The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts run_eval.py and cleanup.py use subprocess.run to execute sibling Python scripts (ask_local_deepworks.py and ask_teams_agent.py) to perform agent queries and session deletions. These calls use argument lists, which is a secure practice to prevent shell injection. While the tool targets specific internal scripts, this behavior is documented and appropriate for its function.
[INDIRECT_PROMPT_INJECTION]: The tool has an ingestion surface for untrusted data as it processes user-provided question sets and log files from previous runs to drive its evaluation logic.
Ingestion points: run_eval.py reads question files via the --questions argument; cleanup.py reads *.log.json files from the output directory.
Boundary markers: The skill treats input as structured JSON data, though it does not explicitly use delimiters to prevent instruction injection within the data fields.
Capability inventory: The skill performs file system reads/writes in C:/code_proj/tmp/dp-eval-runs and executes local Python scripts via subprocesses.
Sanitization: Relies on standard JSON parsing and subprocess.run argument lists to mitigate common injection vectors.

dp-eval-run