The tool-JSON double-escaped-unicode repair now skips control-character \u escapes too, and tracks them in telemetry as controlEscapeSkips
What
Claude Code has a routine that repairs double-escaped unicode characters in tool call JSON. It now additionally detects \u escapes representing control characters (the C0 and C1 ranges) and leaves those unrepaired, on top of the existing behavior of skipping Windows file paths.
This is tracked with a new counter, controlEscapeSkips, reported alongside the existing counts for repaired strings and Windows-path skips in the same telemetry event.
Why
Control characters encoded as \u escapes shouldn't be touched by the double-escape repair logic, since "fixing" them could corrupt otherwise-valid data; tracking how often this happens helps monitor the repair pass's behavior.