adversarial-review
Warn
Audited by Socket on Apr 15, 2026
1 alert found:
AnomalyAnomalySKILL.md
LOWAnomalyLOW
SKILL.md
SUSPICIOUS: the skill’s purpose is coherent, but it routes local code/diffs and guidance files through an external opposite-model CLI, creating meaningful confidentiality and prompt-injection risk. Install trust for `codex exec` appears acceptable from the provided evidence, so this is not confirmed malicious; the main issue is proportional data exposure and reduced transparency from shell-based background execution.
Confidence: 82%Severity: 58%
Audit Metadata