The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external profiling data from multiple sources and interpolates those values into its analysis results and generated reports.
Ingestion points: The skill reads profiling data from several files including kernel_details.csv, trace_view.json, op_summary, communication.json, and operator_details.csv as documented in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore embedded instructions within the processed data fields.
Capability inventory: The skill is designed to perform file system write operations to save reverse-engineered model architecture reports (ARCH_REPORT_SAVE phase in SKILL.md).
Sanitization: The skill lacks explicit sanitization or validation of data extracted from the profiling traces (such as kernel names or task descriptions) before including them in the final markdown output.

ascend-profiling-anomaly