Agents, Tools and MCP
Governed execution, approvals, skills and external capability discovery.
Agent lifecycle
An agent combines instructions, eligible models, Knowledge and approved tools. Publishing an agent makes its configured behavior reusable within an intended scope.
WICSAI_AGENT_UNAVAILABLE can indicate a missing runtime, Group binding, entitlement or allowance. An agent’s instructions cannot grant permissions that the user or workspace lacks.
Reusable procedures
A skill is a reusable procedure with a bounded purpose. Agents may use skills, but a skill does not add permissions by itself.
Keep required tools and source access minimal. If a step requests a consequential action, inspect its target and effect before approval.
Discovery and authorization
MCP exposes external capabilities through a standard protocol. Discovering a tool does not authorize Yasnora or a Group to use it.
Even a read-only tool can expose sensitive material. Reject unexpected requests or excessive scopes; keep credentials server-side.
Consequential actions
Work is the execution inbox for tasks that are waiting, running, need attention, completed or failed. It helps distinguish an active run from a finished result.
A completed run still needs output review. An approval should identify the action, target and effect; reject an unexpected or overly broad request.