research-citation-check
Warn
Audited by Socket on Sep 11, 2026
1 alert found:
AnomalyAnomalyscripts/run_ai_alignment.py
LOWAnomalyLOW
scripts/run_ai_alignment.py
This module itself does not show direct malware behaviors (no secrets, no network/exfiltration, no eval/exec or obfuscated payloads). However, it creates a significant supply-chain/remote-code-execution trust boundary by dynamically resolving and executing rendering scripts from a “skill” dependency mechanism. Additionally, it may read/extract text from PDFs even when paths appear outside the specified work_dir (it only warns, not blocks), which can broaden file-access scope depending on helper implementations. Overall: moderate security risk primarily due to dynamic subprocess execution and path-handling delegation to helpers.
Confidence: 62%Severity: 58%
Audit Metadata