Enum HarnessInternalEvent 

pub enum HarnessInternalEvent {
    AssistantTextChunk {
        msg_id: String,
        delta: String,
    },
    AssistantThinkingChunk {
        msg_id: String,
        delta: String,
    },
    ToolCall {
        id: String,
        name: String,
        input: Value,
    },
    ToolResult {
        id: String,
        output: Result<Value, String>,
    },
    CompactionApplied {
        original_message_count: usize,
        compacted_message_count: usize,
        original_tokens: u64,
        compacted_tokens: u64,
    },
    TurnEnd {
        stop_reason: String,
        usage: Option<HarnessUsage>,
        final_messages: Vec<ChatMessage>,
    },
}

Variants

AssistantTextChunk

Fields

msg_id: String

delta: String

AssistantThinkingChunk

Fields

msg_id: String

delta: String

ToolCall

Fields

id: String

name: String

input: Value

ToolResult

Fields

id: String

output: Result<Value, String>

CompactionApplied

Compaction fired between steps and the running history was folded. Counts let HR estimate how aggressive compaction is at this point in the turn (and whether to raise alerts). Token counts come from the harness’s own estimator — see compaction::estimate_messages_tokens — and are approximate.

Native_adapter currently does NOT project this onto an AdapterEvent (no proto field reserved for it yet). It surfaces only through structured tracing on the RD side; tests can still assert on it in the harness’s mpsc channel.

Fields

original_message_count: usize

compacted_message_count: usize

original_tokens: u64

compacted_tokens: u64

TurnEnd

Fields

stop_reason: String

usage: Option<HarnessUsage>

final_messages: Vec<ChatMessage>

Final messages history at turn end (post-compaction, includes the assistant’s reply and any tool round-trips). RD captures this snapshot to seed the next dispatch’s prior_messages so multi-turn conversations in the same RD process don’t re-prompt from scratch.

This is internal-only state: it never lands on the wire. native_adapter::HarnessEventAdapter::ingest reads it via a side-channel (history_handle) and drops it before projecting to AdapterEvent::TurnEnd. Tests can still assert on the raw HarnessInternalEvent directly.

Trait Implementations

impl Clone for HarnessInternalEvent

fn clone(&self) -> HarnessInternalEvent

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for HarnessInternalEvent

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

Formats the value using the given formatter.

impl PartialEq for HarnessInternalEvent

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

Tests for self and other values to be equal, and is used by ==.

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

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for HarnessInternalEvent