The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes powerful cloud management tools including aws, az, gcloud, and kubectl to perform enumeration and privilege escalation, which grants the agent significant control over cloud environments.
[DATA_EXFILTRATION]: The instructions include commands to access sensitive credentials from cloud metadata services (IMDS) using curl, which can lead to the exposure of the agent's or the target's environment credentials.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for processing untrusted data from external sources, which is a known vector for indirect injection attacks.
Ingestion points: The agent is instructed to search for secrets within untrusted JavaScript bundles, mobile APKs, and git leaks.
Boundary markers: There are no instructions to use delimiters or to ignore potential instructions embedded within the data being processed.
Capability inventory: The agent is empowered to execute privileged CLI commands and network operations based on the content of the untrusted data.
Sanitization: The skill lacks explicit requirements for sanitizing or validating extracted strings before they are used as arguments in shell commands, relying on regex patterns that may not prevent all forms of injection.

cloud-iam-deep