context-decay
Warn
Audited by Socket on Oct 19, 2026
1 alert found:
SecuritySecurityevals/fixtures/approval-blocked-heal/installed/export-data/SKILL.md
MEDIUMSecurityMEDIUM
evals/fixtures/approval-blocked-heal/installed/export-data/SKILL.md
SUSPICIOUS. The skill’s purpose is plausible, but its actual trust chain is not: it asks the agent to run an undocumented local tool and to auto-update the skill itself with no trusted source or verification path. That mismatch makes the footprint higher-risk than a normal data-export guide, even without direct evidence of credential theft or external exfiltration.
Confidence: 85%Severity: 74%
Audit Metadata