Workflows & Self-Service
Workflow Failures
The Workflows screen lists every workflow in your organization's active repositories. It opens on the workflows that need a look. When a GitHub Actions run fails or times out, CodeCargo explains why, about a minute after the run finishes. Failures are grouped by what you have to do about them, so you can start with the ones you can fix.
The Workflows screen
The header counts your workflows, the repositories they live in, and how many are reusable or dispatchable. Inside a project, the same screen covers only that project's repositories.
Five tiles sit above the list. Each one reads from the latest run of every workflow:
| Tile | What it shows | Click it to |
|---|---|---|
| Needs attention | Workflows in the Needs attention list: how many are failing now, and how many failed recently | Open the Needs attention tab |
| Running | Workflows with a run queued or in progress | Filter All workflows to running ones |
| Success rate | The share of finished latest runs that passed | Filter All workflows to passing ones |
| Drift | Workflows that aren't on their repository's default branch | Filter All workflows to drifted ones |
| Coverage | Workflows with guardrails on, and how many ran in the last 24 hours | — |
Click a filter tile again to clear its filter.
The screen has two tabs. Needs attention is described below. All workflows is the full inventory. Search it by name, path or repository, and narrow it with the filters in the column headers: Repository, Traits (Agentic, Standard, Reusable, Dispatchable), Branch, Last run, and, when your workflows have guardrail scores, Guardrails. Open a workflow for its code, its runs and its compliance score.
Needs attention
The screen opens on Needs attention when any workflow needs a look. The list holds two kinds of workflow:
- Workflows whose latest run failed or timed out.
- Workflows that failed in the last day, even if a later run has moved on. An intermittent failure stays in view after the next green run.
Groups
Failures are grouped by the action they need. Groups appear in this order, and empty groups are hidden.
| Group | What the failures have in common | What to do |
|---|---|---|
| Blocked by CargoWall | CargoWall blocked a connection the job needed | If the job needs that host, allow it in the CargoWall policy. The workflow doesn't need to change. |
| Fix in the code | The failure comes from the repository's code, tests, dependencies or workflow file | Troubleshoot opens it with the failure loaded for the AI assistant |
| Fix outside the repository | A secret, a permission, the network or an external service needs to change | Fix it where it lives. Editing the repository won't help. |
| Cause not known | The explanation hasn't run yet, or couldn't name a cause | Open the run on GitHub for the full log |
| Flaky or runner problems | The failure tends to pass on a re-run, already did, or the runner itself failed | Re-run it, and watch for the ones that keep coming back |
A CargoWall block stays in its own group, even when it also looks flaky. A failure that passed on a re-run, or that the runner caused, goes with the flaky ones. Every other failure stays in the group its cause belongs to, including after it recovers.
Within a group, failures that are still failing come first, then failures on the default branch, then the newest. Each group shows five rows; Show more reveals the rest.
Filters
- Default branch keeps only failures on each repository's default branch.
- Triggered by me keeps only runs you started.
Opening a failure
Each row shows the run's status, the workflow and its branch, the repository, the run number, what triggered it, who started it and when. It also shows the explanation's one-line headline and a chip naming the cause. Hover over the chip to read the full explanation.
Open a row for the rest of the explanation, how often the same failure has happened, and a short excerpt of the log. The excerpt comes from the failed job's log, or from GitHub's annotations for the job when the log can't be read. The GitHub icon opens the run on GitHub.
Troubleshoot opens a new workspace in the GenAI Editor with the workflow's repository, and the reusable workflows it calls. The AI assistant starts from the failed run: the explanation, the failed step and the log excerpt. It starts there even if a later run has already passed. When CargoWall caused the failure, the assistant is told to suggest allowing the host instead of editing the workflow. When the fix is outside the repository, it explains what needs to change instead of editing files. A failed Self-Service run has the same Troubleshoot button on its run page.
Recovered
A row is marked Recovered only when a re-run of that failure passed, or a newer run on the same branch passed. A green run on another branch does not recover it, and neither does a run that is still in progress or was cancelled. Recovered rows sort last within their group.
How CodeCargo explains a failure
The explanation names:
- The cause: Tests, Build, Lint / check, Dependencies, Network, Auth / permissions, Runner, Workflow config, External service, or Blocked by CargoWall. When the evidence doesn't point clearly at one cause, the chip reads Unclassified.
- Where it failed: the failed job and step.
- Whether a re-run is likely to pass: Likely flaky is a judgment from the failure itself. Flaky means a later attempt of that run already passed, even if the explanation expected the failure to stick. Build, lint, workflow-definition and CargoWall failures are never marked likely flaky, because they fail the same way every time.
- Whether the fix is in the repository, which decides between Fix in the code and Fix outside the repository.
- How often it has happened: for example, "The same failure happened 3 times in the last 7 days."
AI picks the cause from the run's log, GitHub's annotations and CargoWall's record of the run. It writes none of the text you read: every sentence is fixed wording, filled in with the job, step and host names from GitHub and CargoWall. An explanation is a label only. It never re-runs, cancels or changes anything.
The explanation also appears on the failed run in the workflow's Runs tab, in the Cause column, with the failed step beside it. The assistant in the GenAI Editor can read it too when you ask about the run.
When you don't see an explanation
Explanations are available on plans that include CodeCargo-managed AI. A bring-your-own-key (BYOK) plan does not receive them. Only failed and timed-out runs in active repositories are explained; cancelled runs are not. CodeCargo only adds an explanation if the run finished within the last day and does not have one yet. A burst of failures can take longer than a minute, because each organization gets a limited number of new explanations per hour. Contact your CodeCargo administrator if you expect explanations and never see them.
CargoWall
An explanation attributes the failure to CargoWall only when CargoWall blocked a connection while enforcing in the failed job, and either the failure looks like a connection or dependency problem or the failure text names a host that was blocked. A block in audit mode, or a block that does not show up in the failure, is not reported as the cause. When CargoWall is the cause, the explanation names the blocked host. When CargoWall blocked a host but something else caused the failure, the explanation says so in a separate sentence.
The same CargoWall block in several workflows is one row. Open it for each workflow, then open a workflow for the blocked hosts, the log excerpt and a link to the CargoWall run.
Repeated failures
When a workflow keeps failing the same way on its default branch, CodeCargo notifies your organization: its last three runs there all failed with the same error. You get one notification per failure, however long it lasts; further repeats update it. A different failure on the same workflow is a new notification. Manage it with your other notifications. For a fix you keep making by hand, an Agentic Job can make it for you and open the pull request.
