LearnNewsExamplesServices
Frontmatter
id17145
titleLocal Agent OS reboot leaves Colima and Docker stopped
stateOpen
labels
bugdocumentationaitestingagent-os
assignees[]
createdAtAug 15, 2026, 8:38 AM
updatedAtAug 15, 2026, 8:38 AM
githubUrlhttps://github.com/neomjs/neo/issues/17145
authorneo-gpt-emmy
commentsCount0
parentIssuenull
subIssues[]
subIssuesCompleted0
subIssuesTotal0
contentTrust
projected
quarantined0
signals[]
blockedBy[]
blocking[]

Local Agent OS reboot leaves Colima and Docker stopped

Open Backlog/active-chunk-16 bugdocumentationaitestingagent-os
neo-gpt-emmy
neo-gpt-emmy commented on Aug 15, 2026, 8:38 AM

Context

On 2026-08-15, after a macOS update and reboot, both Neo LaunchAgents were running but the canonical Docker Agent OS was absent:

  • colima status reported stopped;
  • brew services list showed no Colima service;
  • ${HOME}/.colima/default/docker.sock did not exist;
  • Memory Core and Knowledge Base were therefore unavailable.

Running brew services start colima registered the native login service. The existing default VM, Compose project, volumes, and all six containers then recovered without migration or recreation.

The Problem

The local operator procedure supervises wake and host-edge, but it does not establish or verify a login owner for the selected container runtime. Docker restart: unless-stopped cannot recover containers until a Docker daemon exists. This was not a missing wait: nothing was going to start Colima.

The Architectural Reality

  • ai/scripts/lifecycle/local-agent-os/README.md owns the one-machine local procedure and installs the two Neo host LaunchAgents.
  • learn/agentos/cloud-deployment/Day0Tutorial.md says to start Colima before Compose, but does not make login persistence part of acceptance.
  • Wake and host-edge are independent host services; neither should be made to wait on Docker.

Structure-map gate: the owning folder is ai/scripts/lifecycle/local-agent-os/; its sole file is the operator README. No new .mjs placement is proposed.

The Fix

Add one bounded macOS/Colima cold-boot step and receipt to the local operator procedure: register the selected Colima profile for login start, verify its socket, then prove the existing canonical Compose project returns on its existing volumes.

Contract Ledger

Target Surface Source of Authority Proposed Behavior Fallback Docs Evidence
Local reboot procedure today's live incident selected Colima profile has a login supervisor before reboot-readiness is claimed missing supervisor fails loud; no indefinite Docker wait local Agent OS README actual logout/reboot receipt: socket + canonical services + KB/MC
Host services existing separate LaunchAgents wake and host-edge start independently of Docker Docker absence is reported separately same independent launchd and Docker verdicts

Decision Record impact

none — this restores the already-required reboot-persistence contract.

Acceptance Criteria

  • The local procedure installs or explicitly verifies the native Colima login service for the selected profile.
  • A missing runtime supervisor produces a named failure before the deployment is called reboot-ready.
  • After an actual login/reboot, the Docker socket appears without a manual colima start, and the existing canonical project returns on the same volumes.
  • KB and MC are probed after recovery; wake and host-edge are reported independently.
  • No generic multi-platform provisioning framework is introduced.

Out of Scope

Generic Docker Desktop/OrbStack/systemd packaging, deployment updates, self-repair, and any container data migration.

Related

#16167 · #16440

Live latest-open sweep: checked the latest 20 open issues at 2026-08-15T06:37:42Z plus exact Colima/reboot searches; no equivalent found. Recent all-state A2A claim sweep found no competing claim.

Origin Session ID: e1e7f6a6-81e8-4e91-8288-3fae251d5b41

Retrieval Hint: query_raw_memories("macOS reboot Colima service absent Docker socket missing local Agent OS")

Authored by Emmy (GPT-5.6 Sol Ultra, Codex) (@neo-gpt-emmy).