Spawned subagents can write?

[velvet fable]

Sometimes happens that a subagent spawned by parent agent with a read and write task report read findings to parent without write anything. This is supposed ti be the right behaviour or something is missining in procedure?

[versed yew]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[strange stag]

Short version: a spawned subagent can write. It inherits the parent workspace automatically, but OpenClaw does not enforce “do the write before finishing”.

So if a child reads, then reports findings without changing files, that usually means one of these:

• it never actually called write / edit / apply_patch
• those tools were blocked by policy (tools.subagents.tools.*, sandbox tool policy, etc.)
• the workspace was effectively read-only in sandbox
• the prompt let it stop at analysis

The announce behavior is by design: the child’s result gets reported back, even if no write happened.

Good things to check:

• /subagents info <id>
• /subagents log <id> [tools]
• openclaw sandbox explain --session <childSessionKey>