self-improve
Warn
Audited by Socket on Nov 2, 2026
2 alerts found:
AnomalySecurityAnomalySKILL.md
LOWAnomalyLOW
SKILL.md
The workflow is broadly consistent with its stated purpose and uses GitHub for its intended collaboration tasks. However, it can publish transcript-derived material and project/session identifiers without per-action approval or an explicit sensitive-data review. Treat as elevated risk; review and redact evidence before publication.
Confidence: 91%Severity: 68%
Securityevals/files/covered-and-duplicate/project/.claude/skills/self-improve/SKILL.md
MEDIUMSecurityMEDIUM
evals/files/covered-and-duplicate/project/.claude/skills/self-improve/SKILL.md
SUSPICIOUS. The skill is coherent with its stated purpose and uses mainly official GitHub tooling, but it grants an agent autonomy to create Issues/PRs without asking approval and to upload locally derived session evidence to GitHub. That makes it a high workflow-risk skill despite limited malware indicators and relatively normal supply-chain posture.
Confidence: 85%Severity: 74%
Audit Metadata