Unclear It is not clear which MCP results go through the picture step or what it changes about them.
What
A permission request is the question Claude Code asks before it uses a tool. These requests now include an agent_id, taken from askerAgentId. It identifies the subagent (a helper agent Claude starts for part of a task) that made the request.
- Tool permission requests now include
agent_id. - One loop that goes through pending permission requests now skips any request with a non-empty
agent_id, so requests from subagents are not handled there. - Separately, non-image results from MCP servers (outside services that give Claude extra tools) that are saved to a file now pass through
readPicture.
Why
Programs that show or answer permission prompts from outside the terminal, such as remote approval screens, can now tell which subagent is asking. This may also change which pending prompts are shown or replayed.
Names in the bundleagent_id
The entry above is what we published on the day. These lines were added later, as Anthropic's own pages caught up, and they sit beside the original rather than replacing it.
Confirmed since
Anthropic's documentation has since written up agent_id, on Agent SDK reference - TypeScript.
`agent_id` identifies the subagent that produced the message and is absent on main-thread messages. The value equals the `task_id` on that subagent's [`task_started`](#sdktaskstartedmessage) and other task events, and is unchanged when the…agent-sdk/typescript see the edit
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
It is not clear which MCP results go through the picture step or what it changes about them.
Anthropic's documentation agrees
Anthropic's documentation has since written up agent_id, on Agent SDK reference - TypeScript.