The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-controllable input for resource management operations, which creates a potential surface for indirect prompt injection.
Ingestion points: The skill accepts untrusted data through parameters such as search_term in listing operations, and name, description, identifier, and email during resource creation in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat these inputs as literal data rather than instructions.
Capability inventory: The skill leverages harness_create, harness_delete, harness_list, and harness_get tools, which allow for the modification of platform security principals like user groups and service accounts.
Sanitization: No evidence of input sanitization or validation is present within the skill instructions, relying entirely on the underlying MCP server's implementation.

manage-users