Struct ConfidenceEscalationConfig 

pub struct ConfidenceEscalationConfig {
    pub enabled: bool,
    pub confidence_threshold: f64,
    pub always_escalate_tools: Vec<String>,
    pub cost_threshold_usd: f64,
    pub use_llm_confidence: bool,
    pub plan_mode_only: bool,
    pub max_replan_attempts: u32,
    pub max_total_escalations: u32,
    pub prompt_user_on_exhaust: bool,
}

Implements the escalation decision rule from “The Hitchhiker’s Guide to Agentic AI” (Eq. 18.12): Escalate iff p_success < tau_conf OR action in A_irreversible OR cost > B_auto

When enabled, the harness evaluates tool calls before execution and escalates high-risk or low-confidence actions to a blocked-handoff (requiring human review).

Fields

enabled: bool

Master switch. Default: false (opt-in).

confidence_threshold: f64

Minimum p_success threshold (0.0–1.0). Tools whose estimated success probability falls below this threshold trigger escalation.

always_escalate_tools: Vec<String>

Tool names that always trigger escalation regardless of confidence. These are matched against the tool call’s function name.

cost_threshold_usd: f64

Maximum estimated cost in USD before escalation. Tool calls whose estimated cost exceeds this threshold trigger escalation.

use_llm_confidence: bool

Whether to solicit LLM-based confidence estimation alongside heuristics. When false (default), only heuristic signals (error history, tool class) are used to estimate p_success.

plan_mode_only: bool

When true, only escalate during PlanBuildEvaluate orchestration mode. During single-mode or other orchestration modes, escalation is skipped.

max_replan_attempts: u32

Maximum number of re-plan attempts via conversation injection before prompting the user. The agent sees a structured message explaining which tools were blocked and is asked to try a different approach.

max_total_escalations: u32

Maximum total escalations (across all chain steps) before aborting with partial results. Must be >= max_replan_attempts.

prompt_user_on_exhaust: bool

Whether to prompt the user when the escalation chain is exhausted. When false, the chain falls through directly to abort-with-partial-results.

Implementations

impl ConfidenceEscalationConfig

pub fn would_escalate( &self, tool_name: &str, estimated_cost_usd: Option<f64>, ) -> bool

Returns true if any escalation rule would trigger for the given tool name and estimated cost. Used as a fast pre-check before running the full gate.

Trait Implementations

impl Clone for ConfidenceEscalationConfig

fn clone(&self) -> ConfidenceEscalationConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ConfidenceEscalationConfig

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

Formats the value using the given formatter.

impl Default for ConfidenceEscalationConfig

fn default() -> ConfidenceEscalationConfig

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for ConfidenceEscalationConfig

fn deserialize<__D>( __deserializer: __D, ) -> Result<ConfidenceEscalationConfig, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl JsonSchema for ConfidenceEscalationConfig

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 ConfidenceEscalationConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.