The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it processes untrusted data from external sources.
Ingestion points: The workflow involves reading and analyzing production traces, conversation histories, and external datasets (referenced in SKILL.md, conversation-evaluation.md, and rag-evaluation.md).
Boundary markers: The instructions lack explicit boundary markers or 'ignore' directives that would prevent the agent from obeying instructions embedded within the data it is tasked with analyzing.
Capability inventory: The skill is granted access to high-privilege tools such as Bash, Write, and Edit, in addition to the orq* platform tools, which increases the potential impact of a successful injection.
Sanitization: There is no evidence of sanitization or validation logic applied to the external data before it is presented to the agent's context.

run-experiment