The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py uses subprocess.run to execute the system curl command for interacting with the Mimecast API.
Evidence: The mimecast_request function in scripts/agent.py constructs a command list starting with curl and executes it using subprocess.run(cmd, ...).
Context: While the command is structured as a list to mitigate shell injection, invoking external binaries for network operations is a high-risk pattern compared to using native Python libraries like requests.
[PROMPT_INJECTION]: The skill processes untrusted external data (email headers and logs) which constitutes an indirect prompt injection surface.
Ingestion points: scripts/agent.py retrieves log data containing sender addresses, recipient addresses, and display names from the Mimecast API.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands when the script outputs these strings to the console or the report file.
Capability inventory: The script has the capability to execute commands via subprocess.run and write files to the local disk.
Sanitization: No sanitization or escaping is performed on the data fetched from the external API before it is processed or displayed.
[SAFE]: The policy auditing logic in scripts/process.py is implemented using safe data processing techniques and does not perform any network or dangerous system operations.
[SAFE]: Documentation and reference files provide legitimate configuration guidance for Mimecast services and contain no malicious code or instructions.

implementing-mimecast-targeted-attack-protection