The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill facilitates the extraction of sensitive cloud credentials and targets configuration files.
Evidence: Provides commands to query the AWS EC2 metadata service at http://169.254.169.254/latest/meta-data/iam/security-credentials/ to obtain IAM role keys.
Evidence: Includes a workflow for extracting temporary AWS credentials from Cognito Identity Pools using the aws cognito-identity CLI.
Evidence: Directs the agent to probe for sensitive configuration files such as .env and config.json via common web paths.
[COMMAND_EXECUTION]: Provides executable command templates for probing and interacting with remote infrastructure.
Evidence: Includes curl commands for listing S3 bucket contents and writing to Firebase databases via PUT requests.
Evidence: Contains shell loops for brute-forcing common S3 bucket name permutations (e.g., target-backup, target-prod).
[PRIVILEGE_ESCALATION]: Documents methods to escalate privileges or gain unauthorized access within a target environment.
Evidence: Describes the exploitation of Jenkins Script Console for Remote Code Execution (RCE) via Groovy scripts to compromise the server.
Evidence: Details how to abuse over-permissioned guestRoleArn in AWS CloudWatch RUM to gain unauthorized access to other AWS services like S3 or DynamoDB.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection via user-supplied target data.
Ingestion points: User-supplied target URL, IP, or account name in SKILL.md.
Boundary markers: Present. Includes a "Mandatory confirmation gate" requiring user confirmation of targets, authorization, and commands.
Capability inventory: Shell command execution (curl, aws cli), file reading (cat), and container management (docker).
Sanitization: Absent. Command templates interpolate target variables directly without demonstrated escaping or validation logic.

hunt-cloud-misconfig