failure-taxonomy

Fail

Audited by Snyk on Feb 19, 2026

Risk Level: HIGH
Full Analysis

HIGH W007: Insecure credential handling detected in skill instructions.

  • Insecure credential handling detected (high risk: 0.90). The skill tells the LLM to read "ALL annotations" and to produce a re-labeled dataset that includes the original annotations verbatim (JSON/CSV), so any secrets present in traces would be echoed in the output.
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 19, 2026, 10:36 AM