The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The auditing tool in scripts/agent.py retrieves data from AWS environment components such as IAM role descriptions, KMS policies, and CloudTrail event logs. This information, which can be influenced by other users or actors in the AWS account, is incorporated into a JSON report. If an attacker names resources with malicious instructions, those instructions could theoretically influence an AI agent that subsequently processes the report.
Ingestion points: scripts/agent.py (EC2, KMS, IAM, and CloudTrail data retrieval functions).
Boundary markers: The script does not use specific delimiters to isolate external data within the generated audit report.
Capability inventory: The skill performs AWS API read operations, local file system writes for reporting, and uses AWS SSM to inspect configuration files.
Sanitization: No specific prompt-level sanitization is performed on the data retrieved from AWS APIs before it is written to the report file.
[EXTERNAL_DOWNLOADS]: The documentation in SKILL.md references the official AWS Nitro Attestation PKI root certificate hosted on a well-known AWS service domain (aws-nitro-enclaves.amazonaws.com). This is a standard and expected download for the purpose of validating enclave attestation documents.
[COMMAND_EXECUTION]: The agent.py script uses the AWS SSM send_command functionality to execute a specific cat command on a targeted configuration file (/etc/nitro_enclaves/allocator.yaml). This is used solely to verify that the Nitro Enclave allocator has been correctly configured with sufficient memory and CPU resources.

implementing-aws-nitro-enclave-security