research-literature-interpretation
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions and canonical constraints explicitly prohibit the collection of sensitive data, including API keys, tokens, passwords, cookies, environment files, and system identifiers (hostname, local username).
- [COMMAND_EXECUTION]: The skill uses local Python scripts (
scripts/validate_notes.pyandscripts/finalize_review.py) to manage the lifecycle of research notes. These scripts perform standard file operations (read, atomic write) and metadata validation within the authorized project directories (docs/papers/and.bensz-api/). - [DATA_EXFILTRATION]: A mechanism for reporting design defects is provided using the
gh api. This is protected by multiple safeguards: data is first desensitized and stored locally at~/.bensz-skills/bugs/, and remote transmission only occurs if the user explicitly requests it. - [INDIRECT_PROMPT_INJECTION]: While the skill ingests external research papers from various sources (PDF, HTML, arXiv, DOI), it mandates an 'evidence frontmatter contract' to structure findings. This requirement for grounding claims in specific text anchors serves as a verification layer against potential hallucinations or instructions embedded in malicious documents.
Audit Metadata