The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill utilizes a piped shell execution pattern (curl | bash) to install the Google Cloud SDK. While the installation source is a well-known service provider, this pattern executes remote scripts directly without local verification.\n- [COMMAND_EXECUTION]: Provides instructions for executing arbitrary scripts on remote Azure virtual machines using Invoke-AzVMRunCommand. It also includes detailed procedures for establishing persistence through the creation of backdoor service principals and new administrative users.\n- [CREDENTIALS_UNSAFE]: The skill explicitly teaches the agent to harvest sensitive data, including exporting and importing stolen Azure authentication tokens, dumping secrets from Azure Key Vaults, and extracting environment variables from AWS Lambda functions which often contain sensitive API keys.\n- [DATA_EXFILTRATION]: Includes commands to perform mass data extraction by synchronizing entire S3 buckets or Google Cloud Storage buckets to local directories using CLI sync tools.\n- [EXTERNAL_DOWNLOADS]: Fetches various security frameworks and CLI tools from remote sources, including the AWS CLI binary and Python packages such as scoutsuite and pacu.\n- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection (Category 8). It ingests untrusted data from external sources such as cloud API responses, virtual machine metadata, and storage bucket contents without sanitization or boundary markers. This data is processed by the agent which possesses high-impact capabilities including shell execution, network operations, and file system access.

Cloud Penetration Testing