The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data by instructing the agent to read container logs from remote instances.
Ingestion points: vastai logs INSTANCE_ID (SKILL.md).
Boundary markers: None provided in the instructions to delimit or ignore instructions within logs.
Capability inventory: The skill is allowed to use Bash(vastai:*) and Bash(curl:*) for system interaction.
Sanitization: There is no mention of sanitizing or escaping the content retrieved from instance logs before processing.
[COMMAND_EXECUTION]: The skill uses the vastai CLI to perform various operations, including searching for offers, creating instances, and viewing user account data.
The allowed-tools section explicitly permits Bash(vastai:*) and Bash(curl:*) for these operations.
[DYNAMIC_EXECUTION]: The skill uses a Python one-liner to parse JSON output from the command line, which involves executing code at runtime to process data.
Evidence: vastai show user --raw | python3 -c "import sys,json; d=json.load(sys.stdin); print(f'Balance: \${d[\"balance\"]:.2f}')".
[SAFE]: The instructions recommend StrictHostKeyChecking=no for SSH connections. While this reduces protection against man-in-the-middle attacks, it is a documented standard practice for connecting to ephemeral, short-lived cloud instances provided by services like Vast.ai.

vastai-common-errors