The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes npx -y nvd-cve-mcp-server which downloads and runs a package from the npm registry at runtime. This pattern executes remote code without version pinning or source verification, posing a significant risk if the package is compromised.
[COMMAND_EXECUTION]: The audit procedure relies on multiple shell command executions via python3 $MCP_CALL. These commands interface with local scripts and system tools like npx to perform security checks and log results.
[DYNAMIC_EXECUTION]: The skill uses environment variables such as $PYATS_MCP_SCRIPT, $ISE_MCP_SCRIPT, and $GAIT_MCP_SCRIPT to determine the paths of scripts to be executed. Loading and executing code from paths computed at runtime is a risk factor as it could be redirected to malicious files if the environment is compromised.
[EXTERNAL_DOWNLOADS]: The use of npx results in the download of external software components from the public npm registry during the execution of the CVE vulnerability scan.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill captures and processes the full running configuration of network devices via the pyats_show_running_config function in SKILL.md.
Boundary markers: Absent; there are no explicit delimiters or instructions provided to the agent to treat the device configuration as untrusted data or to ignore instructions embedded within it.
Capability inventory: The skill possesses extensive command execution capabilities (via $MCP_CALL) and network access (via npx and ISE integration) across its entire lifecycle.
Sanitization: Absent; the agent is instructed to scan the configuration and use the extracted version strings to perform further searches and record findings in GAIT without sanitizing the input data.

pyats-security