agent-afk 5.264.0 → 5.265.1

package/dist/whatif/index.d.ts

@@ -1,6 +1,6 @@
 export { runWhatif, WhatifBudgetError } from './run.js';
 export { computeStructuralImpact } from './structural.js';
-export type { Change, ChangeKind, ChangeSpec, LaunchSettings, Environment, OperatorContext, ChangeOperator, RequestSnapshot, StructuralImpact, PredictionDirection, Confidence, Prediction, Episode, ToolRequest, EpisodeTrace, RunnerOptions, AgentRunner, EpisodeFeatures, JudgeQuestion, JudgeInput, JudgeResult, Judge, RateComparison, Verdict, VerifiedPrediction, PredictionScope, DiscoveredDifference, FeatureDelta, VerifyResult, WhatifReport, WhatifStage, WhatifProgress, CompleteFn, WhatifOptions, WhatifDeps, } from './types.js';
+export type { Change, ChangeKind, ChangeSpec, LaunchSettings, Environment, OperatorContext, ChangeOperator, RequestSnapshot, StructuralImpact, PredictionDirection, Confidence, Prediction, PredictionObservable, Episode, ToolRequest, EpisodeTrace, RunnerOptions, AgentRunner, EpisodeFeatures, JudgeQuestion, JudgeInput, JudgeResult, Judge, RateComparison, Verdict, VerifiedPrediction, PredictionScope, DiscoveredDifference, FeatureDelta, VerifyResult, WhatifReport, WhatifStage, WhatifProgress, CompleteFn, WhatifOptions, WhatifDeps, } from './types.js';
 export { renderTerminal, renderMarkdown } from './report.js';
 export { createAfkRunner } from './runner/afk-runner.js';
 export { createAnthropicComplete } from './complete.js';

package/dist/whatif/observability.d.ts

@@ -0,0 +1,5 @@
+import type { Prediction, PredictionObservable } from './types.js';
+export declare const INTERCEPTED_INTENT_RULE: string;
+export declare const DEFAULT_DOWNSTREAM_REASON = "needs an intercepted action to complete; episodes stop at the first side-effecting request";
+export declare function observabilityOf(prediction: Pick<Prediction, 'observable'>): PredictionObservable;
+export declare function unobservableReason(prediction: Pick<Prediction, 'observable' | 'observabilityReason'>): string | undefined;

package/dist/whatif/predict.d.ts

@@ -1,4 +1,5 @@
 import type { CompleteFn, Prediction, StructuralImpact } from './types.js';
+import type { RepoManifest } from './repo-manifest.js';
 export interface PredictInput {
     spec: {
         title: string;
@@ -7,5 +8,6 @@ export interface PredictInput {
     changeDescriptions: string[];
     structural: StructuralImpact;
     trackRecord?: string;
+    repoManifest?: RepoManifest;
 }
 export declare function predictChanges(input: PredictInput, complete: CompleteFn, model: string): Promise<Prediction[]>;

package/dist/whatif/probe-grounding.d.ts

@@ -0,0 +1,17 @@
+export interface DroppedProbe {
+    predictionId: string;
+    probe: string;
+    reason: string;
+}
+export interface GroundedPrediction {
+    id: string;
+    probes: string[];
+}
+export interface GroundProbesResult<T extends GroundedPrediction> {
+    predictions: T[];
+    droppedProbes: DroppedProbe[];
+}
+export declare function extractPathTokens(probe: string): string[];
+export type ExistenceChecker = (relPath: string) => boolean;
+export declare function makeSetChecker(allPaths: Set<string>): ExistenceChecker;
+export declare function groundProbes<T extends GroundedPrediction>(predictions: T[], existenceChecker: ExistenceChecker): GroundProbesResult<T>;

package/dist/whatif/repo-manifest.d.ts

@@ -0,0 +1,8 @@
+export interface RepoManifest {
+    languages: string[];
+    paths: string[];
+    allPaths: Set<string>;
+}
+export declare function buildRepoManifest(cwd: string): RepoManifest;
+export declare function formatRepoManifest(manifest: RepoManifest): string;
+export declare function pathExistsInCwd(cwd: string, relPath: string): boolean;

package/dist/whatif/report-verdict.d.ts

@@ -0,0 +1,3 @@
+import type { Verdict, VerifiedPrediction } from './types.js';
+export declare function verdictEmoji(v: Verdict): string;
+export declare function verdictLabel(vp: Pick<VerifiedPrediction, 'verdict' | 'unobservableReason'>): string;

package/dist/whatif/trace-render.d.ts

@@ -0,0 +1,2 @@
+import type { EpisodeTrace } from './types.js';
+export declare function renderTrace(trace: EpisodeTrace): string;

package/dist/whatif/types.d.ts

@@ -95,6 +95,7 @@ export interface StructuralImpact {
 }
 export type PredictionDirection = 'added' | 'removed' | 'strengthened' | 'weakened';
 export type Confidence = 'high' | 'medium' | 'low';
+export type PredictionObservable = 'decision' | 'downstream';
 export interface Prediction {
     id: string;
     behavior: string;
@@ -103,6 +104,8 @@ export interface Prediction {
     reason: string;
     testQuestion: string;
     probes: string[];
+    observable?: PredictionObservable;
+    observabilityReason?: string;
 }
 export interface Episode {
     id: string;
@@ -174,7 +177,7 @@ export interface RateComparison {
         candidate: number;
     };
 }
-export type Verdict = 'confirmed' | 'refuted' | 'unclear';
+export type Verdict = 'confirmed' | 'refuted' | 'unclear' | 'unobservable';
 export interface PredictionScope {
     episodes: {
         baseline: string[];
@@ -188,6 +191,7 @@ export interface VerifiedPrediction {
     rates: RateComparison;
     verdict: Verdict;
     scope?: PredictionScope;
+    unobservableReason?: string;
 }
 export interface DiscoveredDifference {
     id: string;
@@ -224,47 +228,7 @@ export interface WhatifReport {
     runDir: string;
     limits: string[];
     headline: string;
+    droppedProbes?: import('./probe-grounding.js').DroppedProbe[];
     corpusExclusions?: import('./episodes.js').CorpusExclusions;
 }
-export type WhatifStage = 'compile' | 'sandbox' | 'snapshot' | 'predict' | 'episodes' | 'run' | 'judge' | 'discover' | 'report';
-export interface WhatifProgress {
-    stage: WhatifStage;
-    message: string;
-    done?: number;
-    total?: number;
-}
-export type CompleteFn = (req: {
-    system: string;
-    user: string;
-    maxTokens: number;
-    model: string;
-    signal?: AbortSignal;
-}) => Promise<{
-    text: string;
-    costUsd: number;
-}>;
-export interface WhatifOptions {
-    spec: ChangeSpec;
-    realHome: string;
-    realCwd: string;
-    agentModel: string;
-    analystModel: string;
-    verify: boolean;
-    turns: number;
-    samples: number;
-    maxUsd: number;
-    judge: 'auto' | 'jev' | 'claude';
-    concurrency: number;
-    maxTurns: number;
-    episodeTimeoutMs: number;
-    keepSandboxes: boolean;
-}
-export interface WhatifDeps {
-    runner: AgentRunner;
-    complete: CompleteFn;
-    makeJudge(choice: WhatifOptions['judge']): Promise<Judge>;
-    makeCrossCheckJudge(): Promise<Judge | undefined>;
-    onProgress?: (p: WhatifProgress) => void;
-    signal?: AbortSignal;
-    now?: () => Date;
-}
+export type { WhatifStage, WhatifProgress, CompleteFn, WhatifOptions, WhatifDeps, } from './types.orchestration.js';

package/dist/whatif/types.orchestration.d.ts

@@ -0,0 +1,43 @@
+import type { AgentRunner, ChangeSpec, Judge } from './types.js';
+export type WhatifStage = 'compile' | 'sandbox' | 'snapshot' | 'predict' | 'episodes' | 'run' | 'judge' | 'discover' | 'report';
+export interface WhatifProgress {
+    stage: WhatifStage;
+    message: string;
+    done?: number;
+    total?: number;
+}
+export type CompleteFn = (req: {
+    system: string;
+    user: string;
+    maxTokens: number;
+    model: string;
+    signal?: AbortSignal;
+}) => Promise<{
+    text: string;
+    costUsd: number;
+}>;
+export interface WhatifOptions {
+    spec: ChangeSpec;
+    realHome: string;
+    realCwd: string;
+    agentModel: string;
+    analystModel: string;
+    verify: boolean;
+    turns: number;
+    samples: number;
+    maxUsd: number;
+    judge: 'auto' | 'jev' | 'claude';
+    concurrency: number;
+    maxTurns: number;
+    episodeTimeoutMs: number;
+    keepSandboxes: boolean;
+}
+export interface WhatifDeps {
+    runner: AgentRunner;
+    complete: CompleteFn;
+    makeJudge(choice: WhatifOptions['judge']): Promise<Judge>;
+    makeCrossCheckJudge(): Promise<Judge | undefined>;
+    onProgress?: (p: WhatifProgress) => void;
+    signal?: AbortSignal;
+    now?: () => Date;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-afk",
-  "version": "5.264.0",
+  "version": "5.265.1",
   "description": "Open-source coding-agent harness you can actually change — own the loop (prompts, gates, routing, skills, terminal states), use any model, run long tasks while you're away.",
   "main": "dist/index.mjs",
   "types": "dist/index.d.ts",