humanish 0.87.0 → 0.88.1

package/dist/cua-actor-lab.js

@@ -25,6 +25,7 @@ import { randomBytes } from "node:crypto";
 import { describeMissingKeys } from "./key-resolution.js";
 import { readFile, realpath, rm } from "node:fs/promises";
 import path from "node:path";
+import { cuaLaneDiagnostics, summarizeCuaDiagnostics } from "./cua-diagnostics.js";
 import { feedbackProofCommands } from "./feedback-proof.js";
 import { runDesktopCommandOrThrow, toErrorMessage } from "./command-failure.js";
 import { pathToFileURL } from "node:url";
@@ -56,10 +57,11 @@ import { renderTaskPrompt } from "./tasks.js";
 import { attachObserverRuntimeStreamUrls, renderObserver } from "./observer.js";
 import { containsSensitive, digestText, redactedTail, redactText } from "./redaction.js";
 import { participantAssignment } from "./participant-assignment.js";
+import { actorEnding } from "./actor-stop-cause.js";
 import { assertPreparedSelectedOutputDirectory, assertSafeOutputPathSegment, prepareContainedOutputDirectory, prepareSelectedOutputDirectory, writeContainedOutputFile, writePreparedRunLatestPointer } from "./selected-output-paths.js";
 import { prepareRunArtifactPaths, validatePreparedRunArtifactPaths } from "./run-paths.js";
 import { createLocalTreeArchive } from "./source-archive.js";
-import { buildRunSource, loadRunBundle, PUBLIC_TARGET_CWD, REVIEW_SCHEMA, RUN_BUNDLE_SCHEMA, aggregateTaskFunnels, formatParticipantOutcomes, formatStudyTaskFunnel, tallyParticipantOutcomes } from "./run.js";
+import { buildRunSource, loadRunBundle, PUBLIC_TARGET_CWD, REVIEW_SCHEMA, RUN_BUNDLE_SCHEMA, aggregateTaskFunnels, formatParticipantOutcomes, formatStudyTaskFunnel, tallyParticipantOutcomes, withCuaReviewProvenance } from "./run.js";
 import { estimateActorCost, estimateDesktopCost, estimateAllocatedDesktopCost, MODEL_RATES, round6 } from "./pricing.js";
 import { observeDesktopResources } from "./e2b-desktop-resources.js";
 export const CUA_ACTOR_LAB_SCHEMA = "humanish.cua-lab-result.v2";
@@ -2539,13 +2541,14 @@ function toLaneResult(spec, outcome, subject, dryRun) {
         subject
     };
     if (!outcome || dryRun) {
-        return { ...base, status: "contract_proof_only", ok: dryRun };
+        return { ...base, status: "contract_proof_only", ok: dryRun, diagnostics: cuaLaneDiagnostics({ dryRun }) };
     }
     if (outcome.skippedReason !== undefined) {
         return {
             ...base,
             status: "blocked",
             ok: false,
+            diagnostics: cuaLaneDiagnostics({ dryRun, skipped: true }),
             skippedReason: outcome.skippedReason,
             error: { code: "HUMANISH_CUA_LAB_FAILED", message: outcome.skippedReason }
         };
@@ -2557,11 +2560,16 @@ function toLaneResult(spec, outcome, subject, dryRun) {
         ...base,
         status,
         ok: laneOk,
+        diagnostics: cuaLaneDiagnostics({
+            dryRun, executionError: outcome.sessionError !== undefined, noEngagement: outcome.noEngagement,
+            ...(session ? { session: { status: session.status, completionReason: session.completionReason, ...(session.trace.stopCause === undefined ? {} : { stopCause: session.trace.stopCause }) } } : {})
+        }),
         ...(session
             ? {
                 session: {
                     status: session.status,
                     completionReason: session.completionReason,
+                    ...(session.trace.stopCause === undefined ? {} : { stopCause: session.trace.stopCause }),
                     reason: session.reason,
                     screenshots: outcome.screenshots.length
                 }
@@ -3487,6 +3495,7 @@ async function runCuaActorLabInScope(options) {
                 session: {
                     status: firstOutcome.session.status,
                     completionReason: firstOutcome.session.completionReason,
+                    ...(firstOutcome.session.trace.stopCause === undefined ? {} : { stopCause: firstOutcome.session.trace.stopCause }),
                     reason: firstOutcome.session.reason,
                     screenshots: firstOutcome.screenshots.length
                 }
@@ -3498,6 +3507,7 @@ async function runCuaActorLabInScope(options) {
         subject: aggregateSubject,
         plan,
         lanes: laneResults,
+        diagnostics: summarizeCuaDiagnostics({ dryRun, evidenceInvalid: !observer.ok, lanes: laneResults }),
         laneSummary,
         ...(rerunLineage === undefined ? {} : { rerun: rerunLineage }),
         observer,
@@ -4031,8 +4041,19 @@ export function buildCuaCostSummary(args) {
             sumInput += usage.input ?? 0;
             sumOutput += usage.output ?? 0;
         }
-        // An attempted closing request can fail after provider work without reporting usage.
-        // Keep the known interaction estimate and make the additional unknown explicit.
+        // A stalled/ambiguous interaction can remain unreported after a later successful retry.
+        // Keep known token estimates and make the additional unknown explicit.
+        if (lane.trace.interactionUsageIncomplete === true) {
+            breakdown.push({
+                kind: "model-tokens",
+                ...(lane.laneId === undefined ? {} : { laneId: lane.laneId }),
+                ...(lane.trace.providerVersion === undefined ? {} : { modelId: lane.trace.providerVersion }),
+                estimatedCostUsd: null,
+                reason: "interaction_usage_unreported",
+                ratesAsOf: null
+            });
+        }
+        // An attempted closing request has its own accounting boundary.
         if (lane.trace.debrief?.usageReported === false) {
             breakdown.push({
                 kind: "model-tokens",
@@ -4444,7 +4465,7 @@ export function buildCuaBundle(args) {
         : args.credibility?.noEngagement === true
             ? "Not counted as a pass: the participant took no actions and said nothing."
             : "Not counted as a pass: the participant's final message described a blocker.";
-    const review = {
+    const review = withCuaReviewProvenance({
         schema: REVIEW_SCHEMA,
         verdict: args.inProgress === true
             ? "contract_proof_only"
@@ -4465,7 +4486,7 @@ export function buildCuaBundle(args) {
             : args.inProgress === true
                 ? ["Live desktop session is still running."]
                 : ["Live desktop session not yet run (dry-run contract only)."]
-    };
+    }, [stream]);
     return {
         schema: RUN_BUNDLE_SCHEMA,
         runId: args.runId,
@@ -4897,7 +4918,11 @@ export function buildCuaFanoutBundle(args) {
         .map((outcome) => outcome?.session?.trace.taskFunnel)
         .filter((funnel) => funnel !== undefined);
     const studyTasks = args.inProgress === true ? undefined : aggregateTaskFunnels(participantFunnels);
-    const review = {
+    const participantEndings = terminalOutcomes.map((outcome) => {
+        const ending = actorEnding(outcome.session.trace);
+        return { status: outcome.session.status, ...(ending === undefined ? {} : { label: ending.label }) };
+    });
+    const review = withCuaReviewProvenance({
         schema: REVIEW_SCHEMA,
         verdict,
         ...(participants === undefined ? {} : { participants }),
@@ -4906,7 +4931,7 @@ export function buildCuaFanoutBundle(args) {
             ? `Live computer-use fan-out is running (${specs.length} per-lane worlds); terminal lane evidence has not been written yet.`
             : args.dryRun
                 ? `${args.rerun ? `Rerun contract from ${args.rerun.sourceRunId}: ` : ""}Dry-run fan-out contract: ${specs.length} per-lane-world lanes composed for ${args.descriptor.id} against ${args.appUrl}; no desktops launched, $0 spend.`
-                : `${args.rerun ? `Rerun from ${args.rerun.sourceRunId}: ` : ""}Computer-use fan-out (${specs.length} per-lane worlds): ${passedLanes}/${specs.length} lane(s) reached a terminal, engaged verdict${participants ? ` — ${formatParticipantOutcomes(participants)}` : ""}${studyTasks ? `; tasks: ${formatStudyTaskFunnel(studyTasks)}` : ""}.`,
+                : `${args.rerun ? `Rerun from ${args.rerun.sourceRunId}: ` : ""}Computer-use fan-out (${specs.length} per-lane worlds): ${passedLanes}/${specs.length} lane(s) reached a terminal, engaged verdict${participants ? ` — ${formatParticipantOutcomes(participants, participantEndings)}` : ""}${studyTasks ? `; tasks: ${formatStudyTaskFunnel(studyTasks)}` : ""}.`,
         gaps: args.inProgress === true
             ? ["Live fan-out session is still running."]
             : args.dryRun
@@ -4921,7 +4946,7 @@ export function buildCuaFanoutBundle(args) {
                     || outcome.session === undefined
                     || outcome.session.status !== "passed")
                     .map(({ spec, outcome }) => `${spec.laneId}: ${outcome?.skippedReason ?? outcome?.sessionError ?? outcome?.session?.reason ?? "did not pass"}`)
-    };
+    }, streams);
     const anyRaw = (outcomes ?? []).some((outcome) => outcome.session?.trace.redaction.screenshots === "raw");
     const ranLive = (outcomes ?? []).some((outcome) => outcome.session !== undefined || outcome.sessionError !== undefined);
     const configuredBrowser = config.execution?.desktop?.browser;
@@ -5105,7 +5130,7 @@ function renderCuaReviewMarkdown(bundle) {
         "",
         `- run: ${bundle.runId}`,
         `- mode: ${bundle.mode}`,
-        `- verdict: ${bundle.review.verdict}`,
+        `- run gate: ${bundle.review.verdict}`,
         `- summary: ${bundle.review.summary}`,
         ...(provenance ? [`- subject: ${provenance.message}`] : []),
         ...(trace