The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The code examples in references/EXAMPLES.md demonstrate processing data from untrusted external sources (API Gateway, S3, SQS, and DynamoDB Streams) without incorporating boundary markers or sanitization logic. This creates a vulnerability surface where malicious payloads in external data could influence the behavior of the processing logic.
Ingestion points: create_order_handler (API body), s3_event_handler (S3 object content), sqs_batch_handler (SQS message body), and process_order_stream_handler (DynamoDB record data) in references/EXAMPLES.md.
Boundary markers: None present in the code examples to delimit external data from instructions.
Capability inventory: The skill's examples include capabilities to write to S3, write to DynamoDB, publish to SNS, and invoke other Lambda functions.
Sanitization: No validation or sanitization is applied to the data retrieved from external sources beyond basic JSON parsing.
[PRIVILEGE_ESCALATION]: The Serverless Framework configuration in references/EXAMPLES.md defines IAM policy statements that use wildcards (*) for the lambda:InvokeFunction action. This grants the service permissions to invoke any Lambda function within the account and region, which exceeds the principle of least privilege required for the described architecture.

serverless-expert