The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses ssh with the -o StrictHostKeyChecking=no option to run diagnostics on remote instances. This prevents the agent from verifying the server's identity, making the connection vulnerable to man-in-the-middle attacks if the network is compromised.\n- [DATA_EXFILTRATION]: The skill prints the first 8 characters of the user's Vast.ai API key during account diagnostics. While not a full exposure, it reveals partial credential information. It also uses curl to connect to cloud.vast.ai, which is a well-known service relevant to the skill's purpose.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the Vast.ai CLI, including instance logs and status messages. A malicious instance or compromised account could return data designed to influence the agent's behavior. Ingestion points: Data from vastai show user, vastai show instances, vastai logs, and vastai show instance. Boundary markers: None identified. Capability inventory: Uses Bash (for vastai, curl, ssh), Grep, and Read. Sanitization: Uses python3 for JSON structure parsing but lacks content filtering for the data within the fields.\n- [DYNAMIC_CONTEXT_INJECTION]: The skill utilizes dynamic context placeholders to execute vastai --version and python3 --version when the skill is loaded. These are benign environmental checks used for troubleshooting the installation.

vastai-debug-bundle