Enum Runner 

pub enum Runner {
    WsOperator {
        variant: String,
        tools: Vec<String>,
    },
    WsClaudeCode {
        variant: String,
        tools: Vec<String>,
    },
    AgentBlockInProcess {
        tools: Vec<String>,
    },
    Subprocess {
        template: String,
        overrides: SubprocessOverrides,
    },
}

The execution shell an agent’s Worker IMPL runs inside — holding tool grant, model selection, and runtime capabilities. Tier 1 of the GH #46 3-tier Worker model (Runner / Agent / Context).

Runner here is broader than the ADK / OpenAI Agents SDK Runner (a loop driver): it is the execution shell holding tool grant, model selection, and runtime capabilities. Loop driving itself is the backend’s job (Claude Code harness / AgentBlock runtime).

Resolved per-agent by resolve_runner’s 5-step cascade; wiring the resolved value into the launch path is Milestone 3 — this Milestone only declares the shape and the pure resolver.

Variants

WsOperator

Platform-neutral WebSocket Operator backend. The joined execution environment may be Claude Code, Codex, or another MainAI/plugin that implements the common binding and spawn contracts.

Fields

variant: String

Provider-defined launch variant selected by the execution environment.

tools: Vec<String>

Minimum tool grant the provider must enforce.

WsClaudeCode

WS backend: Claude Code subagent wrapper. variant is the wrapper’s subagent_type; tools mirrors the wrapper frontmatter = enforced grant.

Kept as a compatibility backend for existing Blueprints. New platform-neutral declarations should use Self::WsOperator.

Fields

variant: String

The wrapper’s subagent_type (= WorkerBinding.variant in the mlua-swarm core crate).

tools: Vec<String>

Declared (informational) tool list — mirrors the wrapper frontmatter; the actual grant is enforced by the wrapper file itself, not by this list.

AgentBlockInProcess

In-process backend: agent-block runtime. tools is the effective (enforced) tool set for the in-process registry.

Enforcement is per AgentKind::AgentBlock mode (GH #86) and is server-granular: PromptBasedAgent embeds only the spec.mcp_servers entries named by an mcp__<server>__<tool> entry of this list, so an unlisted server is unreachable — but every tool of a listed server is reachable. ScriptBasedAgent (spec.script_path present) cannot be enforced at all — the script drives its own mcp.connect — so declaring mcp__ entries there is a compile error rather than a silent no-op.

Fields

tools: Vec<String>

Effective (enforced) tool set passed to the agent-block runtime’s registry — unlike WebSocket Runner tool requests, this list is not merely informational.

Declaring this Runner at all overrides profile.tools, including when the list is empty: an empty list is an enforced-empty grant (the way a Blueprint revokes an agent.md’s inherited tools: line), not “unset”. An agent that declares no Runner keeps profile.tools as its effective set. The override is applied once, when the Run’s immutable BoundAgent snapshot is projected for the compiler, so it is pinned for resume.

Subprocess

GH #83 — Subprocess EmbedAgent backend: the step runs headless through the ProcessSpawner path, with the invocation described by a SubprocessDef template looked up by name in Blueprint::subprocesses. Name symmetry with AgentKind::Subprocess is deliberate (1:1 — this variant is the Runner-axis face of the same Worker IMPL kind).

Per-agent overrides live HERE (not on SubprocessDef) so the template struct stays flat and shareable across agents.

Fields

template: String

SubprocessDef::name reference into Blueprint::subprocesses.

overrides: SubprocessOverrides

Per-agent overrides applied on top of the referenced template and the agent profile. Empty (all defaults) is omitted on the wire.

Trait Implementations

impl Clone for Runner

fn clone(&self) -> Runner

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Runner

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for Runner

fn deserialize<__D>( __deserializer: __D, ) -> Result<Runner, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Eq for Runner

impl JsonSchema for Runner

fn schema_name() -> Cow<'static, str>

The name of the generated JSON Schema.

fn schema_id() -> Cow<'static, str>

Returns a string that uniquely identifies the schema produced by this type.

fn json_schema(generator: &mut SchemaGenerator) -> Schema

Generates a JSON Schema for this type.

fn inline_schema() -> bool

Whether JSON Schemas generated for this type should be included directly in parent schemas, rather than being re-used where possible using the $ref keyword.

impl PartialEq for Runner

fn eq(&self, other: &Runner) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl Serialize for Runner

fn serialize<__S>( &self, __serializer: __S, ) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for Runner