The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents the opencli plugin install github:user/repo command, which allows the agent to download and execute arbitrary code from any GitHub repository.
[COMMAND_EXECUTION]: The opencli external register command enables the definition of custom installation strings, such as npm i -g my-tool, which are executed directly in the shell. This allows for arbitrary command execution under the guise of tool registration.
[EXTERNAL_DOWNLOADS]: The skill instructs users to download and install the primary tool from an untrusted NPM package (@jackwener/opencli) and GitHub repository (github.com/jackwener/OpenCLI).
[DATA_EXFILTRATION]: Several tool strategies (COOKIE, INTERCEPT, UI) and the opencli browser bind command are designed to capture and use sensitive session cookies from the user's browser. This capability allows an agent to access private authenticated sessions and potentially exfiltrate session data.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an interface for extracting data from external websites through opencli browser extract or site-specific adapters. This creates a significant surface for indirect prompt injection, as malicious instructions embedded in a target website's content could influence agent behavior. There are no documented boundary markers or sanitization procedures for this ingested data, while the agent maintains high-impact capabilities like shell execution and cookie access.

opencli-usage