The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides an extensive list of CLI commands for AWS (aws), Azure (az), GCP (gcloud), and Tencent Cloud (tccli) to perform identity verification, permission enumeration, and privilege escalation.
[EXTERNAL_DOWNLOADS]: The skill downloads the coscli utility from https://cosbrowser.cloud.tencent.com/software/coscli/coscli. This is an official domain for Tencent Cloud services and is considered a well-known service.
[PRIVILEGE_ESCALATION]: A central feature of the skill is detailing specific exploitation paths to elevate privileges, such as iam:CreatePolicyVersion, iam:PassRole combined with Lambda or SCF function creation, and exploiting broad trust policies in sts:AssumeRole configurations.
[PERSISTENCE]: The skill contains instructions for creating long-term persistence in a compromised cloud environment by creating new sub-users and generating "backdoor" Access Keys via the CAM API.
[DYNAMIC_EXECUTION]: The skill includes Python code snippets (in references/tencent-cam-escalation.md) that programmatically interact with cloud APIs to create policies and use base64 to package executable cloud functions (SCF) at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external cloud environment responses (such as IAM policy documents, role descriptions, and resource metadata). This represents a vulnerability surface where malicious content embedded in the cloud infrastructure could attempt to influence the agent's behavior.
Ingestion points: Data entering the context from aws, tccli, and SDK responses (SKILL.md, references/aws-escalation.md).
Boundary markers: None identified.
Capability inventory: Shell command execution, network API requests, and local file operations (SKILL.md).
Sanitization: None identified.

cloud-iam-audit