Enum Op 

pub enum Op {
    SendMessage {

        content: String,
        mode: TurnLoopMode,
        model: String,
        goal_objective: Option<String>,
        reasoning_effort: Option<String>,
        reasoning_effort_auto: bool,
        auto_model: bool,
        allow_shell: bool,
        trust_mode: bool,
        auto_approve: bool,
        approval_mode: ApprovalMode,
        temperature: Option<f32>,
        top_p: Option<f32>,
        max_output_tokens: Option<u32>,
    
},
    CancelRequest,
    ApproveToolCall {
        id: String,
    },
    DenyToolCall {
        id: String,
    },
    SpawnSubAgent {
        prompt: String,
    },
    ListSubAgents,
    ChangeMode {
        mode: TurnLoopMode,
    },
    SetModel {
        model: String,
    },
    SetCompaction {
        config: CompactionConfig,
    },
    SyncSession {
        messages: Vec<Message>,
        system_prompt: Option<SystemPrompt>,
        model: String,
        workspace: PathBuf,
    },
    ApplyKernelResume {
        hints: KernelResumeHints,
    },
    CompactContext,
    Rlm {
        content: String,
        model: String,
        child_model: String,
        max_depth: u32,
    },
    EditLastTurn {
        new_message: String,
    },
    TruncateBeforeLastUserMessage {
        reply: Sender<bool>,
    },
    QueryContext {
        reply: Sender<ThreadContextSnapshot>,
    },
    QueryHarnessTaskGraph {
        reply: Sender<Value>,
    },
    QueryHarnessCycles {
        reply: Sender<Value>,
    },
    Shutdown,
}

Operations that can be submitted to the engine.

Variants

SendMessage

Send a message to the AI.

Fields

content: String

mode: TurnLoopMode

model: String

goal_objective: Option<String>

reasoning_effort: Option<String>

Reasoning-effort tier: "off" | "low" | "medium" | "high" | "max". None lets the provider apply its default.

reasoning_effort_auto: bool

True when the user selected auto thinking, even though the UI sends a concrete per-turn value to the model API.

auto_model: bool

True when the user selected auto model routing.

allow_shell: bool

trust_mode: bool

auto_approve: bool

approval_mode: ApprovalMode

temperature: Option<f32>

top_p: Option<f32>

max_output_tokens: Option<u32>

CancelRequest

Cancel the current request.

ApproveToolCall

Approve a tool call that requires permission.

Fields

id: String

DenyToolCall

Deny a tool call that requires permission.

Fields

id: String

SpawnSubAgent

Spawn a sub-agent.

Fields

prompt: String

ListSubAgents

List current sub-agents and their status.

ChangeMode

Change the operating mode.

Fields

mode: TurnLoopMode

SetModel

Update the model being used.

Fields

model: String

SetCompaction

Update auto-compaction settings.

Fields

config: CompactionConfig

SyncSession

Sync engine session state (used for resume/load).

Fields

messages: Vec<Message>

system_prompt: Option<SystemPrompt>

model: String

workspace: PathBuf

ApplyKernelResume

Restore kernel turn frame from persisted event-log projection (Phase 3b 6e).

Fields

hints: KernelResumeHints

CompactContext

Run context compaction immediately.

Rlm

Run a Recursive Language Model (RLM) turn per Algorithm 1 of Zhang et al. (arXiv:2512.24601). The prompt is stored in the REPL as context; the root LLM only sees metadata.

Fields

content: String

The user’s prompt — stored in REPL, NOT in the LLM context.

model: String

The model to use for root LLM calls.

child_model: String

The model to use for sub-LLM (llm_query) calls.

max_depth: u32

Recursion budget for sub_rlm() calls. Paper experiments use depth=1; defaults set by the /rlm command.

EditLastTurn

Edit the last user message: remove the last user+assistant exchange from the session, then re-send with the new content.

Fields

new_message: String

TruncateBeforeLastUserMessage

Drop the last user message and everything after it (#383 /edit, F4 HTTP).

Fields

reply: Sender<bool>

QueryContext

Return a TUI-aligned context usage snapshot (Zagens / runtime API).

Fields

reply: Sender<ThreadContextSnapshot>

QueryHarnessTaskGraph

Return derived LHT task graph JSON (GET …/harness/task-graph).

Fields

reply: Sender<Value>

QueryHarnessCycles

Return cycle briefings + archive summaries (GET …/harness/cycles).

Fields

reply: Sender<Value>

Shutdown

Shutdown the engine.

Trait Implementations

impl Debug for Op

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

Formats the value using the given formatter.