sessions_spawn Sub-agents lack read/exec permissions ("Tool not found") after main agent is granted
I am experiencing a critical issue with tool permissions after successfully enabling sessions_spawn for my main agent.
1. Main Agent's sessions_spawn Status:
My main agent is now able to successfully call sessions_spawn by omitting the agentId parameter. (It returns {"status": "accepted"}). This indicates the sessions_spawn tool is now functional for the main agent.
2. Sub-agent Tool Access Issue:
However, the sub-agents spawned by the main agent cannot use essential tools like read or exec.
When a sub-agent attempts to read HEARTBEAT.md or execute curl (via exec), it consistently reports: "Tool read not found" or "Tool exec not found".
3. Observed Behavior:
This suggests a system-wide "tool lockdown" where even successfully spawned sub-agents are stripped of basic tool access.
The main agent also reports that read and exec are "not found" when attempting to use them directly.
4. Previous Troubleshooting:
We previously debugged a sessions_spawn "forbidden" error (resolved by adding "sessions_spawn" to tools.agentToAgent.allow in openclaw.json).
The openclaw.json clearly allows sessions_spawn and web.fetch is enabled.
Question:
What is the correct way to grant read, write, exec, and other fundamental tool permissions to all spawned sub-agents (and implicitly, back to the main agent)? Is there a global policy overriding these, or a specific policies configuration needed for sessions_spawn itself that explicitly lists which tools sub-agents inherit or are allowed to use?
1. Main Agent's sessions_spawn Status:
My main agent is now able to successfully call sessions_spawn by omitting the agentId parameter. (It returns {"status": "accepted"}). This indicates the sessions_spawn tool is now functional for the main agent.
2. Sub-agent Tool Access Issue:
However, the sub-agents spawned by the main agent cannot use essential tools like read or exec.
When a sub-agent attempts to read HEARTBEAT.md or execute curl (via exec), it consistently reports: "Tool read not found" or "Tool exec not found".
3. Observed Behavior:
This suggests a system-wide "tool lockdown" where even successfully spawned sub-agents are stripped of basic tool access.
The main agent also reports that read and exec are "not found" when attempting to use them directly.
4. Previous Troubleshooting:
We previously debugged a sessions_spawn "forbidden" error (resolved by adding "sessions_spawn" to tools.agentToAgent.allow in openclaw.json).
The openclaw.json clearly allows sessions_spawn and web.fetch is enabled.
Question:
What is the correct way to grant read, write, exec, and other fundamental tool permissions to all spawned sub-agents (and implicitly, back to the main agent)? Is there a global policy overriding these, or a specific policies configuration needed for sessions_spawn itself that explicitly lists which tools sub-agents inherit or are allowed to use?
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
83,815Members
Resources
Similar Threads
Was this page helpful?
