The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to clone a repository from an untrusted source (https://github.com/noredistribution/mcp-cvp-fun) that is not affiliated with the skill author ('automateyournetwork') or any trusted organization.
[REMOTE_CODE_EXECUTION]: The entry point command uv run --with fastmcp fastmcp run /path/to/mcp_server_rest.py executes a script downloaded from the untrusted repository at runtime.
[COMMAND_EXECUTION]: The skill relies on executing shell commands (git clone, uv run) to install and start the MCP server, providing a path for arbitrary code execution if the remote repository is compromised.
[CREDENTIALS_UNSAFE]: The skill requires a sensitive service account token (CVPTOKEN) for authentication to Arista CloudVision Portal. The documentation acknowledges that 'TLS verification is disabled in source,' which exposes this token to interception via Man-in-the-Middle (MitM) attacks.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection.
Ingestion points: The tools get_inventory, get_events, and get_connectivity_monitor ingest data from the external CloudVision Portal (CVP).
Boundary markers: No boundary markers or 'ignore embedded instructions' warnings are present to prevent the agent from obeying instructions embedded in device names or event descriptions.
Capability inventory: The skill possesses write capabilities through the create_tag tool, which performs a multi-step POST request workflow to the CVP API.
Sanitization: There is no evidence of sanitization or validation of the data retrieved from CVP before it is processed by the agent.

arista-cvp