The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the box_exec tool to run arbitrary shell commands inside the remote Linux container. This is the core functionality intended for building, testing, and running applications in a sandboxed environment.
[REMOTE_CODE_EXECUTION]: The skill facilitates remote code execution within the Upstash Box environment. It provides instructions for running long-lived processes, installing software, and executing complex scripts for video processing using ffmpeg.
[EXTERNAL_DOWNLOADS]: The instructions suggest downloading external tools and packages, such as the ttyd static binary from GitHub releases and various software via package managers like npm, pnpm, and apt-get within the remote box.
[PRIVILEGE_ESCALATION]: The skill describes the use of sudo within the remote container to perform administrative tasks like installing system-level dependencies (apt-get install) and global package management. These operations are confined to the isolated remote box.
[DATA_EXFILTRATION]: The skill includes tools for uploading files from the box to Upstash 'Blob' storage. It describes workflows for making these files publicly accessible via preview URLs or public buckets, which is the intended mechanism for delivering screenshots and videos.
[INDIRECT_PROMPT_INJECTION]: The agent interacts with external, potentially untrusted content through box_git (cloning repositories) and box_browser (visiting websites).
Ingestion points: External code repositories and web page content entered into the agent's context.
Boundary markers: No delimiters or safety warnings for external content are specified in the instructions.
Capability inventory: The skill provides full shell execution (box_exec), file system access, and network operations (blob_upload_url).
Sanitization: No specific sanitization or validation of external content is described before it is processed by the agent.

upstash-box-remote-work