agentfootprint/dist/esm/core/terminalVerdict.d.ts

The explainable agent framework — backtrack a wrong answer to the exact context that caused it (evidence, not guesses). Built on footprintjs.

/**
 * A run's TERMINAL VERDICT: the way an agent's run ended that its caller must
 * receive as a typed error, held as a plain value a chart's state can carry.
 *
 * An agent ends some runs on a record in its state instead of a throw — a
 * reliability fail-fast, a policy halt, a denied message, a refused answer —
 * and only its run boundary (`Agent · finalizeResult`) turns that record into
 * the error. A composition mounts the agent's CHART, so that boundary never
 * runs: the composition asks the agent what the mount handed back
 * (`outcomeOf`, duck-typed — `core-flow/` imports no runner): its answer, or
 * the verdict read off the state its chart ended with. It carries the verdict
 * and raises the SAME error (`terminalErrorOf`). One translation, used by the
 * agent's own boundary and by every composition.
 *
 * Leaf module: the error classes only, no runner.
 */
import { type AnswerValidationReport } from '../answer-validation/types.js';
import { type ReliabilityFailRecord } from '../reliability/failFastRecord.js';
import { type PolicyHaltContext } from '../security/PolicyHaltError.js';
import { type UnsupportedValuesContext } from './agent/evidence/errors.js';
import { type MessageDeniedContext } from './agent/middleware/errors.js';
/** Each verdict keeps exactly what its error is built from. */
export type TerminalVerdict = {
    readonly kind: 'reliability-fail-fast';
    readonly record: ReliabilityFailRecord;
} | {
    readonly kind: 'policy-halt';
    readonly halt: PolicyHaltContext;
} | {
    readonly kind: 'message-denied';
    readonly denial: MessageDeniedContext;
} | {
    readonly kind: 'unsupported-values';
    readonly refusal: UnsupportedValuesContext;
} | {
    readonly kind: 'answer-validation';
    readonly report: AnswerValidationReport;
};
/** The typed error a verdict is raised as — the same one the agent's own run raises. */
export declare function terminalErrorOf(verdict: TerminalVerdict, snapshot?: unknown): Error;
/**
 * What a mount of a runner's chart handed back: the runner's ANSWER, or the
 * terminal verdict its chart ended on. A verdict is read ONLY from the chart's
 * state — handed over when the chart returned no result — never from a result.
 */
export type ChildOutcome = {
    readonly answer: unknown;
} | {
    readonly verdict: TerminalVerdict;
};
/** A runner that reads its own chart's outcome off a mount (an `Agent` does). */
export interface ReadsItsOwnOutcome {
    outcomeOf(sfOutput: unknown): ChildOutcome;
}
/** Whether a runner reads its own chart's outcome. */
export declare function readsItsOwnOutcome(runner: unknown): runner is ReadsItsOwnOutcome;