Enum AgentTraceEvent 

pub enum AgentTraceEvent {
    LlmCall {
        call_id: String,
        model: String,
        input_tokens: i64,
        output_tokens: i64,
        cache_tokens: i64,
        duration_ms: u64,
        iteration: usize,
    },
    ToolExecution {
        tool_name: String,
        tool_use_id: String,
        duration_ms: u64,
        status: String,
        classification: String,
        iteration: usize,
    },
    ToolRejected {
        tool_name: String,
        reason: String,
        iteration: usize,
    },
    LoopIntervention {
        tool_name: String,
        kind: String,
        count: usize,
        iteration: usize,
    },
    ContextCompaction {
        archived_messages: usize,
        new_summary_len: usize,
        iteration: usize,
    },
    PhaseChange {
        from_phase: String,
        to_phase: String,
        iteration: usize,
    },
    LoopComplete {
        status: String,
        iterations: usize,
        total_duration_ms: u64,
        tools_used: Vec<String>,
        successful_tools: Vec<String>,
    },
    SchemaRetry {
        attempt: usize,
        errors: Vec<String>,
        nudge_used: bool,
        correction_prompt: String,
    },
    NativeToolFallback {
        iteration: usize,
        accepted: bool,
        policy: String,
        fallback_index: usize,
        tool_call_count: usize,
    },
    EmptyCompletionRetry {
        iteration: usize,
        attempt: usize,
        error: String,
    },
}

Fine-grained event emitted during agent loop execution. Captures tool calls, LLM calls, interventions, compaction, and phase changes so downstream consumers (portal, IDE hosts, cloud runners) can display execution traces without reconstructing them from raw JSON.

Variants

LlmCall

Fields

call_id: String

model: String

input_tokens: i64

output_tokens: i64

cache_tokens: i64

duration_ms: u64

iteration: usize

ToolExecution

Fields

tool_name: String

tool_use_id: String

duration_ms: u64

status: String

classification: String

iteration: usize

ToolRejected

Fields

tool_name: String

reason: String

iteration: usize

LoopIntervention

Fields

tool_name: String

kind: String

count: usize

iteration: usize

ContextCompaction

Fields

archived_messages: usize

new_summary_len: usize

iteration: usize

PhaseChange

Fields

from_phase: String

to_phase: String

iteration: usize

LoopComplete

Fields

status: String

iterations: usize

total_duration_ms: u64

tools_used: Vec<String>

successful_tools: Vec<String>

SchemaRetry

Emitted when llm_call re-prompts the model after the previous response failed output_schema validation. One event per retry; attempt counts retries (the initial call is attempt 0 and produces no event; the first retry emits attempt: 1).

The retry does not persist the invalid response — the original messages are replayed with a single appended user-role correction that cites the validation errors and schema. That correction text is surfaced here as correction_prompt so transcripts show both why the retry happened and what was sent.

Fields

attempt: usize

errors: Vec<String>

nudge_used: bool

correction_prompt: String

NativeToolFallback

Fields

iteration: usize

accepted: bool

policy: String

fallback_index: usize

tool_call_count: usize

EmptyCompletionRetry

Fields

iteration: usize

attempt: usize

error: String

Trait Implementations

impl Clone for AgentTraceEvent

fn clone(&self) -> AgentTraceEvent

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for AgentTraceEvent

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

Formats the value using the given formatter.

impl Serialize for AgentTraceEvent

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

where __S: Serializer,

Serialize this value into the given Serde serializer.