train-convo-steering
Warn
Audited by Socket on Mar 17, 2026
1 alert found:
AnomalyAnomalySKILL.md
LOWAnomalyLOW
SKILL.md
SUSPICIOUS. The core behavior matches the stated purpose, but the skill routes sensitive conversation data and API credentials to a configurable external gateway instead of requiring the official DeepSeek endpoint. Cross-session storage of per-user behavioral priors is purpose-aligned yet privacy-sensitive. No malicious installer or overt credential theft is shown, but the intermediary API design and persistent profiling make this higher risk than a normal documentation or runtime helper skill.
Confidence: 86%Severity: 66%
Audit Metadata