alterlab-citation-verifier
Pass
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/claim_faithfulness.pyscript executes theclaudeCLI viasubprocess.runto perform its LLM-judge assessment. This is a core component of the skill's functionality used to evaluate if a retrieved abstract supports a specific user claim. - [EXTERNAL_DOWNLOADS]: The skill performs automated network requests to authoritative scholarly databases, including api.crossref.org, api.openalex.org, api.semanticscholar.org, and export.arxiv.org. These requests retrieve metadata and abstracts and include a contact email in the User-Agent as required for polite pool access.
- [PROMPT_INJECTION]: The skill ingests untrusted data from user-supplied bibliographies and external API content, which is then interpolated into LLM prompts within the
scripts/claim_faithfulness.pyscript. The skill mitigates this risk by extracting specific metadata fields (e.g., abstracts) rather than processing arbitrary unstructured text. The ingestion points include BibTeX and free-form references, while its capabilities are restricted to network retrieval and platform CLI invocation.
Audit Metadata