The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/generate_references.py utilizes uvx to run the wenxian package, which is fetched from a public registry. This tool is developed by the skill author.- [COMMAND_EXECUTION]: The skill uses subprocess.run to call external citation tools. By passing arguments as a list rather than a single string, the skill correctly mitigates command injection risks from user-supplied identifiers.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the OpenAlex API and user-specified files, presenting a potential surface for indirect prompt injection within academic metadata.
Ingestion points: Literature search results from api.openalex.org and identifier lists from user-provided files.
Boundary markers: None present; data is passed directly into the agent context as JSON or text.
Capability inventory: Network access via urllib and shell execution of the wenxian tool.
Sanitization: Identifiers are stripped of whitespace and comments, but the academic metadata retrieved via API is not filtered for malicious instructions.

nsfc-literature