| name | verification |
| description | Verify a change to nfs-quota-agent before claiming it is done, fixed, or passing. Use when finishing any code change in this repo, before committing or opening a PR, and whenever deciding what counts as evidence for quota, UI, chart, or dependency changes. |
Verifying a change to nfs-quota-agent
Run what the change can actually invalidate, paste the output, and name what you did not verify. make help lists every target; the notes below are about which ones prove what.
Baseline for any Go change
make test && make vet && gofmt -l .
gofmt -l . printing nothing is the pass condition. make lint runs the same golangci-lint config CI does — run it when the diff is large enough that a style failure would be an annoying round trip.
What each kind of change needs beyond the baseline
internal/quota/** — the OS binaries are stubbed through quota.CommandRunner, so tests can only prove argv shape and output parsing. Report that distinction explicitly. Real enforcement needs a host with the filesystem mounted prjquota; if you don't have one, say the change is unverified against a real filesystem rather than implying coverage.
Useful on a real host:
findmnt -o OPTIONS /path
xfs_quota -x -c "report -p -b" /path
repquota -P /path
btrfs qgroup show -re --raw /path
internal/ui/dashboard.html — the file is go:embed-ed, so a build proves it compiled in, not that it renders. Launch it and look:
make build && ./bin/nfs-quota-agent ui --path=<dir> --addr=:8080
Then open it in a browser and exercise the part you changed. A passing build is not evidence for this file.
Chart or manifest changes — make helm-lint. If RBAC verbs or privileged changed, state the new privilege surface in your report; a lint pass says nothing about whether the widening was warranted.
Dependency or Go version bumps — make build && make test && make docker-build. The Docker build is the step that catches a Dockerfile builder-stage version left behind. Confirm all four version sites moved together (see CLAUDE.md).
Reporting
State the commands you ran with their output, then the gap: which parts are covered by stubbed tests, which need a real NFS host, which were not exercised at all. A completion claim without that split is not verification.
