The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines workflows for ingesting and analyzing data from various external sources, which creates a potential attack surface for indirect prompt injection.
Ingestion points: The references/data-science-expert.md file explicitly identifies external APIs and third-party data as primary sources for the data lifecycle.
Boundary markers: There are no instructions or patterns provided to the agent to use boundary markers or delimiters to isolate untrusted data from the system prompt.
Capability inventory: The skill framework encourages the agent to use powerful execution-capable tools including Python, SQL, Spark, and MLOps platforms for building pipelines and automated decision systems.
Sanitization: While the instructions mention data cleaning for quality purposes (e.g., handling missing values or outliers), they do not include security-specific sanitization or validation to filter out potential malicious instructions embedded within the ingested data.

data-science