citation-management
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFE
Full Analysis
- [DYNAMIC_EXECUTION]: Several scripts (
extract_metadata.py,format_bibtex.py,search_google_scholar.py,search_openalex.py,search_pubmed.py,validate_citations.py) contain__import__('pathlib')calls used to dynamically resolve the script's parent directory for local module imports. This usage is strictly limited to path resolution and does not involve executing untrusted code or dynamic module loading from external sources. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data (e.g., paper titles and abstracts) which is a potential vector for indirect prompt injection. However, the risk is effectively mitigated through several layers:
- Sanitization: The shared utility
_common.pyprovides asanitize_keyfunction that uses regex and Unicode normalization to ensure citation keys are restricted to safe alphanumeric characters, preventing command or path injection. - Guidance:
SKILL.mdexplicitly instructs the agent to treat metadata as untrusted and provides examples of safesubprocessusage to avoid shell injection vulnerabilities. - Limited Capability: While the skill utilizes API data, it does not provide an automated path for this data to influence security-critical operations without the agent following the provided safety guidelines.
- [SAFE]: The skill's architecture is consistent with its stated purpose. Credentials for academic APIs (NCBI, OpenAlex) are managed via environment variables and transmitted only to the intended official domains over HTTPS. No malicious persistence, privilege escalation, or data exfiltration patterns were observed.
Audit Metadata