The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates to ingest and analyze data from external AWS sources such as CloudTrail events and logs using DQL commands like fetch events and fetch logs. This creates a theoretical surface for indirect prompt injection if the ingested log content contains malicious instructions. However, as this is a core requirement for a cloud monitoring skill, and the skill itself performs read-only operations within the Dynatrace environment without high-privilege capabilities or outbound network requests, the risk is categorized as safe.
Ingestion points: references/events.md (fetch events), SKILL.md (fetch logs, fetch bizevents).
Boundary markers: Not explicitly defined in the DQL templates.
Capability inventory: Read-only data retrieval and relationship traversal via Dynatrace DQL; no file-write or external network capabilities.
Sanitization: Relies on DQL parsing (parse, fieldsAdd) to structure input data.
[SAFE]: Analysis of the skill instructions and DQL templates found no evidence of prompt injection, credential exfiltration, obfuscation, or unauthorized remote code execution. All resource access is scoped to the Dynatrace monitoring platform.

dt-obs-aws