The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret external CSV and JSON data produced by validator tools. This creates an attack surface where malicious content within those files could attempt to influence the agent's logic or downstream actions. Evidence: The skill reads issues.csv and summary.json, processing rule names, messages, and suggestions into an agent-facing report.
[DYNAMIC_EXECUTION]: The skill contains instructions to generate a temporary Python script (_summarize_validator_csv.py) on the fly if a pre-existing summarizer tool is not found. While intended as a fallback for data processing, the generation of executable code at runtime is a risk factor. Evidence: SKILL.md specifies building a 'temporary stdlib-only fallback summarizer beside the artifact' and executing it to parse data.
[COMMAND_EXECUTION]: The skill frequently invokes shell commands to run local Python scripts and CLI utilities. Evidence: The skill uses the Bash tool to execute python3 tools/validators/resolve_artifacts.py and python3 tools/validators/summarize_csv.py to handle artifact resolution and summarization.
[DATA_EXPOSURE]: The skill accesses local file system paths to retrieve validation artifacts, including potentially sensitive primitive paths and metadata within the USD assets. Evidence: It reads file paths provided as arguments and looks for sibling .csv and .json files in the artifact directory.

interpret-validators