The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external telemetry data via the scripts/analyze_trace.py script, which represents a surface for indirect prompt injection.\n
Ingestion points: The scripts/analyze_trace.py script reads external CSV trace files provided via the command line (e.g., TRACE.csv).\n
Boundary markers: The script implements strict schema validation against the FIELDS list and checks the schema_version. It also enforces a 25MB file size limit and a 100,000 row limit to prevent resource exhaustion.\n
Capability inventory: The skill uses the uv run command to execute local Python scripts. The script itself performs statistical analysis and outputs results as JSON, which the agent then interprets to form responses.\n
Sanitization: Data values are type-checked (e.g., converted to floats or ints) and length-restricted to 128 characters, reducing the capacity for complex injection payloads.\n- [COMMAND_EXECUTION]: The skill directs the agent to execute a local utility script (scripts/analyze_trace.py) using the uv tool to perform data analysis.

wireless-telemetry