The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides Python templates for the agent to generate and execute scripts via a tool to process large YAML files.
Evidence: Step 1 contains a Python block using import yaml and yaml.safe_load(FILE_CONTENT) for systematic analysis of snapshots.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied AICR snapshot YAML files which may contain instructions or malicious data in metadata or label fields.
Ingestion points: The skill parses metadata, labels, and measurements from large YAML files (up to 80K tokens) provided by the user.
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore instructions embedded inside the snapshot data when generating the cluster assessment report.
Capability inventory: The agent has access to mcp__plugin_context-mode_context-mode__execute_file (Python execution), Grep, and Read tools.
Sanitization: The skill uses yaml.safe_load for secure deserialization, but lacks semantic sanitization or escaping of the values extracted from the YAML before processing them or including them in reports.
[COMMAND_EXECUTION]: The skill provides a template for mapping snapshot data to an aicr recipe CLI command using extracted data.
Evidence: The "AICR Criteria Mapping" section demonstrates a bash command template where extracted fields such as detected_service, detected_accelerator, and detected_os are interpolated. If an agent executes this command without validating the source data, it creates a potential command injection surface.

aicr-analyzing-snapshots