Desktop app renders ReportFindings (code-review results) as raw JSON instead of the findings panel

Bug

In the Claude Desktop app, when a /code-review session reports its results via the ReportFindings tool, the findings are not rendered as the formatted findings panel. The transcript just shows the tool call ("Used ReportFindings") with the raw JSON payload as plain text, which is effectively unreadable — each finding's file/line/summary/failure_scenario/verdict appears as an escaped JSON blob.

The user reports this happens in nearly all code-review sessions in the desktop app, so it looks like the desktop client is missing (or failing to trigger) the renderer for ReportFindings tool calls rather than a one-off glitch.

Steps to reproduce

1. Open Claude Code in the Claude Desktop app (macOS).
2. Ask for a code review (e.g. "do a code review on the last 1 commit on current branch"), which invokes the code-review skill.
3. The review completes and calls ReportFindings with a findings array.
4. Observe the transcript: the tool call shows raw JSON in plaintext instead of a rendered findings list.

Expected behavior

ReportFindings results render as the typed findings UI (per the tool's own description: "Report code-review findings as a typed list so the host UI can render them"), with clickable file:line references, summaries, failure scenarios, and verdicts.

Actual behavior

Plain-text JSON dump of the tool input. The model has to re-print the entire report as markdown for the user to be able to read it (which also contradicts the tool guidance "do not also print the findings as text").

Environment

• Client: Claude Desktop app (macOS)
• OS: macOS (Darwin 25.2.0)
• Model: claude-fable-5
• Review invoked via the code-review skill; findings reported with a single ReportFindings call (10 findings, level: "high")

🤖 Filed with Claude Code on the user's behalf