folded-memory-implementation

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The prompts ask the model to read raw conversation turns and tool outputs and to produce structured episode/semantic summaries (including "tool_results") that are then injected into system prompts and persisted, but provide no explicit redaction or secret-handling rules—so any secrets present in turns or tool outputs could be read and reproduced verbatim and stored, creating an exfiltration risk.