The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources (AWS services like Lambda, RDS, and API Gateway) through log ingestion tools.
Ingestion points: GetLogEvents, FilterLogEvents, and GetQueryResults commands in SKILL.md retrieve log data into the agent's context.
Boundary markers: The skill instructions do not specify any delimiters or safety warnings to distinguish between log content and agent instructions.
Capability inventory: The skill possesses powerful write and delete capabilities, including PutResourcePolicy, PutSubscriptionFilter, DeleteLogGroup, and DeleteResourcePolicy in SKILL.md. If an agent is influenced by malicious log content, it could misuse these tools to modify access controls or delete logs.
Sanitization: There is no evidence of filtering, escaping, or sanitizing the log data before it is presented to the agent.
[COMMAND_EXECUTION]: The skill enables high-privilege management of AWS CloudWatch infrastructure.
Evidence: The tool includes commands such as PutResourcePolicy (modifying IAM-like policies for logs), DeleteLogGroup, and PutSubscriptionFilter (which can redirect logs to other AWS services like Kinesis). These operations are powerful cloud management actions that require appropriate IAM permissions in the environment where the skill is executed.

awsflow-cloudwatch