backlog(B-0790 P1): zero-dev-machines cluster-native architecture — all PRs from cluster; voice as primary operator surface (end-state target)

[AceHack]

Summary

Files the end-state architectural target the substrate is BUILDING TOWARD. Per the maintainer's 2026-05-26 two adjacent signals during the iter-4.2 test session:

"i want all the prs to come from the cluster mostly and dev machines are just conversational interfaces into the cluster and so is alexa"

"0 dev machines everything still works and i can talk to alexa for home automation / homelab persona users we want 0 dev machine needed just cluster and microphone"

Today's substrate has dev machines (Aaron / Max / Addison Macs) as primary substrate-engineering surface; cluster is deployment target. End-state inverts: cluster IS primary substrate-engineering surface; dev machines + Alexa + future microphones are conversational interfaces INTO the cluster.

Why file this as a row (rather than just remembering)

Every iteration between today and end-state needs the target named so it doesn't drift toward "make dev-machine substrate easier" (the wrong axis to optimize). Per the maintainer's broader 2026-05-26 "going for right not fast" discipline. Without B-0790, iter-4.3 → iter-5 → iter-6 risk optimizing the legacy axis.

ServiceTitan-demo-substrate also composes here: a demo where Aaron in front of stakeholders operates a remote cluster via voice with no laptop is substantively different from "look at this CLI tool I built." The end-state IS the demo.

Sub-targets (ship independently)

1. Cluster nodes commit + push to GitHub (per B-0789 iter-5 design — per-node SSH deploy keys auto-registered at install)
2. Autonomous-loop substrate runs ON cluster nodes (same <<autonomous-loop>> cron pattern Aaron's Mac runs today, but firing on cluster nodes)
3. Alexa-speaker → cluster direct integration (_alexa_speaker_acceptance block per the legal-risk-acceptance pattern; voice-mode is Bezos-tier business + voice-math per agent-roster-reference-card)
4. Future microphones connected directly to cluster (homelab persona; broadens B-0759; extends B-0778 hardware reference)
5. Dev machines become conversational interfaces ONLY (read-only observability; send-intent surface; NOT primary work substrate)

Acceptance

End-state-architectural, not single-PR:

• Cluster commits PRs (weekly → eventually daily)
• Cluster runs autonomous-loop substrate (tick shards land in docs/hygiene-history/ticks/ from cluster authorship)
• Alexa-speaker → cluster round-trip works for 3+ distinct operator intents
• Homelab-persona 30-minute zero-dev-machine demo
• docs/cluster-native-architecture.md migration path doc

Composes with

B-0754 / B-0759 / B-0770 / B-0776 / B-0778 / B-0780 / B-0782 / B-0787 / B-0788 / B-0789 + agent-roster-reference-card + tick-must-never-stop + autonomous-loop-per-tick + human-audit-and-legal-risk-acceptance-pattern + maintainers/aaron/legal-entities/inventory.md + memory/persona/max/PERSONA.md (per-maintainer-liability sub-section).

Test plan

• markdownlint clean
• BACKLOG.md regenerated (new B-0790 entry)
• Cross-references resolve to existing substrate
• CI passes (gate workflow + CodeQL)

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.