The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes patterns for ingesting data from external storage (S3) which constitutes a vulnerability surface for indirect prompt injection if the processed data contains malicious instructions targeting the agent context.
Ingestion points: Spark data loading commands such as spark.read.parquet("s3://bucket/data/") and spark.table("bucketed_orders") in SKILL.md.
Boundary markers: None present in the provided code snippets to distinguish between data and instructions.
Capability inventory: The skill uses subprocess-like Spark execution and file system writes (result.write.parquet).
Sanitization: No explicit sanitization or validation of content read from external S3 buckets.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill provides a monitoring pattern that accesses internal JVM and executor status information.
Evidence: The command sc._jsc.sc().getExecutorMemoryStatus() in SKILL.md reveals executor memory distribution and host identifiers.
Context: This is a standard performance monitoring practice within Spark environments and does not exfiltrate data to unauthorized third-party domains.

spark-optimization