The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py executes the curl binary via subprocess.run to communicate with external MDM APIs. While the arguments are passed as a list to prevent basic shell injection, the reliance on external system binaries and the handling of sensitive credentials (tokens) via command-line arguments are notable behaviors. Evidence: cmd = ["curl", "-s", "-X", method, ...] in scripts/agent.py.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests data from external MDM/UEM API endpoints. This data, which includes device names, app inventory, and policy settings, is untrusted and could be crafted to include malicious instructions. 1. Ingestion points: API responses in scripts/agent.py via get_managed_devices, get_managed_apps, and other policy retrieval functions. 2. Boundary markers: None present in the processed JSON output or report generation to distinguish untrusted data from instructions. 3. Capability inventory: Shell command execution (subprocess.run in scripts/agent.py) and local file system write access in both scripts/agent.py and scripts/process.py. 4. Sanitization: No sanitization, validation, or escaping of the ingested strings is performed before they are included in the final audit reports.

implementing-mobile-application-management