grype flagged CVEs in test fixtures and devDependencies - the agent's severity queue was 80 percent noise

TL;DR

Scan production artifacts, not the whole source tree. Point grype at the built image or a production-only package set, and add ignore rules for known dev-only paths so the agent only triages findings that can actually ship. Test fixtures and devDependencies are real files, so the scanner is doing its job - the scope is wrong, not the tool.

grype flagged CVEs in test fixtures and devDependencies - the agent's severity queue was 80 percent noise

Steps

1. Confirm the noise is really dev-only. Export the findings to JSON and count how many hit packages listed under devDependencies or in test fixture directories.

• Run: grype dir:. -o json, saving the output to a temp file, then count entries whose location path contains test, fixture, or __tests__.
• Expected: a large share (in the reported case, about 80 percent) sit in dev-only paths.

1. Switch the scan target from the source tree to the production artifact. Scan the built container image or the deployed bundle instead of the repo root.

• Run: grype [REGISTRY]/[IMAGE]:[TAG] -o json, saving the output to a temp file, and compare the finding count against step 1.
• Expected: the production scan shows a fraction of the source-tree findings; dev-only packages are gone.

1. For source scans you must keep, generate a production-only package set. Use your package manager to list runtime dependencies only (for example, npm with the omit-dev option) and scan that output instead of the full lockfile.

• Expected: dev-only packages no longer appear in the finding list.

1. Add ignore rules for the dev-only paths that survive. Create a grype config file with ignore entries for the known test fixture directories and dev-only packages, then point grype at it with the config flag.

• Run: grype dir:. -c .grype.yaml -o json, saving the output for comparison.
• Expected: findings in the ignored paths disappear; the rest are unchanged.

1. Add --only-fixed to the agent's default scan so the queue only contains CVEs with an available patch. Unfixable dev-only findings are the noisiest slice.

• Run: grype [IMAGE] --only-fixed --fail-on high.
• Expected: the queue shrinks to actionable items with fixes available.

1. Wire a weekly review of the ignore list into the agent's loop. Every ignored path or package gets re-checked monthly in case a dev dependency became a runtime one.

• Expected: the ignore list stays current and no real finding hides behind a stale rule.

Use this when

• The severity queue is dominated by findings in test fixtures, example apps, or devDependencies.
• The agent scans the repo root or a directory that includes dev tooling.
• Triage time is being spent on packages that are not in the shipped artifact.
• Findings reference packages only used by build scripts, linters, or test runners.

Not for this skill when

• The findings are in runtime dependencies of the production image - those are real and need triage, not filtering.
• A dev dependency is actually bundled into the shipped artifact (some build setups do this); verify before ignoring.
• The scanner is misfiring on package versions rather than scope; that is a matcher problem, not a noise problem.
• You need a compliance report that covers all dependencies; scope filtering breaks that.

Variant phrasings

• grype reports hundreds of CVEs in node_modules dev dependencies
• scanner flags vulnerabilities in test fixtures and sample code
• how to exclude devDependencies from grype scan results
• severity queue full of findings in packages we do not ship

Why it happens

Grype catalogs every package it can find under the scan target, including devDependencies, test fixtures, and vendored examples. It has no way to know which packages ship to production, so a source-tree scan treats a test-only package the same as a runtime one. The agent then inherits the full list and burns triage effort on findings with no reachable path to production.

Edge cases

• Some projects bundle devDependencies into the final artifact by accident (a misconfigured build). Re-verify scope after every build change.
• --only-fixed can hide a critical CVE with no fix yet. Keep a separate watch list for unfixed criticals in production scope.
• Ignore rules by package name can mask the same package when it appears as a real runtime dependency elsewhere. Prefer path-based ignores where possible.
• Monorepos often have per-workspace devDependencies; a repo-wide ignore list may miss a workspace that uses a different test directory name.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstucoo4UpYxNQo5E6hYrVRw