467 B
467 B
Smoke tests
Required tests:
- simple chat without tools
- task creates execution graph
- weak model response success
- weak model invalid → retry
- weak invalid → strong fallback
- external disabled by config → no external call
- policy confirm → confirmation card
- policy full_auto → execute immediately
- MCP tool result stored
- local worker command result stored
- event stream emits task/node events
- finalizer creates user result