The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of local Python scripts (compute_labels.py, assemble_widget_proto.py, and validate_chart.py) to process metric metadata and generate chart configuration files. These scripts use standard Python libraries and perform deterministic operations.
[DATA_EXPOSURE]: The validate_chart.py script provides a mechanism to read files from the workspace via the --input_file argument. While this grants the agent read access to files, the script is intended for self-verification of the skill's own generated outputs and contains logic to validate the content against expected textproto schemas.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of PromQL queries and metric display names. These inputs are interpolated into the final chart's titles and labels.
Ingestion points: promql_query and spec_json arguments in assemble_widget_proto.py and compute_labels.py.
Boundary markers: No specific boundary markers are used for the input data.
Capability inventory: The skill can write to the local filesystem (assemble_widget_proto.py) and read from it (validate_chart.py).
Sanitization: The assemble_widget_proto.py script uses a format_proto_string function to escape backslashes and double quotes in titles and queries, preventing breakage of the resulting textproto format.
[SAFE]: The skill includes 'CRITICAL EXECUTION & WORKING DIRECTORY RULES' in its documentation, which explicitly instruct the agent not to use discovery or search tools like grep or find to explore the repository. This reduces the risk of unintended data exposure through agent curiosity.

cloud-monitoring-chart-generation