The Agent Skills Directory

[PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection by processing external data from CloudWatch logs and X-Ray traces. An attacker could potentially embed malicious instructions within application logs or VPC flow logs that the agent might interpret as instructions during analysis.
Ingestion points: The skill utilizes aws logs start-query and aws logs get-query-results to fetch and process log data (found in SKILL.md and references/logs-insights-queries.md).
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings for the agent when processing log content.
Capability inventory: The skill is configured with Bash(aws *) capabilities in the frontmatter, creating a potential path for unauthorized AWS actions if the agent follows instructions found within logs.
Sanitization: No automated sanitization or filtering of external log content is described.
[COMMAND_EXECUTION]: The skill provides a set of operational AWS CLI commands for querying logs, managing alarms, and fetching metrics. While these are necessary for the skill's purpose, they grant the agent significant control over the AWS environment and require user oversight.

observability