4 points | by grumblemumble 5 hours ago ago
2 comments
A teardown of Claude Code's auto-mode safety classifier, looking at the undocumented ruleset that interprets user consent.
Author here, I ran a MITM dump on Claude Code sessions to see what actually gets sent over the wire and what makes the safety classifier tick.
A teardown of Claude Code's auto-mode safety classifier, looking at the undocumented ruleset that interprets user consent.
Author here, I ran a MITM dump on Claude Code sessions to see what actually gets sent over the wire and what makes the safety classifier tick.