Computer Science student at Florida Tech. I build local-first AI operator tools and proof-heavy developer systems for real work on a Mac.

Codex Mission Control is local traffic control for people running multiple Codex chats against real projects, browsers, inboxes, repos, and account surfaces.

It turns projects into missions, gives shared surfaces lane locks, keeps handoffs in outboxes, and forces risky public/account/payment work into exact approval packets before anything leaves the Mac.

• Core loop: projects -> missions -> lane locks -> approval packets -> optional Telegram remote
• Install: git clone, ./scripts/install.sh, then cmc status, cmc lanes, cmc packet, cmc dashboard
• Proof: CI, fresh-clone QA, local demo, smoke tests, launch card, dashboard visuals, and first-builder feedback path
• Boundary: local Mac coordination, not a hosted agent service, not an OpenAI product, not a way around login/MFA/approval gates

• Clawdeck: local-model mode for an existing OpenClaw/Codex workspace, with Ollama defaults, no-hosted-fallback checks, smoke tests, and handoff briefs.
• TaskProof: Playwright task specs into screenshots, DOM captures, console/network evidence, rerun scripts, and static proof reports.
• Boundary Atlas: TS/JS import graphs, cycles, deep imports, boundary drift, dead exports, hotspots, and offline architecture reports.
• Counterexample Studio: local property-testing workbench with deterministic seeds, minimal counterexamples, shrink traces, reruns, and repro snippets.
• Mentor-worker benchmark: local benchmark for mentor/worker LLM collaboration on deterministic coding repair tasks with objective pytest scoring.
• CHATTY Revival: public transparency experiment around a meme/community token, with explicit no-wallet, no-spend, no-fake-engagement, and no-investment-claim boundaries.
• PromptSmith: static browser game about repairing image-generation prompts by reading generated exhibit art.

• I use AI as a build accelerator without hiding that the work is AI-assisted.
• I prefer local-first systems, explicit safety boundaries, and runnable artifacts over vague agent demos.
• I package the proof: screenshots, tests, CI, reports, sample outputs, demo scripts, and install paths.
• My strongest lane is turning messy AI work into operator-grade tooling with clear controls.

Portfolio | LinkedIn | X