Haiku sub agent failing to spawn reverting to main agent
@Krill π¦
Trying to set this up: anthropic/claude-haiku-4-5 as a subagent named "Kobe" for cheap simple tasks.
Kobe spawns successfully but always runs on moonshot/kimi-k2.5 instead of Anthropic Haiku. He knows his default model is anthropic/claude-haiku-4-5 but executes on Kimi.
Current Config
1. Provider (models.providers.anthropic):
"anthropic": {
"baseUrl": "https://api.anthropic.com",
"api": "anthropic-messages",
"models": [{"id": "claude-haiku-4-5", "name": "Claude Haiku 4.5", ...}]
}2. Alias (agents.defaults.models):
"anthropic/claude-haiku-4-5": {"alias": "Kobe"}3. Agent (agents.list):
{
"id": "kobe",
"name": "Kobe",
"model": "anthropic/claude-haiku-4-5",
"identity": {"name": "Kobe", "emoji": "
"}
}4. Allowlist (main agent subagents):
"allowAgents": ["deepseek", "kobe", "main"]5. Auth (/agents/kobe/agent/auth-profiles.json):
"anthropic:default": {
"type": "api_key",
"provider": "anthropic",
"key": "sk-ant-api03-..."
}βββ
What We've Verified
Model ID claude-haiku-4-5 is valid (appears in openclaw models list)
API key is present and probed successfully (openclaw models status --probe)
Auth profile exists in Kobe's agent directory
baseUrl corrected to https://api.anthropic.com (no /v1)
Agent ID "kobe" is in allowlist
Cleared stale session cache (sessions.json removed)
Gateway restarted multiple times
Security audit shows only WARN (no enforcement)
Session creation shows modelOverride immediately:
{"type":"session",...}
{"type":"model_change","provider":"moonshot","modelId":"kimi-k2.5"}Kobe's response confirms:
"Current model: moonshot/kimi-k2.5"
"Default model: anthropic/claude-haiku-4-5"No errors in logs β auth is working, but model fallback happens before any API call.
1. Why is modelOverride being set to Kimi at session creation time?
2. spawn-time parameter or default subagent model overriding the agent config?
3. model resolution cache or runtime config we haven't found?
version: 2026.2.17
Trying to set this up: anthropic/claude-haiku-4-5 as a subagent named "Kobe" for cheap simple tasks.
Kobe spawns successfully but always runs on moonshot/kimi-k2.5 instead of Anthropic Haiku. He knows his default model is anthropic/claude-haiku-4-5 but executes on Kimi.
Current Config
1. Provider (models.providers.anthropic):
"anthropic": {
"baseUrl": "https://api.anthropic.com",
"api": "anthropic-messages",
"models": [{"id": "claude-haiku-4-5", "name": "Claude Haiku 4.5", ...}]
}2. Alias (agents.defaults.models):
"anthropic/claude-haiku-4-5": {"alias": "Kobe"}3. Agent (agents.list):
{
"id": "kobe",
"name": "Kobe",
"model": "anthropic/claude-haiku-4-5",
"identity": {"name": "Kobe", "emoji": "
"}}4. Allowlist (main agent subagents):
"allowAgents": ["deepseek", "kobe", "main"]5. Auth (/agents/kobe/agent/auth-profiles.json):
"anthropic:default": {
"type": "api_key",
"provider": "anthropic",
"key": "sk-ant-api03-..."
}βββ
What We've Verified
Model ID claude-haiku-4-5 is valid (appears in openclaw models list)
API key is present and probed successfully (openclaw models status --probe)
Auth profile exists in Kobe's agent directory
baseUrl corrected to https://api.anthropic.com (no /v1)
Agent ID "kobe" is in allowlist
Cleared stale session cache (sessions.json removed)
Gateway restarted multiple times
Security audit shows only WARN (no enforcement)
Session creation shows modelOverride immediately:
{"type":"session",...}
{"type":"model_change","provider":"moonshot","modelId":"kimi-k2.5"}Kobe's response confirms:
"Current model: moonshot/kimi-k2.5"
"Default model: anthropic/claude-haiku-4-5"No errors in logs β auth is working, but model fallback happens before any API call.
1. Why is modelOverride being set to Kimi at session creation time?
2. spawn-time parameter or default subagent model overriding the agent config?
3. model resolution cache or runtime config we haven't found?
version: 2026.2.17
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
131,281Members
Resources
Was this page helpful?
