Review findings now reported in 'spawn' or 'message' mode, chosen by a new check

What

When Claude Code assembles the prompt it uses for reviewing code, it now picks one of two ways to report findings (the problems or observations the review turns up). The previous check is gone. That check only looked at whether the CLAUDE_CODE_REPORT_FINDINGS environment variable was set and whether a findings tool was in the list of available tools. An environment variable is a setting passed to a program from outside it.

• spawn mode is used whenever CLAUDE_CODE_REPORT_FINDINGS is set.
• message mode is used otherwise, but only when all of these hold: the client is desktop_app or web_claude_ai, the session was started from a remote entrypoint, and the tengu_purring_micali gate is true. A gate is a switch Anthropic can turn on or off remotely. Nothing has been read about this gate in this release, so whether it is on is unknown.
• In message mode the prompt gains an extra section asking for one line per finding, each with its outcome.
• message mode is skipped when the output format is text or json.
• The prompt's uses_report_findings_tool value is now worked out from this new mode choice instead of the old check, and some parts of the review prompt were moved around.

Why

In practice, the way review findings are presented can now differ between the terminal and the desktop app or claude.ai on the web. In those apps, findings may arrive as a short one-line-per-finding message. Anyone who relies on CLAUDE_CODE_REPORT_FINDINGS still gets the spawn behaviour. Because the review prompt itself changed, the wording and layout of review output may also shift.