humanish 0.88.0 → 0.88.1

package/README.md

@@ -201,9 +201,19 @@ Mobile viewport and touch flags do not certify gesture equivalence. The
 qualifies the historical phone-lane results: they describe Humanish's measured input path,
 not established physical-device app behavior.
 
-#### Adapters: drive a local app via its JS state contract (no E2B, no vision)
+#### Drive an already-running local app
 
-See [state-driven local adapters](https://humanish.dev/docs/computer-use#state-driven-local-adapters).
+Use a custom executor and non-vision provider to drive your app's state contract
+without E2B. The [runnable npm example](docs/architecture/examples/state-driven-local-app/README.md)
+includes a synthetic loopback app, deterministic provider, verification and cleanup:
+
+```bash
+node node_modules/humanish/docs/architecture/examples/state-driven-local-app/runner.mjs
+```
+
+Run it after installing `humanish`. It makes no model calls; it proves the
+integration, not persona behavior. See [state-driven local adapters](https://humanish.dev/docs/computer-use#state-driven-local-adapters)
+for the supported library seam.
 
 ## Browser Scenario Manifests
 

package/dist/actor-goal-source.d.ts

@@ -0,0 +1,5 @@
+/** Presentation provenance for computer-use completion; never reclassifies the stored outcome. */
+export type CuaGoalSource = "participant_report" | "condition_matched" | "unavailable";
+export declare function isCuaTrace(value: unknown): boolean;
+export declare function cuaGoalSource(value: unknown, recordedStatus?: unknown): CuaGoalSource | undefined;
+export declare const CUA_COMPLETION_NOTE = "Participant reports alone do not establish task success. A matched stop condition establishes only its declared condition. Run gate and share-safety results are separate.";

package/dist/actor-goal-source.js

@@ -0,0 +1,32 @@
+export function isCuaTrace(value) {
+    if (typeof value !== "object" || value === null)
+        return false;
+    const trace = value;
+    return trace.lane === "computer-use" || trace.protocol === "cua-loop";
+}
+export function cuaGoalSource(value, recordedStatus) {
+    if (!isCuaTrace(value))
+        return undefined;
+    const trace = value;
+    if (trace.status !== "passed")
+        return recordedStatus === "passed" || recordedStatus === "complete" ? "unavailable" : undefined;
+    if (trace.lane !== "computer-use" || trace.protocol !== "cua-loop"
+        || trace.completionReason !== "goal_satisfied")
+        return "unavailable";
+    if (!Array.isArray(trace.items) || !trace.items.every((item) => typeof item === "object" && item !== null
+        && !Array.isArray(item) && typeof item.id === "string" && typeof item.kind === "string"
+        && typeof item.title === "string" && (item.lifecycle === "started" || item.lifecycle === "completed"))) {
+        return "unavailable";
+    }
+    // These notices are written by the loop after a matched predicate/window, not by the model.
+    if (trace.items.some((item) => item.kind === "notice"
+        && item.lifecycle === "completed" && item.status === "matched"
+        && typeof item.title === "string"
+        && (item.title.startsWith("stopWhen matched:") || item.title === "dwell window complete"))) {
+        return "condition_matched";
+    }
+    // A legacy trace without enough detail cannot establish which endpoint ended the loop.
+    return trace.items.length > 0 ? "participant_report" : "unavailable";
+}
+export const CUA_COMPLETION_NOTE = "Participant reports alone do not establish task success. A matched stop condition establishes only its declared condition. Run gate and share-safety results are separate.";
+//# sourceMappingURL=actor-goal-source.js.map

package/dist/actor-goal-source.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"actor-goal-source.js","sourceRoot":"","sources":["../src/actor-goal-source.ts"],"names":[],"mappings":"AAGA,MAAM,UAAU,UAAU,CAAC,KAAc;IACvC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,KAAK,CAAC;IAC9D,MAAM,KAAK,GAAG,KAAgC,CAAC;IAC/C,OAAO,KAAK,CAAC,IAAI,KAAK,cAAc,IAAI,KAAK,CAAC,QAAQ,KAAK,UAAU,CAAC;AACxE,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,KAAc,EAAE,cAAwB;IACpE,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IACzC,MAAM,KAAK,GAAG,KAAgC,CAAC;IAC/C,IAAI,KAAK,CAAC,MAAM,KAAK,QAAQ;QAAE,OAAO,cAAc,KAAK,QAAQ,IAAI,cAAc,KAAK,UAAU,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS,CAAC;IAC/H,IAAI,KAAK,CAAC,IAAI,KAAK,cAAc,IAAI,KAAK,CAAC,QAAQ,KAAK,UAAU;WAC7D,KAAK,CAAC,gBAAgB,KAAK,gBAAgB;QAAE,OAAO,aAAa,CAAC;IACvE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,IAAI;WACpG,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,OAAO,IAAI,CAAC,EAAE,KAAK,QAAQ,IAAI,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ;WACpF,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,IAAI,CAAC,IAAI,CAAC,SAAS,KAAK,SAAS,IAAI,IAAI,CAAC,SAAS,KAAK,WAAW,CAAC,CAAC,EAAE,CAAC;QACzG,OAAO,aAAa,CAAC;IACvB,CAAC;IACD,4FAA4F;IAC5F,IAAI,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAA6B,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,QAAQ;WACzE,IAAI,CAAC,SAAS,KAAK,WAAW,IAAI,IAAI,CAAC,MAAM,KAAK,SAAS;WAC3D,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ;WAC9B,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,mBAAmB,CAAC,IAAI,IAAI,CAAC,KAAK,KAAK,uBAAuB,CAAC,CAAC,EAAE,CAAC;QAC7F,OAAO,mBAAmB,CAAC;IAC7B,CAAC;IACD,uFAAuF;IACvF,OAAO,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,aAAa,CAAC;AACvE,CAAC;AAED,MAAM,CAAC,MAAM,mBAAmB,GAAG,4KAA4K,CAAC"}

package/dist/cua-actor-lab.js

@@ -61,7 +61,7 @@ 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";
@@ -4465,7 +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"
@@ -4486,7 +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,
@@ -4922,7 +4922,7 @@ export function buildCuaFanoutBundle(args) {
         const ending = actorEnding(outcome.session.trace);
         return { status: outcome.session.status, ...(ending === undefined ? {} : { label: ending.label }) };
     });
-    const review = {
+    const review = withCuaReviewProvenance({
         schema: REVIEW_SCHEMA,
         verdict,
         ...(participants === undefined ? {} : { participants }),
@@ -4946,7 +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;
@@ -5130,7 +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