The Agent Skills Directory

[COMMAND_EXECUTION]: The skill exposes the --doctor mode of ncu, which allows for the execution of custom installation and test commands via the --doctorInstall and --doctorTest flags. These parameters are executed in the shell, enabling arbitrary command execution.
[REMOTE_CODE_EXECUTION]: The tool supports loading configuration from .ncurc.js files, which can contain executable JavaScript logic. An attacker could exploit this by placing a malicious configuration file in a repository that the agent is asked to analyze.
[EXTERNAL_DOWNLOADS]: The --registry flag permits the tool to fetch version information from arbitrary remote URLs, introducing a risk of ingesting untrusted data from third-party sources.
[DATA_EXFILTRATION]: Project dependency names and versions can be sent to external servers by configuring a custom registry URL, which could lead to information leakage from private projects.
[PROMPT_INJECTION]: The skill processes untrusted project files, creating an attack surface for indirect prompt injection. 1. Ingestion points: package.json files, .ncurc configuration files, and --packageData string input. 2. Boundary markers: No delimiters or protective instructions are used to separate file content from agent instructions. 3. Capability inventory: The skill can perform filesystem writes, shell command execution, and network requests. 4. Sanitization: No sanitization or validation of the ingested file content is performed.

npm-check-updates