Struct AgentHarnessConfig 

pub struct AgentHarnessConfig {

    pub max_tool_calls_per_turn: usize,
    pub max_tool_wall_clock_secs: u64,
    pub max_tool_retries: u32,
    pub max_parallel_tool_calls: usize,
    pub auto_compaction_enabled: bool,
    pub auto_compaction_threshold_tokens: Option<u64>,
    pub auto_compaction_instructions: Option<String>,
    pub auto_compaction_pause_after: bool,
    pub compact_on_model_switch: bool,
    pub tool_result_clearing: ToolResultClearingConfig,
    pub max_budget_usd: Option<f64>,
    pub budget_warning_threshold: f64,
    pub continuation_policy: ContinuationPolicy,
    pub context_reset_mode: ContextResetMode,
    pub context_reset_stall_threshold: u32,
    pub event_log_path: Option<String>,
    pub orchestration_mode: HarnessOrchestrationMode,
    pub max_revision_rounds: usize,
    pub confidence_escalation: ConfidenceEscalationConfig,
    pub skeptic_panel: SkepticPanelConfig,
    pub verification: VerificationAutoRecoveryConfig,
    pub continuation: TrackerContinuationConfig,
    /* private fields */
}

Fields

max_tool_calls_per_turn: usize

Maximum number of tool calls allowed per turn. Defaults to 120. Set to 0 to disable the cap.

max_tool_wall_clock_secs: u64

Maximum wall clock time (seconds) for tool execution in a turn

max_tool_retries: u32

Maximum retries for retryable tool errors

max_parallel_tool_calls: usize

Maximum number of tool calls that may execute concurrently within a single parallel batch. Set to 0 to disable the cap (unlimited concurrency). Default: 4.

auto_compaction_enabled: bool

Enable automatic context compaction when token pressure crosses threshold.

Enabled by default. When disabled, normal threshold-triggered automatic compaction is skipped; the bounded post-tool recovery path may still compact the older prefix as a safety fallback after a provider failure.

auto_compaction_threshold_tokens: Option<u64>

Optional absolute compaction threshold (tokens) for native and local compaction.

When set, this may lower the trigger but cannot bypass the resolved model capacity or context.max_context_tokens safety ceiling. The next response budget is reserved before deriving the trigger.

auto_compaction_instructions: Option<String>

Optional custom instructions for the compaction summarization prompt. When set, replaces the default Anthropic compaction prompt entirely. Useful for tool-use scenarios to prevent the model from calling tools during summarization. Only applies to Anthropic provider.

auto_compaction_pause_after: bool

Whether to pause after compaction (Anthropic only). When true and compaction triggers, the API returns early with stop_reason: "compaction" and only the compaction block. The caller can then insert additional messages before the model generates its text response.

compact_on_model_switch: bool

Automatically compact conversation context when the main session model or provider is switched mid-conversation, so the newly selected model starts from a summary instead of the outgoing model’s raw trace. Default: true. Disable to keep the full history across a model switch.

tool_result_clearing: ToolResultClearingConfig

Provider-native tool-result clearing policy. When enabled, old tool results are stripped from the context once it grows past trigger_tokens, protecting recent results and assistant tool-call batches and always retaining at least clear_at_least_tokens. This bounds per-turn context growth and is enabled by default.

max_budget_usd: Option<f64>

Optional maximum estimated API cost in USD before VT Code stops the session.

budget_warning_threshold: f64

Fraction of max_budget_usd at which VT Code emits a one-time near-budget warning. Ignored when max_budget_usd is unset.

continuation_policy: ContinuationPolicy

Controls whether harness-managed continuation loops are enabled.

context_reset_mode: ContextResetMode

When to trigger a context reset — starting a clean session from external artifacts only, discarding conversation history. Distinct from compaction (which preserves conversational continuity). Default: off (carry forward history as before).

context_reset_stall_threshold: u32

Number of consecutive stall turns before on_stall context reset triggers. Ignored unless context_reset_mode = "on_stall". Default: 2.

event_log_path: Option<String>

Optional compatibility/export JSONL path for harness events. Canonical events are always stored under the workspace session store; unset configuration creates no global harness file.

orchestration_mode: HarnessOrchestrationMode

Select the exec/full-auto harness orchestration path.

max_revision_rounds: usize

Maximum generator revision rounds after evaluator rejection.

confidence_escalation: ConfidenceEscalationConfig

Confidence-based escalation for autonomous decisions.

Implements the escalation decision rule: Escalate iff p_success < tau_conf OR action in A_irreversible OR cost > B_auto

When a tool call is classified as irreversible or below the confidence threshold, the harness escalates to blocked-handoff instead of proceeding autonomously. Opt-in (default: disabled).

skeptic_panel: SkepticPanelConfig

Adversarial multi-model evaluator panel. When enabled, the harness runs the evaluator prompt against every listed model in parallel and aggregates the strictest verdict/scorecard across the panel. Opt-in (default: disabled).

verification: VerificationAutoRecoveryConfig

Autonomous recovery for the anti-blind-editing verification gate. When the model emits text instead of running a verifier, the harness grants bounded directive retries and (optionally) runs the detected project verifier itself instead of forcing manual continue.

continuation: TrackerContinuationConfig

Tracker-aware auto-continuation: when task_tracker still has incomplete steps, keep looping / auto-queue the next turn instead of ending and nudging the user to resume.

Trait Implementations

impl Clone for AgentHarnessConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for AgentHarnessConfig

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

Formats the value using the given formatter.

impl Default for AgentHarnessConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for AgentHarnessConfig

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl JsonSchema for AgentHarnessConfig

fn schema_name() -> Cow<'static, str>

The name of the generated JSON Schema.

fn schema_id() -> Cow<'static, str>

Returns a string that uniquely identifies the schema produced by this type.

fn json_schema(generator: &mut SchemaGenerator) -> Schema

Generates a JSON Schema for this type.

fn inline_schema() -> bool

Whether JSON Schemas generated for this type should be included directly in parent schemas, rather than being re-used where possible using the $ref keyword.

impl Serialize for AgentHarnessConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.