The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/setup.sh executes remote content by piping a download from https://install.microsandbox.dev directly into the shell (curl -fsSL ... | sh). This pattern allows for the execution of arbitrary, unverified code from an external source.
[EXTERNAL_DOWNLOADS]: The skill downloads the msb CLI and libkrunfw library from the microsandbox.dev domain during setup. It also fetches OCI container images from remote registries during normal operation.
[COMMAND_EXECUTION]: The skill provides extensive capabilities to run arbitrary commands and shell scripts on the host or inside sandboxed environments via msb run, msb exec, and msb shell. This allows the agent to execute any instructions provided to it.
[PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection (Category 8). Ingestion points: Untrusted code or scripts are passed to execution commands in SKILL.md and references/cli-reference.md. Boundary markers: The skill uses microVM hardware isolation as a security boundary. Capability inventory: Full CLI access to sandbox and network management via subprocess calls to msb. Sanitization: The skill does not sanitize or validate input code strings, relying solely on the VM boundary for host protection.

microsandbox