The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/ask_teams_agent.py executes a local Python script sync_deepworks_iam_token.py using the subprocess.run method. This script is located within the repository path provided by the user via the --repo argument.
Evidence: subprocess.run([sys.executable, str(repo / "scripts" / "sync_deepworks_iam_token.py")], check=False, capture_output=True)
[DYNAMIC_EXECUTION]: The script dynamically modifies the Python module search path (sys.path) to include the src directory of the user-specified repository. This allows the skill to load and execute the agent_eval library from an external directory.
Evidence: sys.path.insert(0, str(repo / "src"))
[DATA_EXPOSURE]: The skill reads the .env file in the provided repository to verify the existence of the AGENT_TEAMS_TOKEN credential, which is required for authentication with the cloud service.
Evidence: if not env_file.is_file() or "AGENT_TEAMS_TOKEN" not in env_file.read_text(encoding="utf-8"):
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text and note content, which is then transmitted to a cloud-based agent. This represents an injection surface if the external agent processes the data without sufficient isolation.
Ingestion points: The --message and --note CLI arguments in scripts/ask_teams_agent.py.
Boundary markers: None present in the script to separate user input from system instructions.
Capability inventory: The skill can execute local subprocesses (subprocess.run) and write files to the local filesystem (answer_path.write_text, log_path.write_text).
Sanitization: No explicit sanitization or escaping of the user-provided input strings was identified.

dp-agent-eval-teams