import type { Scenario, AgentReasoningState } from "./scenarios.js";
export interface TurnResult {
    turn: number;
    command: string[];
    intent?: string;
    agentReasoningState?: AgentReasoningState;
    inputs: Record<string, unknown>;
    outputs: Record<string, unknown>;
    stateBefore: Record<string, unknown>;
    stateAfter: Record<string, unknown>;
    stateChange: Record<string, unknown>;
    evidenceGenerated: unknown[];
}
export interface SessionArtifact {
    schemaVersion: "1.0.0";
    sessionId: string;
    scenarioId: string;
    humanPrompt: string;
    description: string;
    cliPath: string;
    repositoryRoot: string;
    startedAt: string;
    completedAt?: string;
    turns: TurnResult[];
}
export interface RunOptions {
    /** Path to the synth CLI executable. */
    cliPath: string;
    /** Optional working directory; a temp directory is created if omitted. */
    repositoryRoot?: string;
    /** Optional session identifier; a UUID is generated if omitted. */
    sessionId?: string;
}
/** A single entry in the intent reconstruction log. */
export interface IntentReconstructionEntry {
    turn: number;
    command: string;
    intent?: string;
    understoodAs: string;
    confidence: number;
    unknowns: string[];
    evidenceConsumed: string[];
    modelDelta?: {
        from: string;
        to: string;
    };
}
/** A failure-taxonomy classification for a session. */
export interface FailureTaxonomyEntry {
    category: "intent-confusion" | "missing-context" | "governance-confusion" | "vocabulary-mismatch" | "none";
    severity: "low" | "medium" | "high";
    description: string;
}
/** Derived report from a session artifact. */
export interface SessionReport {
    schemaVersion: "1.0.0";
    sessionId: string;
    scenarioId: string;
    intentReconstruction: IntentReconstructionEntry[];
    failureTaxonomy: FailureTaxonomyEntry[];
}
/**
 * Run a first-contact scenario and return a session artifact.
 *
 * The runner materializes the scenario repository, optionally initializes
 * SYNTH, executes each scripted turn through the CLI, and captures the
 * command output plus state change telemetry.
 */
export declare function runScenario(scenario: Scenario, options: RunOptions): Promise<SessionArtifact>;
/**
 * Persist a session artifact to disk.
 *
 * Returns the path of the written file.
 */
export declare function saveSessionArtifact(artifact: SessionArtifact, outputDir: string): Promise<string>;
/**
 * Persist a session artifact to disk with an explicit filename.
 *
 * Returns the path of the written file.
 */
export declare function saveSessionArtifactAs(artifact: SessionArtifact, outputDir: string, filename: string): Promise<string>;
/**
 * Persist a session report to disk next to its artifact.
 *
 * Returns the path of the written file.
 */
export declare function saveSessionReport(report: SessionReport, outputDir: string): Promise<string>;
/**
 * Persist a session report to disk with an explicit filename.
 *
 * Returns the path of the written file.
 */
export declare function saveSessionReportAs(report: SessionReport, outputDir: string, filename: string): Promise<string>;
/**
 * Compute a simple Semantic Alignment Score from a completed session artifact.
 *
 * This is a placeholder rubric; real scoring will be derived from observed
 * agent trajectories in EXP-FIRSTCONTACT-011.
 */
export declare function computeSemanticAlignmentScore(artifact: SessionArtifact): {
    intentRecognition: number;
    boundaryRecognition: number;
    evidenceSeeking: number;
    governanceCompliance: number;
    recoveryAbility: number;
    compressionEfficiency: number;
    overall: number;
};
/**
 * Reconstruct the agent's intent model across the session.
 *
 * Produces a chronological log of how the agent's understanding changed
 * with each piece of evidence, including the model delta between turns.
 */
export declare function reconstructIntent(artifact: SessionArtifact): IntentReconstructionEntry[];
/**
 * Classify a session against the first-contact failure taxonomy.
 *
 * Categories mirror the misinterpretation patterns observed in the
 * Windows agent experiment and the canonical scenarios:
 *   - intent-confusion: the agent models the project as the wrong kind of thing
 *   - missing-context: the request is underspecified and the agent lacks criteria
 *   - governance-confusion: the agent treats SYNTH/project boundaries incorrectly
 *   - vocabulary-mismatch: a feature request is treated as an immediate coding task
 */
export declare function classifySession(artifact: SessionArtifact, scenario?: Scenario): FailureTaxonomyEntry[];
/**
 * Build a derived session report from an artifact and its source scenario.
 */
export declare function buildSessionReport(artifact: SessionArtifact, scenario?: Scenario): SessionReport;