{"version":"2.1.318","anchor":"neutralizer-now-also-applies-a-control-tag-pass-sanitizer-r","canonical_anchor":"neutralizer-now-also-applies-a-control-tag-pass-sanitizer-r","heading":"Text cleaning before the model sees input uses a changed rule set","tier":"internal","area":"Prompt Safety","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.318\/e\/neutralizer-now-also-applies-a-control-tag-pass-sanitizer-r","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.318","markdown":"### Text cleaning before the model sees input uses a changed rule set\n\nClaude Code's text cleaner now draws rules from three sources, adds a pass for control tags, and fixes replacements to apply to the text being cleaned\n\n**Unclear.** What the rule sources contain and what the extra control-tag step removes is not clear.\n\n**What**\n\nClaude Code cleans some text before the model sees it, removing or changing parts that could be mistaken for instructions. That cleaner has changed:\n\n- Its rules now come from three sources, and one source is skipped in some cases\n\n- When something it finds is a control tag, it runs an extra step\n\n- Replacements now apply to the text being cleaned, fixing earlier code that matched against the wrong text\n\n**Why**\n\nThis can change what text is removed from what the model reads.\n\n- Area: Prompt Safety\n- Tier: Under the hood\n- Useful: 1\/5\n- Signal: 1\/5\n- Scope: individual\n- Heads-up: no"}