Start with the run Details, then use Activity and Diagnostics to locate the phase that failed.

Admission

SymptomCheck
401 authentication_errorBearer header, key value, key mode, revocation and environment
403 permission_errorRequired API-key scope
Model not foundActive agent slug and workspace ownership
Billing readiness failureBilling-owner workspace and payment state
List active model identifiers:
curl "$SALAMBO_BASE_URL/api/v1/models" \
  -H "Authorization: Bearer $SALAMBO_API_KEY"

Deployment

SymptomCheck
Manifest rejectedsalambo manifest --path . --json diagnostics
Source upload failedAPI authority, archive size and network
Managed build rejectedUnsupported image or project configuration
Missing secret variableExport the fromEnv variable before deploy
Deployment never becomes readyDeployment status and build Diagnostics

Sandbox

SymptomCheck
Sandbox cannot startDeployment readiness, region and machine availability
Runtime contract errorManaged image health and runtime update guidance
Tool command deniedRequested path and workspace permissions
External request deniedEgress policy and host-only allowlist

Provider

SymptomCheck
Authentication failureProvider secret and exposure target
Provider unavailableProvider status and bounded retry
Model rejectedProvider name, model identifier and account access
Missing provider key in sandboxThe secret may be runtime-only by design

Extensions and tools

SymptomCheck
Extension missingextensions path and manifest diagnostics
Tool inactiveagent/settings.json active tool list
Extension startup failedEntrypoint, loading mode and contract validation
Tool timed outDiagnostic deadline and external dependency

Workspace and checkpoints

SymptomCheck
Response failed after outputPost-turn checkpoint may have failed
Follow-up cannot resumeLatest ready checkpoint and sandbox recreation
Upgrade failedCheckpoint, provisioning, restore or runtime validation phase
New seed file missingSeeds apply only to new runs, not existing durable workspaces

Cancellation and cleanup

A completed response followed by a cancelled or stopped run can represent successful work followed by intentional cleanup. Check response outcomes before treating the final run status as a failure.

Stale browser after deployment

If a save action fails immediately after a web deployment, hard refresh the page. A page loaded from the previous application bundle can submit an outdated server-action identifier.