The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the devbox CLI installer from get.namespace.so, which is the vendor's official distribution domain.
[REMOTE_CODE_EXECUTION]: Executes the vendor's installation scripts directly via bash for Linux/macOS and iex for Windows PowerShell environments.
[COMMAND_EXECUTION]: Instructs the agent to generate shell scripts locally and execute them on remote devboxes using the devbox exec command.
[DATA_EXFILTRATION]: Forwards local GitHub authentication tokens to remote devbox instances using the devbox setup-github command to enable repository access.
[PERSISTENCE]: Configures local SSH settings by modifying the ~/.ssh/config file via the devbox configure-ssh command, ensuring persistent connectivity to the devboxes.
[DYNAMIC_EXECUTION]: Dynamically generates and executes task-specific scripts on the remote environment to automate complex workflows.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Untrusted data enters the environment through repository checkouts (--checkout in SKILL.md) and git clone operations in references/devboxes-run-tests.md.
Boundary markers: None present; the skill does not define specific delimiters or warnings to ignore instructions within the ingested repositories.
Capability inventory: Provides extensive capabilities including arbitrary command execution (devbox exec), file system management (devbox upload), and network port exposure (devbox url expose).
Sanitization: No evidence of content sanitization or validation of the repository data before it is processed by the agent.

devboxes