Vulnerability Scans
Scan your Docker images with Trivy on a schedule and see what to patch first.
Find and fix vulnerabilities
Section titled “Find and fix vulnerabilities”1. Turn on scanning
Section titled “1. Turn on scanning”- Open Agents and click the agent.
- On the Features tab, make sure Vulnerability management is on.
- Under Settings → Automations, turn on the vulnerability scan job.
- Optional: change the job’s schedule (5-field cron). The default
0 0 * * *runs daily at midnight. Changes apply without a restart.
2. Scan your images
Section titled “2. Scan your images”- Open Security in the sidebar.
- Click Scan all images.
3. Inspect the results
Section titled “3. Inspect the results”- Open Security. The overview combines every agent you can read into one risk score, tagged by agent.
- Check Highest-priority fix and Images to prioritize for what to fix first.
- Open Vulnerabilities to see every finding.
- Optional: turn on Fixable to show only findings with a known fixed version.
- Optional: click Export CSV to download every result that matches the current search and filters, across all pages.
See Overview cards for what each card shows.
4. Patch the image
Section titled “4. Patch the image”Arcane patches fixable OS-package vulnerabilities with Copacetic (copa), so you don’t have to wait for the publisher’s fixed version. You need the images:patch permission.
Patching needs Docker’s containerd image store. Without it, Arcane refuses to patch. Check with docker info -f '{{ .DriverStatus }}': it should list io.containerd.snapshotter.v1.
To turn it on:
-
Note the images you need, and plan a time when your containers can stop.
-
Add this to
/etc/docker/daemon.json, keeping any settings already there:{"features": {"containerd-snapshotter": true}} -
Restart Docker:
sudo systemctl restart docker. -
Pull or rebuild the images you need, then redeploy your containers.
Then patch:
- Open Security → Patches. Each row shows an image’s fixable CVE count, last scan, last patch status, and patched tag.
- Click Patch on a row. You can also patch from the image detail page or the Images table row menu.
- Follow progress in the Activity Center. Patches run in the background, one at a time.
- Redeploy your containers using the new tag (for example
nginx:1.25-patched).
copa can’t patch language dependencies (npm, pip, Go modules, and so on), locally built images with no registry source, or untagged images. Rebuild those images instead.
5. Check the fix
Section titled “5. Check the fix”- Arcane re-scans the patched image automatically. Check its result in the Patches table.
- For rebuilt images, click Scan all images again.
- Confirm the risk score dropped on the Security overview.
Schedule patching
Section titled “Schedule patching”- Under Settings → Security → Image Patching, turn on Scheduled Image Patching.
- In Automations, set the patch job’s schedule. The default
0 3 * * *runs daily at 03:00.
Each run patches local scanned images that have a stored report and no patch since their latest scan.
Block vulnerable updates
Section titled “Block vulnerable updates”Arcane can scan each new image before it replaces a running container, and block the update when the image has too many vulnerabilities. This is off by default and needs Vulnerability management on for the agent.
- Open Settings → Security → Vulnerability Scanning and pick the agent.
- Under Scan Images Before Updating, turn on Enabled.
- Set the highest count allowed for each severity.
-1means no limit. - Click Save.
| Severity | Default | Meaning |
|---|---|---|
| Critical | 0 |
Any critical vulnerability blocks. |
| High | -1 |
No limit. |
| Medium | -1 |
No limit. |
| Low | -1 |
No limit. |
The limits apply to scheduled auto updates and to manual updates, for containers and Compose projects. An update goes ahead only when every severity is at or under its limit.
- Arcane pulls the new image by digest, scans it, and deploys that same digest. The container keeps running on its current image until the scan passes.
- A blocked update stays on the Updates page with a Blocked status and the counts that exceeded the limits. Arcane removes the image it pulled for the scan.
- Scheduled runs skip a blocked update until a newer image is published or you change the limits. A manual update scans it again.
- If the scan itself fails, the update fails and is tried again on the next run.
- Scan settings such as .trivyignore and Only Report Fixable Vulnerabilities apply to these scans too.
Configure scans
Section titled “Configure scans”Scan settings are under Settings → Security → Vulnerability Scanning and apply to every scan. Re-scan to refresh saved results after you change them.
- Only Report Fixable Vulnerabilities records only vulnerabilities with a known fixed version. The Fixable table filter, by contrast, only hides saved results.
- Trivy Config (YAML) takes the contents of a
trivy.yamlfile. - .trivyignore takes one vulnerability ID per line. Lines starting with
#are comments.
Arcane’s own options win over the YAML: output format, timeout, cache directory, database or server settings, exit code, and fixable-only. To cap resources, use CPU Limit (cores), Memory Limit (MB), and Concurrent Scan Containers. See Settings.
Use a remote Trivy server
Section titled “Use a remote Trivy server”A Trivy server holds the vulnerability database, so each host doesn’t download and load it. Use one on small or 32-bit hosts, especially armv7, where the local database may fail to load.
- Turn on Use Remote Trivy Server.
- Set Trivy Server URL, for example
http://trivy.example.com:4954. - Optional: set Trivy Server Token if the server requires authentication.
Scan containers must reach the server URL. Trivy Network sets their network. The token is passed in the TRIVY_TOKEN environment variable, so it stays out of process arguments.
Threat intelligence
Section titled “Threat intelligence”Arcane raises the priority of vulnerabilities that are exploited in practice, using two public sources:
- CISA Known Exploited Vulnerabilities (KEV): CVEs the US Cybersecurity and Infrastructure Security Agency knows are exploited.
- FIRST EPSS (Exploit Prediction Scoring System): the probability a CVE will be exploited soon.
The Vulnerability Risk job syncs both at most every 20 hours. Run it from Automations to sync now. KEV comes from www.cisa.gov; EPSS scores come from api.first.org, for scanned CVE IDs only. A failed sync keeps the previous data.
For air-gapped installs, turn off Threat Intelligence under Settings → Security → Vulnerability Scanning. Scores then reflect only severity and container usage.
Reference
Section titled “Reference”Overview cards
Section titled “Overview cards”| Card | What it shows |
|---|---|
| Risk score | The combined score, its band, and the change from 7 days ago. |
| Highest-priority fix | The fixable finding with the highest risk, with its image, package, and fixed version. |
| Priority trend | Daily scores over 30 or 90 days. |
| Known exploited, Overdue KEV | Findings in the CISA KEV catalog, and those past CISA’s remediation due date. |
| Likely exploited | Findings with an EPSS probability of 10% or more. |
| Running critical/high, Images scanned | Critical and high findings in running containers, and how many images have been scanned. |
| Findings by severity, Findings by exposure | All findings, including ones without a fix. |
| Top vulnerabilities | The 10 highest-risk CVEs. Sort by Prevalence to see which CVEs appear in the most images. |
| Images to prioritize | The 5 highest-scoring images. |
In the Vulnerabilities table, Rows per page → All asks for confirmation, since loading every finding can slow the browser. It resets to 20 next time.
How scans run
Section titled “How scans run”- Scans use Trivy from the bundled tools image (
ghcr.io/getarcaneapp/tools), which pins the Trivy version. - Each scan uses the agent’s cache, network, resource, and server settings.
- Arcane saves findings, severity counts, and scan duration per image.
How patching works
Section titled “How patching works”- copa runs against the Docker daemon’s built-in BuildKit, with no helper container. It pulls with Arcane’s stored registry credentials.
- With a stored report from the latest scan, only its vulnerabilities are patched. Without one, every outdated OS package is updated.
- The result gets a tag suffix (
nginx:1.25→nginx:1.25-patchedby default) in the local daemon only. Nothing is pushed, and running containers aren’t touched.
Settings
Section titled “Settings”Find these under Settings → Security (Vulnerability Scanning and Image Patching). Schedules are on the Settings → Automations jobs.
| Setting | Purpose |
|---|---|
| Scheduled Vulnerability Scan | Turn scheduled scans on or off. |
| Vulnerability Scan Interval | Cron schedule for scans (default 0 0 * * *). |
| Trivy Config (YAML) | YAML configuration applied to every scan. |
| .trivyignore | Vulnerability IDs to exclude, one per line. |
| CPU Limit (cores) | CPU cores per scan container, such as 0.5; 0 for no limit. Hosts using CPU affinity use whole cores. |
| Memory Limit (MB) | Memory limit per scan container. |
| Concurrent Scan Containers | Maximum scan containers running at once, across manual and scheduled scans. |
| Only Report Fixable Vulnerabilities | Only record vulnerabilities that have a known fix. |
| Use Remote Trivy Server | Scan through a remote Trivy server. |
| Trivy Server URL | Remote Trivy server URL. |
| Trivy Server Token | Optional remote Trivy server token. |
| Trivy DB Registry | Registry the Trivy database is pulled from: ghcr.io (default) or docker.io. |
| Trivy Privileged | Run scan containers in privileged mode when the host security policy requires it. Off by default. |
| Threat Intelligence | Fetch CISA KEV and FIRST EPSS data for the risk score (default on). |
| Scan Images Before Updating | Scan each updated image and block it when it exceeds the limits (default off). |
| Critical to Low limits | Highest count allowed per severity for an update; -1 for no limit (defaults 0, -1, -1, -1). |
| Patched Image Tag Suffix | Tag suffix for patched images (default patched). |
| Patch Timeout (seconds) | Timeout for a single patch run (default 600). |
| Patch All Platforms | Patch every platform of a multi-platform image, not only the server’s own. |
| Scheduled Image Patching | Turn scheduled patching on or off. |
| Image Patch Interval | Cron schedule for patching (default 0 3 * * *). |
How risk scores work
Section titled “How risk scores work”Every finding, image, and agent gets a 0–100 score: severity weighted by exploitation and container usage.
Finding score
Section titled “Finding score”risk = min(100, 10 × severity × threat × usage)Severity is the CVSS (Common Vulnerability Scoring System) v3 score, or v2 if there’s no v3 score. Without either, Arcane uses the severity label: Critical 9.5, High 8, Medium 5.5, Low 2.5. Unknown is not scored.
Threat comes from threat intelligence:
| Evidence | Threat |
|---|---|
| In CISA KEV | 1.0, and severity is raised to 10 |
FIRST EPSS p |
0.7 + 0.3 × √p |
| No data | 0.7 |
A KEV finding in a running container always scores 100. KEV and EPSS only cover CVE IDs, so other advisory IDs, such as GHSA (GitHub Security Advisory), DLA (Debian LTS Advisory), and DSA (Debian Security Advisory), always use 0.7. With threat intelligence off, every finding uses 0.7.
Usage is 1.0 when a running, paused, or restarting container uses the image, 0.8 when only stopped containers do, and 0.5 when none do. If Arcane can’t list containers, usage is 1.0.
| Finding | Risk |
|---|---|
| KEV, running | 100 |
| CVSS 9.8, no threat data, running | 69 |
| CVSS 7.5, EPSS 36%, running | 66 |
| CVSS 7.5, EPSS 36%, stopped | 53 |
| CVSS 9.8, EPSS 1%, unused | 36 |
Image score
Section titled “Image score”Only fixable findings count, each CVE once per image at its highest risk. The score weights the top four findings, plus a term for the fixable count n:
score = 0.70 × 1st + 0.15 × 2nd + 0.05 × 3rd + 0.02 × 4th + 8 × (1 − e^(−n/40))The count term adds at most 8 points. For example, 12 fixable findings whose top four score 100, 66, 57, and 36 give 70 + 9.9 + 2.85 + 0.72 + 2.07 ≈ 86.
An image with no fixable findings scores 0. If none of its findings can be scored, the score is unavailable. Unrated non-CVE advisories are skipped.
Agent score
Section titled “Agent score”Start with the average of the agent’s scanned image scores, weighted by usage: running 3, stopped 2, unused or unknown 1. Images with no fixable findings count as 0. Then add 60% of the gap between that average and the agent’s highest image score, so one bad image isn’t averaged away.
A fixable KEV finding in a running or unknown-usage image sets a floor of 70. Each further such image adds 4, and one past its KEV due date adds 10, up to 95.
For example, a running image at 57, a stopped image at 53, and a clean unused image average (57 × 3 + 53 × 2 + 0) / 6 ≈ 46, and score 46 + 0.6 × (57 − 46) ≈ 53. Scores only include images still on the host with a completed scan, and skip ignored findings.
Combined score
Section titled “Combined score”The Security overview combines every agent into one score the same way: the usage-weighted average across all agents, pulled toward the highest image score on any agent. A large agent counts for more than a small one. The KEV floor applies across all agents.
Risk bands
Section titled “Risk bands”| Score | Band |
|---|---|
| 90–100 | Critical |
| 70–89 | High |
| 40–69 | Medium |
| 1–39 | Low |
| 0 | None |
Score history
Section titled “Score history”The Vulnerability Risk job computes scores hourly and records one daily snapshot in the TZ timezone. The 7-day change uses the latest snapshot that is 7–10 days old. Snapshots are kept 400 days. The trend restarts when the scoring formula changes.