The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a wide range of shell and PowerShell commands to manage cloud infrastructure resources and interact with platform APIs. It includes advanced features like Invoke-AzVMRunCommand for running scripts on remote virtual machines.
[EXTERNAL_DOWNLOADS]: Downloads cloud administration and security testing tools from external sources, including official software distributions from Amazon and Google, as well as specialized offensive security frameworks hosted on GitHub.
[REMOTE_CODE_EXECUTION]: Contains instructions for fetching and executing installation scripts directly from the web (curl | bash). It also facilitates the execution of arbitrary commands on target cloud instances.
[DATA_EXFILTRATION]: Outlines procedures for retrieving data from cloud storage buckets and synchronizing it to the local environment, and includes steps for exporting authentication tokens to local files.
[CREDENTIALS_UNSAFE]: Provides scripts for automated password spraying and for harvesting plaintext credentials from resource attributes within cloud environments.
[PROMPT_INJECTION]: The skill is open to indirect prompt injection because it consumes data from external cloud services and security tool outputs. This could potentially allow malicious data within the target environment to influence the AI agent's decision-making process.
Ingestion points: Reads output from discovery tools and cloud metadata services.
Boundary markers: Relies on high-level confirmation gates for user interaction, but specific delimiters for external data are not used in command snippets.
Capability inventory: Possesses capabilities for subprocess execution, network access, and remote instance command execution.
Sanitization: No evidence of input sanitization or verification for the data retrieved from external tools or cloud APIs.

cloud-penetration-testing