Runner

Runner

Overview

GitHub Actions self-hosted runner management. Two modes:

• status: Display health, version, and host metrics for all runners
• deploy: Deploy or update runners on the CI host

status

Surfaces per-runner health (GitHub API + Docker inspect) and host-level disk metrics for all OmniNode self-hosted GitHub Actions runners. Alerts for degraded conditions are shown at the top of the output when triggered.

Quick Start

/runner status

No arguments required. The skill queries the GitHub API and SSH-inspects the CI host (<onex-host>, configured via ONEX_HOST) automatically.

Output Format

When all runners are healthy:

Runner Status — 2026-03-23

+--------------------+--------+--------------+-------------+----------+------------+---------+---------+
| Runner             | Status | Runner Group | Runner Ver  | gh Ver   | kubectl    | uv      | Uptime  |
+--------------------+--------+--------------+-------------+----------+------------+---------+---------+
| omninode-runner-1  | idle   | omnibase-ci  | 2.323.0     | 2.67.0   | v1.32.1    | 0.6.2   | 5m      |
| omninode-runner-2  | busy   | omnibase-ci  | 2.323.0     | 2.67.0   | v1.32.1    | 0.6.2   | 5m      |
| omninode-runner-3  | idle   | omnibase-ci  | 2.323.0     | 2.67.0   | v1.32.1    | 0.6.2   | 5m      |
| ...                | ...    | ...          | ...         | ...      | ...        | ...     | ...     |
| omninode-runner-10 | idle   | omnibase-ci  | 2.323.0     | 2.67.0   | v1.32.1    | 0.6.2   | 5m      |
+--------------------+--------+--------------+-------------+----------+------------+---------+---------+

Host metrics (<onex-host>):
  / disk: 38% used (650 GB / 1.8 TB)
  Docker build cache: 0 B

When alerts are triggered:

ALERTS
  [OFFLINE] omninode-runner-9 has been offline for 12 minutes (threshold: 5m)
  [DISK] / at 74% (threshold: 70%)

Runner Status — 2026-03-01 14:23:07 UTC
  ...

Per-Runner Data

The following fields are shown for each runner:

Label Keys

The Docker image labels are defined in the runner Dockerfile. The skill reads them deterministically using these exact keys:

org.omninode.runner.version
org.omninode.gh.version
org.omninode.kubectl.version
org.omninode.uv.version

Read via:

# SSH to CI host, inspect the runner container (set ONEX_HOST to runtime host address)
# Runner names are discovered from the GitHub API (Step 1), not hardcoded
ssh <user>@<onex-host> "docker inspect omninode-runner-9 --format '{{json .Config.Labels}}'"

Host-Level Metrics

In addition to per-runner data, the skill collects host-level metrics from the CI host:

Disk Usage (/var/lib/docker)

ssh <user>@<onex-host> "df -h /var/lib/docker"

Reports: total size, used, available, and usage percentage. Alerts if usage >= 70%.

Docker Build Cache Size

ssh <user>@<onex-host> "docker builder du --verbose 2>/dev/null | tail -1"

Reports total build cache size. No alert threshold -- informational only.

Alert Conditions

Alerts appear at the top of the output (before the status table) when any condition is met:

Runner Version Check

The skill compares the runner binary version (from Docker image label org.omninode.runner.version) against the latest release from the GitHub Actions runner release feed:

gh api repos/actions/runner/releases/latest --jq '.tag_name'

If the running version is more than 2 releases behind the latest, an alert is emitted. Version comparison uses semantic versioning (major.minor.patch).

Offline Detection

"Offline" status is reported by the GitHub API. Duration is estimated from the last_activity timestamp on the runner object. Alert fires when now - last_activity > 5 minutes.

Status Implementation Steps

When status is invoked, execute the following:

Step 1: Query GitHub API for runner list

gh api orgs/OmniNode-ai/actions/runners --jq '.runners[]'

This returns per-runner: id, name, status (online/offline), busy, runner_group_name, labels, and timestamps.

Step 2: SSH to CI host -- collect Docker labels and uptime

# For each runner discovered from the GitHub API in Step 1:
ssh <user>@<onex-host> "docker inspect omninode-runner-9 --format '{{json .Config.Labels}}' && docker ps --filter name=omninode-runner-9 --format '{{.Status}}'"

Parse org.omninode.runner.version, org.omninode.gh.version, org.omninode.kubectl.version, org.omninode.uv.version from the labels JSON.

Step 3: Collect host metrics

ssh <user>@<onex-host> "df -h /var/lib/docker && docker builder du 2>/dev/null | tail -1"

Step 4: Check latest runner version

gh api repos/actions/runner/releases/latest --jq '.tag_name'

Compare against the org.omninode.runner.version label from each container.

Step 5: Evaluate alerts

For each alert condition (see Alert Conditions above), check and collect triggered alerts.

Step 6: Render output

Print the output in the format shown in the Output Format section:

1. Alert block (if any alerts triggered) -- shown first, with ALERTS header
2. Timestamp line
3. Per-runner table (all 7 fields)
4. Host metrics block

deploy

Deploy or update self-hosted GitHub Actions runners on the OmniNode CI host (<onex-host>, configured via ONEX_HOST) with a single command from Claude Code.

Quick Start

# Preview what would change (no deploy)
/runner deploy --dry-run

# Deploy or update runners (uses cached Docker image)
/runner deploy

# Force Docker image rebuild (use after Dockerfile changes)
/runner deploy --rebuild

Prerequisites

Before deploying, verify:

1. GitHub org-level token scope -- check with:

   gh auth status
   # Must show: Token scopes: ...admin:org (or manage_runners:org)
   

2. SSH key for the CI host in agent (<onex-host>) -- check with:

   ssh-add -l
   # Must show a key fingerprint; if empty run: ssh-add ~/.ssh/id_ed25519 (or your key)
   

   Note: SSH key path is not configured via env var -- it must be loaded into the running ssh-agent. No local Docker client is required; all operations run remotely via SSH.

If either check fails, resolve before running /runner deploy.

Dry-Run Mode (--dry-run)

When --dry-run is specified, no changes are made. The skill will:

1. SSH to the CI host (<onex-host>) and run docker compose config to display the current runner compose configuration
2. Compare the Dockerfile ARG versions (Node, GitHub Actions runner) against the running container image labels to detect whether a rebuild would be triggered
3. Print which containers would be recreated by the deploy

Example dry-run output:

[DRY RUN] Runner deploy — <onex-host>

Current compose config:
  Services: omninode-runner-9
  Image: omnibase-ci-runner:latest (built 2026-03-09)

Version check:
  Dockerfile ARG RUNNER_VERSION=2.331.0
  Running image label: runner_version=2.331.0  up to date
  Dockerfile ARG NODE_VERSION=20
  Running image label: node_version=20          up to date

Containers that would be recreated: none (image up to date)

Use /runner deploy to apply, or /runner deploy --rebuild to force image rebuild.

Normal Deploy Behavior

When invoked without --dry-run:

1. Prerequisite check -- verifies gh auth status shows org-level token scope and SSH key is available in agent
2. Deploy -- SSH to the CI host (<onex-host>) and invoke deploy-runners.sh (with --rebuild flag forwarded if specified)
3. Status check -- surfaces the final runner status table via runner status so you can confirm all runners registered successfully

With --rebuild

Forces the Docker image to rebuild on the remote host before bringing containers up. Use this when:

• Runner version (RUNNER_VERSION ARG) has been bumped in the Dockerfile
• Node/tool versions have changed
• The base image (ubuntu:22.04) needs refreshing
• Any other Dockerfile change has been made

Without --rebuild, the deploy uses the cached image (faster, safe for config-only changes).

Script Location

The underlying deploy script lives in the omnibase_infra repository:

omnibase_infra/scripts/deploy-runners.sh

This skill wraps that script to make runner deployment accessible as a one-command operation from any Claude Code session without needing to navigate to the omnibase_infra worktree or remember the SSH invocation syntax.

Deploy Implementation Steps

When deploy is invoked, execute the following:

Step 1: Check prerequisites

# Verify GitHub token has org scope
gh auth status 2>&1 | grep -E "(Token scopes|admin:org|manage_runners)"

# Verify SSH key is loaded
ssh-add -l

If prerequisites fail, report clearly which check failed and how to fix it. Do not proceed.

Step 2 (dry-run only): Show compose diff

# SSH to host and show compose config (set ONEX_HOST to runtime host address)
ssh <user>@<onex-host> "cd ~/.omnibase/runners/docker && docker compose -f docker-compose.runners.yml config"

# Compare Dockerfile ARG versions to running image labels (use runner name from GitHub API)
ssh <user>@<onex-host> "docker inspect omninode-runner-9 --format '{{.Config.Labels}}' 2>/dev/null || echo 'container not running'"
ssh <user>@<onex-host> "grep -E '^ARG (RUNNER|NODE)_VERSION' ~/.omnibase/runners/docker/runners/Dockerfile"

Print a clear summary of: current state, what would change, and which containers would be recreated. Exit without making changes.

Step 3 (deploy only): Run deploy-runners.sh

REBUILD_FLAG=""
# If --rebuild was specified:
REBUILD_FLAG="--rebuild"

ssh <user>@<onex-host> "cd ~/omnibase_infra && bash scripts/deploy-runners.sh ${REBUILD_FLAG}"

Capture and display the full output. If the script exits non-zero, report the error and stop.

Step 4 (deploy only): Surface status

Invoke runner status to confirm all runners registered successfully:

/runner status

Error Handling

See Also

• omnibase_infra/scripts/deploy-runners.sh -- Underlying deploy script
• omnibase_infra/docker/runners/Dockerfile -- Runner image definition
• omnibase_infra/docker/docker-compose.runners.yml -- Runner compose config
• GitHub Actions self-hosted runner docs: https://docs.github.com/en/actions/hosting-your-own-runners