paper-spine-citation
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted research data from 'research_dossier.md', 'sota_gap_map.md', 'source_index.md', and user-provided PDFs (ingestion points). No boundary markers or 'ignore' instructions are present to delimit this content. The skill possesses capabilities to write files and execute shell commands for its local validation scripts. No sanitization or filtering of the external content is performed, creating a surface for indirect prompt injection.
- [COMMAND_EXECUTION]: The skill invokes 'citation_bank_check.py' and 'citation_quality_audit.py' using shell commands. These are internal scripts provided with the skill.
- [EXTERNAL_DOWNLOADS]: The verification scripts perform network requests to 'api.crossref.org' (well-known service) to resolve DOI information. No sensitive data is exfiltrated during these requests.
Audit Metadata