UNI Universal Natural Intelligence

Wiki · The Builder — colony cold start

START HERE — launch the entire UNI system

The Builder — colony cold start · START_HERE.md @ 44baf03d5041 (gen2-runtime) — opens the published snapshot ac338733bbba
1 value was removed from this page. Each one is marked in place as [redacted: category] 1 private address. Nothing else was altered. The document is otherwise exactly as it is written in the repository, and the sha256 below is of the original, so what was ingested stays checkable.

How to read this page

A Plain and a Clear version of this page have not been written yet. What follows is the document itself.

Precise — the source document

This is the document. Rendered from the repository at the commit above, with nothing rewritten for the web. A gate re-renders it on every deploy and fails the build if a single byte differs.

You (or an agent) just cloned the UNI repo. This one file is the complete launch guide. It has the one-line kickoff, the clean-clone steps, and the exact prompt + order for all five agents. Everything is on the default branch (gen2-runtime = main) — no feature branch to hunt down; a plain git clone gives you all the work.

Reading this means the clone worked. Confirm with git log --oneline -1 (a recent commit) and git status (clean, on gen2-runtime). CLAUDE.md is the binding contract and auto-loads for any Claude Code session opened in this repo folder — read it once.


⭐ THE ONE LINE that kicks it all off

Paste this to a fresh Claude Code agent in an empty folder, on any machine:

Run: git clone https://github.com/TMDLRG/UNI.MineCraft.git .    then read START_HERE.md in full and follow it exactly to bring the entire UNI 5-agent system online.
  • Private repo: the clone will ask for the operator's GitHub login if this machine has no stored credentials — authenticate as TMDLRG (or a collaborator).
  • If the clone says the folder is not empty (Claude may have created a hidden file), clone into a subfolder instead: git clone https://github.com/TMDLRG/UNI.MineCraft.git uni then work inside uni/ and read uni/START_HERE.md.
  • The clone checks out gen2-runtime (main) by default — that IS the full, current, stable repo.

What you cloned

  • Branch gen2-runtime (main). The lab/ozone-life-uni-hard-science feature branch was merged into main on 2026-07-13 — main now carries all the work (111 commits: the FE engine + gates, the DNS + observability + LLM discovery surface, the ops manual, and every handoff). Both branch names point to the same commit; work on main. No cross-shipping of a feature branch.
  • The binding docs: CLAUDE.md (rules) · docs/GAIA.md (the read-only, signal-only observability mirror — the studio track's "extra D," mirroring every track's state with full provenance) · docs/AGENT_HYDRATION_2026-07-14.md (the reboot control sheet: the roster + seams + the verified current true state) · docs/OPERATIONS_MANUAL.md (the full ops/tech map + the LLM REST surface at /api/discovery).

The 5 agents — who runs where, and which prompt

# Agent Runs on Launch prompt (in this repo)
1 Science / builder (the mind, the gates) a Claude Code session, in this repo folder docs/prompts/SCIENCE_AGENT_LAUNCH_PROMPT.md
2 Studio / Producer (the broadcast) a Claude Code session on the GPU/studio box (THINKER) docs/STUDIO_AGENT_LAUNCH_PROMPT.md
3 Legal-auditor (the evidence cross-examination — new) a Claude Code session, in this repo folder docs/prompts/LEGAL_AUDITOR_LAUNCH_PROMPT.md
4 Custom UNI-GPT (design advisor) Chrome — a custom GPT docs/prompts/CUSTOM_GPT_HYDRATION.md
5 OS / Mind (the on-chip UNI-OS) a Claude Code session on the remote chip PC / other account docs/prompts/UNI_OS_MIND_ENGINEER_PROMPT.md

Launch sequence — do these in order

Step 0 — confirm the clone (any shell, in the repo)

git status              # expect: on gen2-runtime, clean, up to date with origin
git log --oneline -1    # expect: a recent commit

Step 1 — Science / builder agent (start this first — it owns the mind + the gates)

  1. Open a new Claude Code session with THIS repo folder as the project (so CLAUDE.md auto-loads).
  2. Open docs/prompts/SCIENCE_AGENT_LAUNCH_PROMPT.md, copy the fenced block, and paste it as the first message.
  3. Its first job: verify the colony with the gates, then the stale-mind redeploy — which is gated on the operator's go-ahead for the streamed genome (live-stream guard) and a /lab-team-review MERGED VERDICT. Before touching uni-colony at all (redeploy or podman rm), it must run the mandatory capture-before-destroy procedure in docs/handoffs/GAIA_CAPTURE_BEFORE_DESTROY_2026-07-14.md — the colony minds live in the container's ephemeral filesystem and any podman rm destroys them.

Step 2 — Studio / Producer agent (the broadcast — on THINKER, the GPU box)

  1. On THINKER, open a Claude Code session in the repo folder.
  2. Paste the fenced block from docs/STUDIO_AGENT_LAUNCH_PROMPT.md.

    The studio CODE hardcodes the THINKER repo path (viewer/launcher.cjs, viewer/studio_up.ps1), so the studio agent's clone must live at that path on THINKER. The science/legal agents are path-portable.

Step 3 — Legal-auditor agent (the evidence — this prompt instantiates the role, which does not exist yet)

  1. Open a Claude Code session in the repo folder.
  2. Paste the fenced block from docs/prompts/LEGAL_AUDITOR_LAUNCH_PROMPT.md.

Step 4 — Custom UNI-GPT (Chrome)

Follow docs/prompts/CUSTOM_GPT_HYDRATION.md: paste Part A into the GPT's Instructions; upload the Part B cookbook files (from the repo root you cloned) to its Knowledge; start a fresh conversation.

Step 5 — OS / Mind agent (remote PC / other Claude account)

  1. On the remote machine, clone the repo (same one-line), open a Claude Code session in it.
  2. Paste the fenced block from docs/prompts/UNI_OS_MIND_ENGINEER_PROMPT.md.

    This agent works cross-box — it coordinates via git commits + docs/handoffs/*.md + operator relay, NOT a live channel. Hand off proof (gate output + commit hashes), never prose.


The flow (every agent lands on this)

One repo, one discipline. Every agent runs the same active-inference OODA loop: OBSERVE with gates (never process existence) → ORIENT by minimizing the gap between measured state and documented truth (VFE) → DECIDE the one next act with the most expected free-energy reduction (EFE) → ACT as code + doc + gate. One cure at a time. Receipts beat rhetoric — every claim carries its machine gate or says NOT VERIFIED. The claim fence is binding: a passing gate demonstrates the named behaviour, never experience, awareness, or life — never percent-scored. Keep warranted claims and over-claims visibly separate — that separation is the product. Gaia (http://127.0.0.1:8096/gaia on THINKER, MCP at viewer/gaia/gaia_mcp.cjs) is the live, queryable OBSERVE surface every agent can consult instead of re-deriving state by hand — it is a read-only, signal-only mirror of the gates and probes, never itself a source of truth beyond what it mirrors. Full statement: CLAUDE.md + docs/AGENT_HYDRATION_2026-07-14.md §0.

The current true state (verified 2026-07-13 — do not inherit stale docs; verify yourself)

  • The colony is UP but running a ~3-week-STALE mind. mc-server + uni-colony are up on the chip (ssh uni@[redacted: private-address]), RCON shows 6 UNIs + Director — but the running image uni-colony:v2 (2026-06-22) predates the Producer/Director/SP.Show layer, so GET :4000/producer/health 404s (deploy drift, not a source bug) and no Producer flies the camera. The science agent's first task is the redeploy from HEAD — behind the genome guard + owner go-ahead, and behind the mandatory mind-capture procedure below.
  • Gate ladder: 9 PASS · 6 PARTIAL · 8 PENDING · 0 FAIL (23 rows; source of truth: evidence/gates.ndjson → rendered docs/GATES.md). All 8 PENDING runners are pre-existing scaffolds, unrelated to Gaia. Critical- path gate = forage-pureworld-graduation (PENDING) — the on-air colony claim is fenced to it.
  • Honest science line: the emergent-forage loop is closed live only WITH a developmental metab_scale 0.2 runway (PARTIAL); pure-world self-sufficiency at scale 1.0 is NOT proven; the hunt-motor fix, not the FE cure, was the binding constraint.
  • Gaia (the observability mirror) is live — built this session, did not exist before. It lives at viewer/gaia/** (canonical doc docs/GAIA.md): a READ-ONLY, SIGNAL-ONLY MCP + UI mirroring every track — repo/git, gate ledger, infra registry, science-source excerpts, studio + colony probes, sessions, its own code, drift. GAIA LAW (enforced in code): every output is a direct signal with a full provenance triple (locator, captured_at, sha256, byte_len) — Gaia never summarizes, scores, ranks, or authors a verdict, and is read-only over science (never touches lib/sp/**, never sets a gate). It runs on THINKER at http://127.0.0.1:8096/gaia. Gates: gaia-slice1-live PASS, gaia-litigation-hold PASS (a WORM, hash-chained chain-of-custody store for the colony's brain .bin files — which live only in the colony container's ephemeral filesystem — with off-box replication verified byte-identical), and gaia-boot-persistent PARTIAL: crash-restart and a cold-start boot-launcher run are PROVEN; the one remaining leg is the literal reboot trigger, which is PENDING until the operator's next real reboot (an autonomous arbiter, viewer/gaia/gaia_boot_proof.ps1, will confirm it the moment that happens — no human judgment call needed). Anyone about to redeploy or podman rm uni-colony must read docs/handoffs/GAIA_CAPTURE_BEFORE_DESTROY_2026-07-14.md first and get a PASS from evidence_hold.cjs verify before destroying anything — Gaia cannot enforce this herself (that would break her read-only law over the colony), so the procedure is manual today.

The fence is the product. Every gate demonstrates a behaviour — none of it is evidence of experience, awareness, or life. Five agents, one discipline: measure the same world, refuse to claim past it, hand off proof not prose. Hold the fence and the vision holds with it. Now go to Step 0.

sha256 d083b39a357dc3ce — of the original file, so what was ingested stays checkable.