The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to install the InstaBox CLI by downloading and piping a remote script directly into a shell environment (curl -fsSL https://api.instabox.sh/install.sh | sh). This provides a mechanism for arbitrary, unverified code execution from an external server.\n
Evidence found in SKILL.md, README.md, and references/cli.md.\n- [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by ingesting data from remote machines that could contain malicious instructions designed to override agent behavior.\n
Ingestion points: The agent can read file contents (instabox files), capture command output (instabox exec), and retrieve metadata (instabox get) from the remote instances.\n
Boundary markers: The instructions do not specify the use of delimiters or warnings to treat this external content as untrusted.\n
Capability inventory: The skill provides powerful capabilities including shell command execution (exec), file system writes, and desktop environment control, which could be abused if an indirect injection is successful.\n
Sanitization: No explicit sanitization or filtering of the remote content is mentioned before it is processed by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill relies on fetching and executing external components, including the instabox CLI and the @insforge/instabox-sdk Node.js package, which represent unverified dependencies from third-party sources.\n- [COMMAND_EXECUTION]: The skill's primary function is the execution of commands on remote Linux instances. While remote, this capability requires the agent to handle command construction carefully to avoid command injection vulnerabilities if incorporating untrusted input.

instabox