What
Claude Code runs some small background requests that should only produce text. A tool call, where the model asks to run one of its tools, now ends both of them without a result:
- Agent summaries: when Claude Code writes a short progress summary of a subagent (a helper Claude starts to work on part of a task), the request is stopped as soon as the model starts a tool call or anything other than text. No summary is produced, and the skip is recorded with the reason
tool_call. Before, the request returned whatever the model sent. - Prompt suggestions: the request that suggests what you might type next now reports whether it called a tool. If it did, the suggestion is dropped and the outcome is recorded as
tool_call. Before, the only recorded outcomes were aborted, empty and unused.
Why
In practice you may sometimes see no subagent summary or no suggestion, rather than one built from a request that went off course into a tool call.