Enum HarnessEvent 

pub enum HarnessEvent {
    RunStarted {
        run_id: RunId,
        agent_id: AgentId,
    },
    ContextBuilt {
        run_id: RunId,
        turn_id: TurnId,
        context: ContextPack,
    },
    ContextCompacted {
        run_id: RunId,
        turn_id: TurnId,
        estimated_tokens_before: u32,
        estimated_tokens_after: u32,
        dropped_turn_start: u64,
        dropped_turn_end_exclusive: u64,
    },
    ModelRequested {
        run_id: RunId,
        turn_id: TurnId,
        step: u32,
        model: ModelName,
    },
    ModelFailed {
        run_id: RunId,
        turn_id: TurnId,
        step: u32,
        reason: String,
    },
    ModelResponded {
        run_id: RunId,
        turn_id: TurnId,
        step: u32,
        output: Message,
        proposed_calls: Vec<ToolProposal>,
        served_model: Option<ModelName>,
        usage: Option<ModelUsage>,
    },
    ToolProposalsRejected {
        run_id: RunId,
        turn_id: TurnId,
        reason: String,
    },
    ToolCallProposed {
        run_id: RunId,
        turn_id: TurnId,
        call: ToolCall,
    },
    PolicyEvaluated {
        run_id: RunId,
        call_id: ToolCallId,
        decision: PolicyDecision,
    },
    ApprovalGranted {
        run_id: RunId,
        call_id: ToolCallId,
        actor_id: ActorId,
    },
    ApprovalDenied {
        run_id: RunId,
        call_id: ToolCallId,
        actor_id: ActorId,
        reason: String,
    },
    ToolStarted {
        run_id: RunId,
        call_id: ToolCallId,
    },
    ToolFinished {
        run_id: RunId,
        result: ToolResult,
    },
    ToolFailed {
        run_id: RunId,
        call_id: ToolCallId,
        reason: String,
    },
    RunFinished {
        run_id: RunId,
    },
    RunFailed {
        run_id: RunId,
        reason: String,
    },
}

Durable event log entries. Transcript, metrics, replay, and audit views are derived from this log.

The tagged JSON schema rejects unknown fields rather than discarding durable data.

Variants

RunStarted

Binds a fresh state machine to one run and agent.

Fields

run_id: RunId

Durable run identifier shared by every later event.

agent_id: AgentId

Agent identity selected by the host for this run.

ContextBuilt

Records the exact bounded context selected for a model turn.

Fields

run_id: RunId

Run receiving the context.

turn_id: TurnId

New turn identifier; concluded turn identifiers cannot be reused.

context: ContextPack

Lane-labeled context that must pass budget validation.

ContextCompacted

Records prior turns omitted from the next model context.

Fields

run_id: RunId

Run receiving the compacted context.

turn_id: TurnId

Turn whose context will reflect the compaction.

estimated_tokens_before: u32

Estimated tokens before prior turns were dropped.

estimated_tokens_after: u32

Estimated tokens after prior turns were dropped.

dropped_turn_start: u64

Zero-based first dropped prior-turn position.

dropped_turn_end_exclusive: u64

Exclusive end of the dropped prior-turn range.

ModelRequested

Records that a model request crossed the provider boundary.

Fields

run_id: RunId

Run making the request.

turn_id: TurnId

Turn whose context is being submitted.

step: u32

Monotonic model step expected by the run state.

model: ModelName

Host-selected model used for the request.

ModelFailed

Records that the pending model request failed without terminating the run.

Fields

run_id: RunId

Run containing the pending request.

turn_id: TurnId

Turn of the pending model request.

step: u32

Model step of the pending request.

reason: String

Durable host-reported failure reason.

ModelResponded

Records the normalized result returned by the pending model request.

Fields

run_id: RunId

Run receiving the response.

turn_id: TurnId

Turn of the pending model request.

step: u32

Model step of the pending request.

output: Message

Normalized model-authored message.

proposed_calls: Vec<ToolProposal>

Unvalidated model proposals; an empty list concludes the turn.

served_model: Option<ModelName>

Provider-reported model that served the response, when available.

This is an auditable response fact. The host-selected request model remains recorded separately on HarnessEvent::ModelRequested. Unknown values are omitted so pre-field 0.3.0 records round-trip without gaining a field.

usage: Option<ModelUsage>

Provider-reported token usage, or None when usage is unknown.

Reported zero counts remain known usage rather than None.

ToolProposalsRejected

Rejects every unvalidated tool proposal in the pending model response.

Fields

run_id: RunId

Run containing the pending proposals.

turn_id: TurnId

Turn that produced the pending proposals.

reason: String

Non-empty host explanation for rejecting the whole proposal batch.

ToolCallProposed

Records a model proposal after host validation and effect classification.

Fields

run_id: RunId

Run containing the proposal.

turn_id: TurnId

Turn that produced the proposal.

call: ToolCall

Host-validated call; its tool and input must match a pending proposal.

PolicyEvaluated

Records the policy decision for the pending validated call.

Fields

run_id: RunId

Run containing the call.

call_id: ToolCallId

Pending call evaluated by policy.

decision: PolicyDecision

Durable allow, approval, or denial decision.

ApprovalGranted

Records who granted a pending approval.

Fields

run_id: RunId

Run containing the call.

call_id: ToolCallId

Pending call approved for execution.

actor_id: ActorId

Human or host actor that granted approval.

ApprovalDenied

Records who denied a pending approval and why.

Fields

run_id: RunId

Run containing the call.

call_id: ToolCallId

Pending call denied before execution.

actor_id: ActorId

Human or host actor that denied approval.

reason: String

Durable denial reason for audit and continuation context.

ToolStarted

Records that the host began executing the approved call.

Fields

run_id: RunId

Run containing the call.

call_id: ToolCallId

Approved call that crossed the execution boundary.

ToolFinished

Records the structured result returned by the running call.

Fields

run_id: RunId

Run containing the call.

result: ToolResult

Result whose call id must match the running call.

ToolFailed

Records that the running call failed without a result.

Fields

run_id: RunId

Run containing the call.

call_id: ToolCallId

Running call that failed.

reason: String

Durable host-reported failure reason.

RunFinished

Terminates a concluded turn as a successful run.

Fields

run_id: RunId

Run that completed.

RunFailed

Terminates any started, nonterminal run as failed.

Fields

run_id: RunId

Run that failed.

reason: String

Durable terminal failure reason.

Implementations

impl HarnessEvent

pub fn run_id(&self) -> &RunId

Returns the owning run id without validating event order or phase.

pub fn name(&self) -> &'static str

Returns the stable snake-case event name used in transition diagnostics.

Trait Implementations

impl Clone for HarnessEvent

fn clone(&self) -> HarnessEvent

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for HarnessEvent

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for HarnessEvent

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for HarnessEvent

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

Equality operator ==.

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

Inequality operator !=.

impl Serialize for HarnessEvent

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

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for HarnessEvent