The leaked codebase contains references to fake tools that simulate capabilities without executing external actions. This is documented directly in the source publication (https://alex000kim.com/posts/2026-03-31-claude-code-source-leak/).

Frustration regexes are present in the code to pattern-match user dissatisfaction in dialogue. The same primary source lists these regex patterns as part of the model's response logic (https://alex000kim.com/posts/2026-03-31-claude-code-source-leak/).

Undercover modes appear in the leaked files as hidden behavioral states for the model. Coverage is limited to the original article and associated discussion thread (https://alex000kim.com/posts/2026-03-31-claude-code-source-leak/; https://news.ycombinator.com/item?id=47586778).