Problem

Visible OpenClaw session output can replace tool-result text with an attachment placeholder even when the underlying tool result is structured non-image content.

Impact

This makes local tool verification unreliable because ordinary tool output can disappear from the visible transcript and be replaced by a placeholder.

Suspected source

extractToolResultText in src/agents/embedded-agent-subscribe.tools.ts originally only collected explicit text content blocks. Structured non-image blocks like JSON/resource content therefore produced no visible text in the subscription/display path.

Expected

• Structured non-image tool-result blocks should be serialized into safe visible text.
• True image-only blocks should remain skipped or handled by media delivery.

Local fix

• Add a narrow structured-block fallback inside extractToolResultText.
• Skip text and image blocks in the fallback.
• Add a focused unit test covering structured non-image output and mixed text/image output.

Local evidence

• git diff --check
• node scripts/run-vitest.mjs src/agents/embedded-agent-subscribe.tools.test.ts

Resolution status

• Canonical fix candidate: fix(agents): preserve structured tool result visible text #97268 (fix(agents): preserve structured tool result visible text).
• fix(agents): preserve structured tool result visible text #97268 now uses Fixes #97267 so this issue should close automatically when the fix merges.
• Current local installed OpenClaw verification shows the structured-only fallback, mixed text-first behavior, image-only suppression, and inline data-URI redaction all pass.
• Sibling same-root candidate fix(agents): serialize structured non-image tool-result blocks in extractToolResultText (fixes #97267) #97276 exists; maintainers should land one helper implementation, not both.