cml-packet-capture
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill operates by invoking the
cml-mcpcommand-line utility (Model Context Protocol server) to communicate with the CML environment. - [EXTERNAL_DOWNLOADS]: The skill documentation specifies a dependency on the
cml-mcppackage, which is intended to be installed via the Python package manager (pip). - [INDIRECT_PROMPT_INJECTION]: The skill captures and analyzes network traffic (PCAP files), which constitutes untrusted external data. This presents a potential surface for indirect prompt injection if the captured packets contain adversarial text instructions processed by the agent during analysis.
- Ingestion points: Raw network traffic captured from CML links and subsequent analysis of protocol hierarchies and conversation data via the Packet Buddy integration.
- Boundary markers: The skill does not explicitly define delimiters or instructions to ignore embedded content within the packet data.
- Capability inventory: The skill has the ability to write files to the local system (
download_capture), access sensitive environment variables (CML_PASSWORD), and execute shell commands through the MCP server. - Sanitization: There is no evidence of content sanitization or instruction filtering applied to the captured packet data before it is presented to the agent for interpretation.
Audit Metadata