agents-skills-feedback-loop
Warn
Audited by Socket on Oct 17, 2026
1 alert found:
SecuritySecuritySKILL.md
MEDIUMSecurityMEDIUM
SKILL.md
SUSPICIOUS: the core learnings-file workflow is coherent and mostly local, but the optional capture hook expands scope materially by installing a machine-global persistent hook that reads session transcripts and forwards trimmed transcript content to the Claude CLI for reflection. That data flow is relevant to the feature, yet proportionately riskier than the base skill description and should be treated as medium/high operational risk rather than benign documentation tooling.
Confidence: 90%Severity: 72%
Audit Metadata