Expand description
Tool annotations — the single source of truth for tool semantics.
These types describe what a tool does at a semantic level. The VM
consumes them to make policy decisions (read-only vs mutating, which
argument holds the workspace path, which aliases to normalize, etc.)
without hardcoding tool names or file-extension lists. Pipeline
authors declare a ToolAnnotations value per tool in their
CapabilityPolicy.tool_annotations registry; everything downstream
is driven by that declaration.
This alignment is ACP-compliant: ToolKind matches the canonical
tool-kind vocabulary from the [Agent Client Protocol schema]
(https://agentclientprotocol.com/protocol/schema) one-for-one.
Structs§
- Tool
Annotations - Full annotations for one tool. Pipelines populate one of these per tool in the capability-policy registry; the VM consults the registry on every tool call.
- Tool
ArgSchema - Declarative description of a tool’s argument shape. The VM uses this to:
Enums§
- Side
Effect Level - Rough side-effect taxonomy for the capability-ceiling check.
- Tool
Kind - Canonical tool-kind vocabulary. Matches the ACP
ToolKindenum so harn-cli’s ACP server can forward the value unchanged insessionUpdatevariants.