fix(deepagents): filter invalid content blocks from subagent ToolMessage

[Christian Bromann (christian-bromann)]

Summary

• Filter out tool_use, thinking, and redacted_thinking blocks from subagent response content in returnCommandWithStateUpdate before constructing the ToolMessage. When using Anthropic models, AIMessage.content can be an array containing these block types, which are invalid as ToolMessage content and cause downstream errors.
• Falls back to "Task completed" when filtering removes all content blocks.
• Adds four unit tests covering: tool_use filtering, thinking/redacted_thinking filtering, mixed-invalid-block fallback, and string passthrough.

Fixes #239, fixes #245

[changeset-bot]

⚠️ No Changeset found

Latest commit: 604b64e

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

This PR includes no changesets

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[Anton Nak (antonnak)]

This PR filters tool_use blocks but does not filter thinking or redacted_thinking blocks, which cause the same issue. See #245.

[Christian Bromann (christian-bromann)]

This PR filters tool_use blocks but does not filter thinking or redacted_thinking blocks, which cause the same issue. See #245.

Anton Nak (@antonnak) thanks for the feedback, added more blocks to the filter list.