Find the symptom, follow the link.
Work is not starting
| Symptom | Likely cause | Where to look |
|---|---|---|
Task stuck in PENDING | All agent slots busy | Capacity and agent slots |
| Nothing starts at all | Agent-hour allowance exhausted | Limits and quotas |
| Lower-priority work never runs | Queue saturated by P0 tasks | Task statuses |
Work is failing
| Symptom | Likely cause | Where to look |
|---|---|---|
FAILED early in the run | Build or test command wrong for a clean workspace | Build and test configuration |
| Tests pass locally, fail in Coroid | Suite depends on a service or setup step | Build and test configuration |
| Task hits the run-time ceiling | Task too large | Tasks, plans, sprints and lanes |
Repeated NEEDS_HUMAN_REVIEW | Under-specified work | Specifications |
| Stuck mid-run | Several causes | When a task gets stuck or fails |
| Dependency install fails with an auth error | Private registry credentials not configured | Language support and the agent environment |
| A tool the build needs is missing | Toolchain outside the default workspace image | The agent environment |
Output is not what you wanted
| Symptom | Likely cause | Where to look |
|---|---|---|
| Solves the wrong problem | Acceptance criteria too loose | Specifications |
| Change is much larger than asked | No out-of-scope line | Specifications |
| Ignores your conventions | Missing project context | Project context |
| Plans are consistently weak | Work running under too small a profile | AI profiles |
| Passes every gate but is still wrong | Gates check properties, not intent | Quality gates |
Repository and access
| Symptom | Likely cause | Where to look |
|---|---|---|
| Repository missing from the list | App not granted that repo, or token scope too narrow | Connect your code |
| Connection worked, now does not | Token owner lost access, or token expired | Connect your code |
| Pull requests open against the wrong branch | Base branch misconfigured | Repository and branch settings |
Releases and deployments
| Symptom | Likely cause | Where to look |
|---|---|---|
Release says Preparing after a gate pass | Required approval or current release-note approval is still missing | Configure the release gate |
Release says Blocked | Current environment gate has blockers, or a temporary checklist exception expired | Configure the release gate |
| Deployment request is unconfirmed | Provider run could not be verified; avoid a new dispatch until you inspect it | Connect deployment events |
| Pipeline succeeded but no shipment appears | Callback missing, late, mismatched, or quarantined | Connect deployment events |
| Deployment shipped but health needs attention | Post-deploy canary failed | Follow health and recover |
| Need to return to a prior commit | Choose a previously healthy release SHA and request rollback | Follow health and recover |
Billing
| Symptom | Likely cause | Where to look |
|---|---|---|
| Model calls stopped | Credits exhausted | Model credits |
| Unexpected spend | Expensive profile on a high-volume agent | AI profiles |
| Cannot attribute cost | Use per-project reporting | Usage and reports |
Still stuck
Collect the task id, the status it is in, and the stage that failed before contacting support. Those three facts resolve most questions immediately.