The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that ingests data from the Vast.ai REST API (via the vastai CLI tool) and uses it to drive automated actions such as instance creation and recovery.
Ingestion points: Data enters the execution flow via subprocess.run(["vastai", "show", "instances", "--raw"]) and subprocess.run(["vastai", "search", "offers", ...]) in SKILL.md.
Boundary markers: The skill does not implement explicit delimiters or instructions to ignore potential natural language instructions that could be embedded in externally-controlled metadata (such as GPU names, instance tags, or image IDs) returned by the API.
Capability inventory: The skill leverages subprocess.run to execute CLI commands and is granted Write and Edit permissions in its configuration.
Sanitization: While the Python code uses json.loads to parse API responses (which protects against basic command injection), it does not include logic to filter or sanitize the content for potential downstream interpretation by an AI agent that might be monitoring the script's output or logs.

vastai-webhooks-events