nsfc-proposal
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The installation instructions in
SKILL.mdadvise users to download a shell script from the author's GitHub repository and pipe it directly into a bash shell (curl -fsSL ... | bash). This is a common but high-risk delivery method for executable content. - [EXTERNAL_DOWNLOADS]: The skill requires the installation of a dependency, the
qinyan-paper-searchskill, to perform its literature search functions. It fetches resources from the author's GitHub repository (LeonChaoX/qinyan-academic-skills) and interacts with external academic databases. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user-provided data, including research topics, scientific problems, and literature content.
- Ingestion points: User-supplied values for keywords, research problems, and existing research foundations in
SKILL.md(Phase 1). - Boundary markers: The instructions lack explicit delimiters or instructions for the agent to ignore potential malicious prompts embedded in the processed research data.
- Capability inventory: The agent can perform literature searches and generate markdown content based on the input.
- Sanitization: There is no evidence of sanitization or validation of the input data before it is interpolated into writing tasks.
Audit Metadata