openclaw/Kova
openclaw/Kova
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
main
Kova/docs/SCENARIO_HIERARCHY.md
Shakker 7bb812304e
docs: document scenario hierarchy
2026-05-07 12:23:56 +01:00

1.6 KiB
Raw Permalink Blame History

Kova Scenario Hierarchy

Kova measures OpenClaw behavior by runtime path, not by whichever client first exposed that path.

Gateway Session

gateway-session-send-turn is the shared Gateway session messaging benchmark. It calls sessions.create, sessions.send, and chat.history directly over Gateway RPC with mock auth/provider. Use it for the core Control UI, dashboard, and channel message path.

This scenario owns:

  • cold and warm Gateway session turn timing
  • pre-provider, provider, and post-provider timing
  • Gateway session diagnostics timeline attribution
  • Gateway process RSS/CPU as the primary product resource signal
  • helper process cost under gateway-session-client

Dashboard

dashboard-readiness is dashboard-specific coverage. It verifies the dashboard command, URL/websocket entry, and post-dashboard Gateway health. It does not own the shared message-turn benchmark.

Channels

Channel scenarios should prove adapter behavior: the channel can create or route a user turn into the shared Gateway session path and recover/report adapter-specific failures. They should compare against gateway-session-send-turn instead of duplicating the core Gateway session benchmark.

Agent CLI

Agent CLI scenarios measure CLI client behavior separately:

  • agent-cli-local-turn covers short-lived local agent CLI turns.
  • agent-gateway-rpc-turn covers openclaw agent as a CLI client crossing the Gateway agent RPC path.

CLI process cost should stay on CLI roles. It should not be used as the primary acceptance signal for Control UI, dashboard, or channel users.