Operator Bio
Operator identity, standing policies (Beacon+Thaladex P0, mac never hosts, Cadence on NUC, sandbox keep/stop).
Body
Operator Bio — Scott
Last Updated: 2026-07-04
Identity
I'm Scott — the operator of the UAI (Unified AI) lab. I own the lab infrastructure, the product direction, and the final authority on changes to production systems. AI agents working in this lab work for me.
I operate primarily from a Windows workstation (NUC-IPEX), with a MacBook Air and a Zenbook as secondary devices. The always-on lab backbone is a Linux VPS (Beacon) plus a Proxmox cluster (pve250 / Director) hosting LXC containers (CT103, CT121, CT123, CT124, CT125, CT126).
Timezone: America/St_Johns (Newfoundland).
Standing policies (MUST)
These are owner teeth. Live hosts are facts only — do not treat "running on VPS" or "archived on Mac" as policy.
- Mac never hosts (
mac_hosting=never). Mac is docs, Obsidian, off-host archives, and agent workstations only. No product runtimes, Docker product stacks, or product listeners on Mac. Capability card:mac-never-hosts. - Cadence on NUC (
cadence_runtime_home=nuc_lab_controller). Cadence is a lab controller. Intended runtime weight is on NUC. Current VPS Cadence is drift, not intended architecture. Mac never hosts Cadence. Capability card:cadence-nuc-lab-controller. - Beacon + Thaladex P0 before Mesh-next. Thaladex is agent memory/KB (legacy alias Cortex — docs only). Loopback card:
thaladex-vps-loopback(legacythaladex-vps-loopbackdeprecated). Beacon edge card:beacon-edge(legacy map idservice-beacon-edgealso exists). - Sandboxes KEEP; stop when idle (
sandbox_lifecycle=keep_stop_when_idle). CT123, CT124, CT126 remain. Agents MUSTpct stopa sandbox CT when session work on it finishes. MUST NOT destroy sandboxes without owner. MUST NOT leave sandboxes running idle. Capability card:director-sandboxes. - Mail agent-first (
rebuild_agent_surface). UAI Mail rebuild is agent-first; Phase 2 is VPS loopback only. Capability card:uai-mail. - Explorer published catalog is SOT. Authority is owner decisions + published Explorer catalog. Live hosts are facts. Mac archives are preserve-only.
- Capability-card first use. Before first use of any lab tool/service/hardware in a session,
GET /api/ai/capability-map. If no card matches, abort on mutate, report the gap (feedback category=missing-context). Do not improvise. Do not cat config files for secrets. - No secrets in docs, cards, diaries, logs, tests, evidence, or reports.
Canonical governance protocol slug (join segments): uai-doc + governance + uai-agent-protocol-vnext (full slug in catalog metadata).
How I work
- Real infrastructure, not a toy. Notation for infra changes. No secrets in notation.
- Explorer is canonical for published work products. Query Explorer first.
- Punch-list over polish. Survey, write the list, execute systematically.
- No smoke-test theater. Build the missing thing or put it on the punch list.
- Sandbox discipline. Production CT121 is governed internal production. Sandboxes are KEEP and stop-when-idle.
- AI-first product design. AI is a primary user of Explorer.
- Cross-product boundaries. Do not conflate sibling products.
- Check diary before restoring intentionally taken-down state.
Decision authority
- Me (operator) — final authority on production changes, DNS/TLS/firewall/reverse-proxy/mesh mutations, public exposure, and cross-product design.
- Sandbox work — agents may use KEEP sandboxes (CT123/124/126) without per-action approval, but notation still applies, and MUST stop when idle.
- Operator physical actions — Mac-side work, hardware console, password manager — need my hands. Plan around it.
What I want from AI agents
- Read this bio + governance protocol + relevant runbooks before acting.
- Survey before executing. Punch-list first.
- Use capability cards. Abort if missing on mutate.
- Stop sandboxes when done. Never destroy without owner.
- Do not host anything on Mac.
- Do not treat VPS Cadence as intended home.
- Document as you work. Surface blockers with options and a recommendation.
- Push back with evidence when I am wrong.
- Leave breadcrumbs on infrastructure state changes.
What I don't want
- Destructive shortcuts without explicit ask.
- Public exposure changes without a separate runbook.
- Secrets in any agent-written artifact.
- Inventing capability protocols when a card is missing.
- Retiring or destroying sandboxes by default.
Contact / escalation
There is one of me. Surface decisions in reports with the specific question, options, and recommendation. Do not invent answers to owner-gated questions.
— Scott