Summary
When calling openclaw agent with --to but without explicit --agent, routing resolves to main.
With explicit --agent steward --to, routing resolves to steward.
This creates a default/explicit routing mismatch for the same business intent.
Environment
- Observed on:
v2026.3.12 and v2026.3.24
- Default entry agent in config:
steward
Minimal Reproduction
- Verify default agent/bindings:
openclaw agents list --bindings --json
- Default path (
--to without --agent):
openclaw agent --to +15550001111 --session-id probe-default --message test --json
- Explicit path (
--agent steward --to):
openclaw agent --agent steward --to +15550001111 --session-id probe-explicit --message test --json
- Compare:
result.meta.systemPromptReport.sessionKey
- workspace/session metadata
Observed
- default path:
agent:main:main
- explicit path:
agent:steward:main
Expected
Please confirm intended behavior when --to is provided without --agent.
Two acceptable outcomes for callers:
- default
--to honors configured default entry agent semantics, or
- CLI fails loudly and requires explicit
--agent when --to is used.
Why this matters
In multi-agent setups, this mismatch can silently route workloads to main unless every caller enforces explicit --agent.
Context
This is a fresh report after closing previous discussion, because the behavior is still reproducible and blocks downstream close criteria.
Summary
When calling
openclaw agentwith--tobut without explicit--agent, routing resolves tomain.With explicit
--agent steward --to, routing resolves tosteward.This creates a default/explicit routing mismatch for the same business intent.
Environment
v2026.3.12andv2026.3.24stewardMinimal Reproduction
--towithout--agent):openclaw agent --to +15550001111 --session-id probe-default --message test --json--agent steward --to):openclaw agent --agent steward --to +15550001111 --session-id probe-explicit --message test --jsonresult.meta.systemPromptReport.sessionKeyObserved
agent:main:mainagent:steward:mainExpected
Please confirm intended behavior when
--tois provided without--agent.Two acceptable outcomes for callers:
--tohonors configured default entry agent semantics, or--agentwhen--tois used.Why this matters
In multi-agent setups, this mismatch can silently route workloads to
mainunless every caller enforces explicit--agent.Context
This is a fresh report after closing previous discussion, because the behavior is still reproducible and blocks downstream close criteria.