Bug type

Regression (worked before, now fails)

Summary

session_spawn subagent runtime fails with "Model Not Exist" for all non-DeepSeek providers. Only deepseek-chat works without provider prefix.

Steps to reproduce

# Works:
sessions_spawn(task="test", runtime="subagent", model="deepseek-chat")

# All fail:
sessions_spawn(task="test", runtime="subagent", model="ollama:llama3.2:3b")
sessions_spawn(task="test", runtime="subagent", model="gpt4free:gpt-4o")
sessions_spawn(task="test", runtime="subagent", model="oracle-ollama:llama3.2:3b")
sessions_spawn(task="test", runtime="subagent", model="llama3.2:3b")

Expected behavior

All provider:model syntax should work in subagent runtime, enabling specialized teams:

• GPU Teams (Ollama/LM Studio)
• Cloud Teams (Oracle)
• Free API Teams (GPT4Free)
• DeepSeek subagents

Actual behavior

Only deepseek-chat works without provider prefix. All other providers fail with "Model Not Exist" error. Breaks Office Manager architecture with specialized teams.

OpenClaw version

v2026.3.2

Operating system

Ubuntu/Linux

Install method

npm global

Logs, screenshots, and evidence

Impact and severity

Affected: All users with multi-provider setups (GPU, cloud, free APIs)
Severity: High (breaks multi-provider delegation)
Frequency: 100% reproducible
Consequence: Office Manager architecture broken - cannot delegate to specialized teams

Additional information

Workaround: Use Standard Runtime instead of Subagent Runtime:

openclaw run --agent [team-id] --task-file ./tasks/job.txt

Suspected Cause: Regression in v2026.3.1 subagent system changes. Provider prefix parsing broken in isolated runtime.

Labels: bug, regression, subagent, models