Struct FighterLoopParams 

pub struct FighterLoopParams {

    pub manifest: FighterManifest,
    pub user_message: String,
    pub bout_id: BoutId,
    pub fighter_id: FighterId,
    pub memory: Arc<MemorySubstrate>,
    pub driver: Arc<dyn LlmDriver>,
    pub available_tools: Vec<ToolDefinition>,
    pub max_iterations: Option<usize>,
    pub context_window: Option<usize>,
    pub tool_timeout_secs: Option<u64>,
    pub coordinator: Option<Arc<dyn AgentCoordinator>>,
    pub approval_engine: Option<Arc<PolicyEngine>>,
    pub sandbox: Option<Arc<SandboxEnforcer>>,
}

Parameters for the fighter loop.

Fields

manifest: FighterManifest

The fighter’s manifest (identity, model config, system prompt, capabilities).

user_message: String

The user’s message to process.

bout_id: BoutId

The bout (session) ID.

fighter_id: FighterId

The fighter’s unique ID.

memory: Arc<MemorySubstrate>

Shared memory substrate for persistence.

driver: Arc<dyn LlmDriver>

The LLM driver to use for completions.

available_tools: Vec<ToolDefinition>

Tools available for this fighter to use.

max_iterations: Option<usize>

Maximum loop iterations before forced termination (default: 50).

context_window: Option<usize>

Context window size in tokens (default: 200K).

tool_timeout_secs: Option<u64>

Per-tool timeout in seconds (default: 120).

coordinator: Option<Arc<dyn AgentCoordinator>>

Optional agent coordinator for inter-agent tools.

approval_engine: Option<Arc<PolicyEngine>>

Optional policy engine for approval-gated tool execution. When present, the referee checks every move before the fighter can throw it.

sandbox: Option<Arc<SandboxEnforcer>>

Optional subprocess sandbox (containment ring) for shell and filesystem tools. When present, commands are validated and environments are sanitized before execution.