Summary

OutputScanner exists as a standalone class (security/output_scanner.py) for post-tool output scanning and redaction. Per D5, it should be integrated into ToolInvoker to scan tool results after execution. This integration is not done.

Design Spec Reference

• §12.3 D5 — add post-tool-call scanning for sensitive data in outputs

Scope

• Wire OutputScanner into ToolInvoker.invoke_all() post-execution path
• Scan ToolResult.content for sensitive data patterns
• Redact or flag findings before returning results to the LLM
• Configurable per autonomy level