deep-research
Warn
Audited by Socket on May 14, 2026
1 alert found:
AnomalyAnomalySKILL.md
LOWAnomalyLOW
SKILL.md
SUSPICIOUS. The skill’s capabilities mostly fit its research purpose, but it combines untrusted web ingestion with Bash and Write permissions, creating meaningful indirect prompt-injection risk. Its use of external MCP memory/pattern stores is proportionate to the goal yet expands data exposure beyond a purely local research assistant.
Confidence: 84%Severity: 66%
Audit Metadata