The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides detailed command-line instructions for using cloud provider CLIs and standard utilities to execute attack chains.\n
Evidence: Provides commands for aws iam, az role, gcloud auth, and kubectl to perform enumeration and privilege escalation.\n
Evidence: Lists specific commands for exploiting IAM privileges, such as iam:CreateAccessKey and iam:PassRole combined with ec2:RunInstances.\n- [DATA_EXFILTRATION]: Provides procedures for exploiting SSRF vulnerabilities to access sensitive cloud metadata services.\n
Evidence: Includes curl commands targeting 169.254.169.254 (AWS, Azure, and GCP IMDS endpoints) to extract temporary IAM credentials and tokens.\n
Evidence: Instructions for extracting Kubernetes ServiceAccount tokens (JWTs) from the API or internal pod environment.\n- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection and includes instructions that mimic security verdict logic.\n
Ingestion points: Cloud credentials and tokens sourced from external, untrusted data (git-leaks, JS bundles, APKs, or SSRF responses).\n
Boundary markers: Absent; there are no instructions to validate or sanitize the content of extracted credentials before use in shell commands.\n
Capability inventory: Extensive execution of subprocesses including aws, az, gcloud, kubectl, and curl across multiple files/sections.\n
Sanitization: Absent; found data is interpolated directly into command arguments without escaping.\n
Evidence: Includes a 'Severity scoring guidance' table that defines findings as 'Critical' or 'High', which could be used to influence automated security analyzers.

cloud-iam-deep