Self-hosted runner laptop-extra: multi-repo CI fit (tracking)

[Coldaine]

Summary

This issue tracks which Coldaine repos are a good match for the repo-registered self-hosted runner laptop-extra on Nobara/Fedora (labels: self-hosted, Linux, X64, fedora, nobara).

Smoke test: Runner Test workflow run completed successfully on this runner (2026-04-13).

Operational notes

• Single listener: Do not run ./run.sh while actions.runner.Coldaine-ColdVox.laptop-extra.service is active; a second Runner.Listener causes broker session Conflict / "session already exists".
• Install: /home/coldaine/actions-runner, runner 2.333.1 (matches current actions/runner latest as of scan).
• ColdVox main: Self-hosted jobs use runs-on: [self-hosted, Linux, X64, fedora, nobara] (see ci.yml, release.yml, runner workflows).

Repo fit (from automated scan)

Good fit

Repo	Notes
Coldaine/ColdVox	Already targets this runner; some jobs remain ubuntu-latest by design.

Possible with workflow edits (ubuntu-only CI today)

Repos whose workflows use ubuntu-latest only (no macOS/Windows matrix in the scanned files). Candidate for runs-on: [self-hosted, Linux, X64, fedora, nobara] if you register the runner for that repo (or use an org runner) and accept Nobara vs Ubuntu differences:

• ActuarialKnowledge, ColdOracle, ColdSearch, ColdTrace, IdeaEmergence, NetworkConfig, PersonalKnowledgeBase, PinterestDownloader, ProjectBroadsideStudio, ShortcutSage, TastyTrade, coldaine-infra, coldaine-workflows, ddr5-rdimm-oracle, llm-archiver, python-roborock, the-watchman

Poor fit (multi-OS / matrix / containers called out in scan)

• OpenCode-Config-Manager — Windows + macOS + Ubuntu in CI.
• inbox-zero — mac + Linux release matrix.
• prism-mcp, rtabmap, rtk — multi-OS matrices and/or container usage per scan.

No .github/workflows on default branch (32 repos)

Including: AgentVisualCrazy, AlexaChat, ClaudePlanning, IngestEVERYTHING, KnowledgeGraphSystem, LocalLargeLanguageModels, MoneyWizard, OffhandProjects, OneDriveCleanup, PolyformPipeline, SpeciationViz, Uigraphnotes1, auto-accept-agent, azurlane-nav-state-machine, coldaine-control-plane, coldaine-vscode-helpers, colossus-blender-mcp-viga, frozenSkillz, kimi-codex-spark, llm-governed-store-skeleton, loominary, mem0-chrome-extension, oasis-docs-scraper, obsidian-vault-backup, rc-race-replay-poc, recall-pipeline, repo-investigation, state-cartographer, sunscope, the-Librarian, tv-control, wispr-word-corpus, zo-workqueue, zoask.

---

Next steps (optional): For each "Possible" repo, either add a repo-level runner registration or move to an org-level runner group, then PR runs-on (and validate apt vs dnf steps).

Subagent scan reference: broad gh api pass over Coldaine repos; ~80 REST calls, no rate limit errors.

[Coldaine]

Automation added (no manual discipline required)

This machine now enforces the single-listener rule automatically:

1. ~/actions-runner/run.sh wrapper — If actions.runner.Coldaine-ColdVox.laptop-extra.service is active, ./run.sh exits immediately with instructions (prevents accidental duplicate Runner.Listener / broker Conflict). Real script is run.upstream.sh.

2. Systemd timer actions-runner-dedupe.timer — Every 2 minutes (and ~90s after boot), /usr/local/sbin/actions-runner-dedupe.sh kills any stray Runner.Listener run process that is not --startuptype service while the unit is active.

Multi-repo reality check

GitHub user Coldaine is a user account, not an org — there is no org-level runner token without moving repos under an organization and admin:org scopes. This registered runner only serves Coldaine/ColdVox until you add separate runner installs (other directories + config.sh per repo) or use an org with shared runners.