The Agent Skills Directory

Command Execution: The skill utilizes the AWS CLI and MCP server to perform infrastructure management tasks. It incorporates several safety controls to mitigate risks:
Permission Validation: Uses aws iam simulate-principal-policy to verify effective permissions before executing write operations.
Policy Safety Rules: Implements a mandatory workflow for put-bucket-policy that includes retrieving the existing policy, creating a timestamped backup, and merging new statements rather than overwriting blindly.
Syntax Validation: Employs python3 -m json.tool to validate JSON policy syntax before application.
User Confirmation: Instructs the agent to display write commands and wait for explicit user confirmation before execution.
Indirect Prompt Injection Surface: The skill ingests data from existing S3 bucket policies during audit and remediation workflows. While this presents a potential surface for indirect prompt injection, the risk is managed by treating the data as structured JSON and applying validation steps.
Ingestion Points: Reads existing policies via aws s3api get-bucket-policy (documented in SKILL.md and references/remediation.md).
Boundary Markers: No explicit boundary markers are used for the ingested policy content.
Capability Inventory: The skill has capabilities to modify S3 configurations, create KMS keys, and update CloudTrail settings.
Sanitization: Validates JSON structure using Python's standard library before processing.
Infrastructure Management: The skill interacts with various AWS services including S3, KMS, IAM, CloudTrail, and GuardDuty. All documented endpoints (e.g., s3.amazonaws.com) are official vendor resources.

securing-s3-buckets