run step, compute steps and GitHub actions. No step calls an AI model, so it costs no tokens and needs no AI account.
Before you begin
- The
reliantCLI, v1.21 or later, signed in (Sign in and tokens), with a connected machine that holds the repository (Run a machine from the terminal). The scan is a command, so this workflow always needs a machine. - The GitHub integration connected, with the Reliant GitHub App installed on the repository and push access for your own account. Issues are written as
reliant-labs[bot]by default. - A scanner on the machine. The default,
reliant forge ci vuln-scan, runs govulncheck andnpm auditin a forge project (Scan dependencies for vulnerabilities). Any command that exits non-zero on findings works.
1
Write the workflow
.reliant/workflows/dependency-sweep.yaml
0 ends at clean, 127 (command not found) ends at misconfigured, and anything else files. compute nodes evaluate CEL with no model and no shell. open_issue re-checks the search hits, because GitHub search matches words, not the exact prefix.2
Test both issue paths with scenarios
Mock the scan and the GitHub calls. This scenario covers a second night with the issue still open:Add scenarios for a clean night, a first finding, a missing scanner and a refused write, then run them:Test workflows with scenarios explains the format.
.reliant/workflows/dependency-sweep/scenarios/open_issue_gets_a_comment.yaml
terminal
3
Turn on the schedule
Commit the workflow, then activate its trigger with your repository:This test used a sandbox repository and a frontend-only project, hence
terminal
--npm. Enable under Workflows → Library does the same in the app.4
Fire it now and verify the issue
Don’t wait for 3am:Firing again two minutes later ran
terminal
scan shows FAIL because the scanner exited 1. That’s the finding, not a broken run. The issue, opened by reliant-labs[bot] with the labels security and dependencies:comment instead of file: the issue got a Still failing on 2026-11-20. comment and no duplicate was opened.5
Clean up
terminal
What this does not do
- It does not fix anything. Upgrading the dependency, or recording an exemption under
ci.vuln_scan.exemptionsinforge.yaml, is yours. An agent step could propose the upgrade, but then the run needs an AI account. - It does not close the issue. A clean night ends at
cleanand leaves the issue open. Close it when the fix merges, or add agithub/issue.update@1step on the clean path. - The exit code is the only signal. A scanner that fails for another reason also files an issue, with the reason in its body.
- The search can miss a fresh issue. GitHub indexes new issues with a short delay, so two runs seconds apart could each file one.
- The machine must be online at 3am. A suspended cloud machine is woken. Your own computer must be on, and firings missed for longer than the catch-up window (10 minutes by default) are dropped.
Troubleshooting
govulncheck is not on PATH, so the Go vulnerability gate cannot run
govulncheck is not on PATH, so the Go vulnerability gate cannot run
The gate refuses to pass a scan it could not run. Install govulncheck on the machine (
go install golang.org/x/vuln/cmd/govulncheck@latest), or put --npm in scan_command. In forge v1.6.1 a site project with no Go module asks for it too.The scan failed but no issue was filed: not found
The scan failed but no issue was filed: not found
GitHub answers “not found” when the Reliant GitHub App is not installed on the repository. Install it, or check the
repo input. A bot write also fails if your own account cannot push to the repository.Next steps
Scan dependencies for vulnerabilities
What
reliant forge ci vuln-scan gates on, and how to record an exemption.Agent on a schedule
Schedule triggers, overlap and the catch-up window.
Post a daily digest to Slack
Report to a channel instead of an issue.
How automations fire
What happens between the 3am slot and the run’s first step.