The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external platform and has capabilities to modify security configurations, which presents an indirect prompt injection risk.
Ingestion points: The skill fetches role and resource group data using the harness_list and harness_get tools (SKILL.md).
Boundary markers: No delimiters or specific instructions are provided to the agent to treat data from the platform as potentially untrusted or to ignore instructions embedded within role descriptions.
Capability inventory: The skill has the ability to create, update, and delete roles and assignments via harness_create, harness_update, and harness_delete (SKILL.md).
Sanitization: There is no defined process for sanitizing or validating descriptions or names retrieved from the Harness API before they are processed by the agent.

manage-roles