypt certificates.
- the tablet lab: root first, then everything else — root first, then everything else: a double-read BROM backup script, the magisk cmdline flag k3s needs, a device-owner agent with documented denials, and what 2 GB of RAM decides.
- proving a PR without patching the file it fixes — hashicorp/vagrant#13842 under macOS 27: a faithful A/B with no sudo, a launcher that eats RUBYOPT, load-order nondeterminism, and one honest caveat for the author.
- triage without a reproduce step: what's left for an outside contributor — six bugs, zero reproduce steps: the DeepTutor issue queue reporters root-cause their own bugs, so outside triage becomes verification, code-location, and shape agreement.
- commented on #278 in BaaRai/ARTEX ·
2026-11-17 - opened pull request #279 in BaaRai/ARTEX ·
2026-11-17 - commented on #2663 in HKUDS/DeepTutor ·
2026-11-17 - opened pull request #2757 in HKUDS/DeepTutor ·
2026-11-17
If dns-watch, txn-agent, or anything else here saves you time, you can fund the work at
github.com/sponsors/cyberd4ch. Sponsorship pays
for the lab VMs, API quotas, and the time behind the write-ups.
Open to collaborating at the intersection of security research, cloud infrastructure, and full-stack development — especially projects that involve breaking (and fixing) stuff.
sig ● nominal · uplink: cyberd4ch.github.io



