🚀 Portfolio · ✉ Email · 🐦 X · 👥 LinkedIn
📍 Illinois · Agent systems · Devtools · Verification · Linux · Android
| Signal | Why It Matters |
|---|---|
| 7 shipped proof points | Portfolio covers the VERITAS Omega Agent Trust Lab, Hermes Agent Mobile Companion, Breakaway, Shell Trials, Omega SMC, coders-bible-desktop, and YubiKey/FIDO2 + Tailscale mesh. |
| 22 direct external PR merges | Live GitHub readback confirms direct merges across GitHub Awesome Copilot, freedesktop-rs nmrs, AgentDoctor, Drift, Trail of Bits Dylint, Kaoto forms, Karakeep, Annotorious, Ansvisor, Amber, ShellCheck, Codex Desktop Linux, three AI-security catalogs, awesome-mcp-servers, and harehare/mq. Hermes Agent maintainer-reapplied authorship remains a separate proof category. |
| 1 formal QA report | Sibyl Memory Plugin beta QA deliverable with sandboxing, local-storage, trust/control, and bug-finding evidence. |
| Before code | Cleanroom certification, field service, industrial maintenance, thermal spray, quality control, and audit-ready documentation. |
Builder-operator: I build local-first agent systems meant to be operated, inspected, approved, tested, and recovered -- not just demoed.
| Start Here | What To Look For |
|---|---|
 |
Evidence index for seven shipped proof points; this profile carries the live 22-direct-PR readback, separate Hermes authorship evidence, and formal QA link. |
 |
Blind six-case agent-assurance challenge with 62 passing tests and downloadable evidence. Protocol v2 records 10/50 scoped outside actions from 10 validators, with 40 remaining, 7/10 technical events, zero independent verifier runs, and $0 settled revenue. The mix is three curator decisions, one independent technical reproduction, three integrations, and three substantive reviews; the qualifying Rask review is an unfavorable root-cause rejection, while the separate AgentTrust catalog decline stays at zero weight. The RCL review is not an independent verifier run. Efficacy, adoption, endorsement, certification, and paid demand remain INCONCLUSIVE. Source · Help verify or break it → |
 |
Maintainer-grade PR intake: repo context, overlap checks, dry-run scoring, and reproducible proof gates. |
 |
Kotlin control surface for supervising agent infrastructure from a phone. |
 |
Memory, verification gates, audit ledger, and human approval flow. |
 |
Public product shipping, deterministic scoring, and SHA-256 sealed credential proof. |
| Project | Contribution |
|---|---|
Hermes Agent #65682998e35313 authored commit |
Original PR #43861 closed unmerged after upstream fallback restructuring. Teknium's surgical reapply preserved my authorship, centralized per-entry key_env resolution in resolve_entry_api_key(), aligned TUI/desktop, cron, gateway, and CLI setup, and passed 565/565 targeted tests. |
Hermes Agent #69696OPEN · CREDITED |
Teknium's broader login-shell PATH implementation explicitly subsumes and credits my earlier PR #43834 for independently mapping the GUI/login-shell bug family. After another startup refactor, the August 6 refresh preserved login-shell resolution before backend selection and spawn, passed 12/12 shell-path tests, and returned required CI to green. This open PR is excluded from the merged totals above. |
Agent Security Harness #323PERMANENT CREDIT · NOT A DIRECT MERGE |
The maintainer's accuracy sweep permanently records my separate pinned 11/11 RCL reproduction in the upstream README. The broader sweep's DOI, guard, count, citation, and metadata findings remain the maintainer's work; this is one external reproduction, not endorsement or adoption. |
| GitHub Awesome Copilot #2476 | Added the non-executing verify-agent-action skill and generated install index after human approval and passing required checks. This is a community-skill integration, not GitHub certification or endorsement of VERITAS. |
| freedesktop-rs nmrs #521 | Replaced conflicting EAP certificate path/blob panics with InvalidInput. After comparing two contributor implementations, the maintainer selected my original error-return design, authored the final revision on my branch, and merged the VrtxOmega-authored PR. |
| AgentDoctor #18 | Added a GitHub Action while preventing output-path and symlink escapes; the repository owner reproduced the defect, required a focused remediation matrix, re-verified the correction, and merged the PR. That lifecycle remains one scoped external event. |
| Drift #792 | Fixed stale-index sampling so medium repositories are sampled across the full file range instead of only near the beginning; added a 39-file regression and merged. |
| Trail of Bits Dylint #2016 | Prevented a misleading-variable-name false positive when the type is explicit; synchronized rustdoc and generated README evidence after review, then merged. |
| awesome-ai-agent-governance #27 | Independent curator accepted the narrowly scoped Trust Lab entry. This establishes catalog fit, not VERITAS efficacy. |
| awesome-ai-security #18 | A second independent curator accepted the Trust Lab into a separate AI-security catalog, with the same discovery-only boundary. |
| awesome-ai-security-tools #29 | SCADA StrangeLove merged the Trust Lab into its new-project watchlist with the limitations preserved in the listing. |
| Kaoto forms #104 | Fixed the shared forms-library empty state so an already-selected All tab no longer suggests switching to itself; added focused regression coverage and merged into 1.x. |
| Karakeep #2864 | Fixed vertical scrolling for tall screenshot previews; shipped in Karakeep 0.33.1 with contributor credit. |
| Karakeep #2863 | Fixed BROWSER_WEB_URL IPv6 host resolution; shipped in Karakeep 0.33.1 with contributor credit. |
| Annotorious #610 | Made OpenSeadragon polygon simplification tolerance configurable while preserving default behavior. |
| Ansvisor #237 | Grouped prompt detail results by platform to avoid provider model slug drift. |
| Ansvisor #235 | Grouped insights results by platform to avoid provider model slug drift. |
| Amber #1116 | Added -r to generated shell-version prelude reads for ShellCheck warning coverage. |
| ShellCheck #3484 | Excluded _ from subshell assignment warnings. |
| codex-desktop-linux #495 | Fixed Linux feature parity for the Codex 26.609 app bundle. |
| codex-desktop-linux #302 | Added Wayland GPU rendering profile for COSMIC / Pop!_OS. |
| codex-desktop-linux #236 | Added GitHub CLI auth workaround docs for isolated XDG / DBus keyring paths. |
| awesome-mcp-servers #4922 | Added VrtxOmega/Ollama-Omega to the Tools section. |
| awesome-mcp-servers #4539 | Added omega-brain-mcp to Developer Tools. |
| harehare #1865 | Allowed pipeline expressions as function arguments via implicit block parsing. |
Also: Sibyl Memory Plugin beta QA report, a formal external QA deliverable for local-first agent memory.
Open the full ecosystem map
| Repository | What It Does |
|---|---|
| premature-contribution-firewall | Maintainer-grade PR/issue intake: dry-run scoring, repo context, overlap checks, adversarial tests, and reproducible proof gates. |
| omega-brain-mcp | MCP governance core: cross-session memory, verification gates, cryptographic audit ledger, and human approval flow. |
| hermes-mobile-android | Android/Kotlin control surface for supervising agent infrastructure from a phone. |
| breakaway | VibeJam 2026 2nd Place · zero-dependency, browser-native credentialing engine with deterministic scoring and SHA-256 sealed proof. Live → |
| coders-bible-desktop | Offline Tauri/Rust knowledge engine with local search and no cloud dependency. |
| yubikey-fido2-linux-auth | Practical Linux hardening guide for FIDO2/U2F login, sudo, SSH, and recovery workflows. |
| Repository / Artifact | What It Proves |
|---|---|
| veritas-agent-trust-lab | Public blind challenge for forged verdicts, parameter swaps, replay, correlated evaluators, evidence deletion, and silent monitoring. Live → |
| veritas-portfolio | Public evidence index for shipped proof points, external contributions, QA work, and operating principles. Live → |
| Shell Trials | Educational systems: a terminal RPG that teaches Linux by doing, not reading. |
| Omega SMC | Android mesh command center for private Tailscale operator infrastructure. |
| Sibyl QA Report | Formal beta QA deliverable for local-first agent memory, sandboxing, and trust/control behavior. |
| Repository | What It Does |
|---|---|
| Gravity-Omega | AI-powered IDE and local operator workbench. |
| OmegaWallet | Desktop Ethereum wallet with renderer-cannot-sign architecture and adversarial validation history. |
| SovereignMedia | Local media platform with cinematic desktop UI. |
| veritas-vault | Local-first AI knowledge retention with SQLite, RAG, briefs, and searchable memory. |
| VERITAS-Omega-CODE | Deterministic software verification specification. |
I do not treat AI output as truth.
I treat it as a claim.
A claim has to survive verification.
| Principle | What It Means |
|---|---|
| Local-First | Offline operation is the default. Network access is scoped and opt-in. |
| Deterministic Validation | AI outputs are treated as claims that need tests, logs, receipts, and repeatable checks. |
| Operator Model | High-risk actions need visible state, review, approval, and recovery paths before execution. |
| Maintainer Discipline | External contributions should be narrow, issue-fit, test-backed, and easy to review. |
Before software, I worked in cleanroom and industrial environments where mistakes have physical consequences: procedures, audits, containment, thermal spray, certification, quality control, and failure modes mattered before I ever touched a repo.
That shaped how I build now. I care whether a system can be checked, operated, repaired, and trusted under pressure.
| Lane | Tools |
|---|---|
| Languages | Python, TypeScript, JavaScript, Rust, Kotlin, Shell, Solidity |
| Systems | Linux, Android, Tauri, Electron, SQLite, GitHub Actions, Tailscale, YubiKey/FIDO2 |
| AI / agents | MCP, local LLM workflows, verification gates, audit logs, approval surfaces |
| Product surfaces | Desktop apps, mobile companions, local-first dashboards, documentation systems |
