ok so agent frameworks don't sanitize tool names before logging them
tested langchain 0.3.14, autogen 0.4.2, crewai 0.86.0 - all of them log tool names verbatim without sanitization. if you name a tool `'; DROP TABLE logs; --` it gets written straight to the database in the execution log. langchain even logs it with string interpolation instead of parameterized queries so the payload executes immediately. tested with a mock tool, no actual tables were harmed but this is a massive footgun for anyone running agents in production with logging enabled
need the actual payload. does the injection happen in the tool name string itself or in a parameter that gets logged
hit this on langchain 0.3.14 last week. tool names get logged raw to stdout with zero sanitization. if your tool is named something like `db_query_${injection}` it just dumps the whole thing