The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the external BRENDA database, specifically fields like commentary and literature. This data is processed and presented to the agent without specific boundary markers, creating a standard surface for indirect prompt injection if the source data were to contain malicious instructions. However, this is an inherent risk of data-retrieval skills and no active exploit was found.
Ingestion points: Data is retrieved via the SOAP API in scripts/brenda_client.py and parsed in scripts/brenda_queries.py.
Boundary markers: None explicitly used for API response strings.
Capability inventory: The skill has file-writing capabilities (pandas.to_csv, plt.savefig, etc.) and network access via the zeep SOAP client.
Sanitization: The skill performs structural parsing (regex/split) but does not sanitize content for potentially malicious natural language instructions.
[COMMAND_EXECUTION]: The skill utilizes Python scripts for data visualization and pathway construction. These scripts are local to the skill and perform intended analytical tasks such as generating plots using matplotlib and reports using networkx. No arbitrary command injection vulnerabilities were identified in the script implementations.

alterlab-brenda