nature-article-writer
Warn
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: MEDIUMMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [METADATA_POISONING]: The skill metadata identifies the author as 'OpenAI', which is a deceptive claim intended to misappropriate the trust associated with that organization, as the actual author is 'tristanmanchester'.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted research notes and manuscript drafts as primary inputs but lacks boundary markers or sanitization to prevent the agent from potentially executing instructions hidden within that data. Mandatory Evidence Chain: (1) Ingestion points: Manuscript drafts and research notes processed in SKILL.md; (2) Boundary markers: Absent; (3) Capability inventory: Execution of local Python scripts via shell commands; (4) Sanitization: Absent.
- [COMMAND_EXECUTION]: The SKILL.md instructions direct the agent to execute shell commands to run Python scripts included in the package. While these scripts perform legitimate analysis, execution of code on user-supplied files is a sensitive capability.
Audit Metadata