The Agent Skills Directory

External Resource Downloads: The skill fetches the Apache Spark source code from its official GitHub repository and installs the pyspark package from the standard Python Package Index (PyPI). These operations involve trusted organizations and well-known services.
Local Command Execution: To perform the audit, the skill utilizes standard development tools such as git, grep, find, and cargo. These commands are used to analyze the codebase and run verification tests as part of the intended auditing process.
Dynamic Script Generation: The skill generates a temporary Python helper script (spark_ground_truth.py) to run queries against a local Spark session for establishing ground truth. This is a common pattern for setting up controlled test environments.
Indirect Prompt Injection Surface: The skill ingests external data from Spark source code and GitHub API responses. While this represents a surface where untrusted data enters the agent context, the context of use is technical analysis from trusted sources.
Ingestion points: Spark source code retrieved via git clone and repository content via gh api.
Boundary markers: Not explicitly utilized for ingested code analysis.
Capability inventory: Execution of shell commands (grep, find, etc.), Python scripts, and cargo tests.
Sanitization: No specific sanitization of ingested source code strings before analysis.

audit-datafusion-spark-expression