humanish 0.107.0 → 0.109.0

package/dist/comms/types.js

@@ -1,8 +1,8 @@
-// The addressed message bus for captured mail (#297): the email or SMS a persona receives off the
+// The addressed message bus for captured mail: the email or SMS a persona receives off the
 // app, as a testable surface. One port, addressed by participant, so the inbox surface and the evidence
 // writer read messages without knowing how they were captured.
 //
-// PUBLIC-SAFETY: raw address values, message bodies, links, and codes are RUNTIME-ONLY. Only the
-// address DIGEST (sha256-short, via redaction.digestText) is ever meant to reach a persisted bundle;
+// Public safety: raw address values, message bodies, links, and codes are runtime-only. Only the
+// address digest (sha256-short, via redaction.digestText) is ever meant to reach a persisted bundle;
 // a verification link / OTP has "no secret shape" (like the lobby code) → literal-scrub + digest.
 export {};

package/dist/evidence/frame-signature.d.ts

@@ -1,6 +1,6 @@
 /**
  * A perceptual hash of a PNG frame for no-progress detection. Decodes the PNG, area-averages it to
- * a SIGNATURE_GRID x SIGNATURE_GRID grayscale grid, CONTRAST-NORMALIZES that grid, quantizes each
+ * a SIGNATURE_GRID x SIGNATURE_GRID grayscale grid, contrast-normalizes that grid, quantizes each
  * cell to SIGNATURE_LEVELS, and packs the cells into a hex string. Deterministic (no Date, no
  * random). Returns SIGNATURE_FALLBACK on any decode failure so two unreadable frames compare equal.
  * The hash cannot be reversed to the image, so exposing it in a trace is public-safe.

package/dist/evidence/frame-signature.js

@@ -6,7 +6,7 @@ const SIGNATURE_FALLBACK = "unreadable";
 const ANCHOR_LEVELS = 16;
 /**
  * A perceptual hash of a PNG frame for no-progress detection. Decodes the PNG, area-averages it to
- * a SIGNATURE_GRID x SIGNATURE_GRID grayscale grid, CONTRAST-NORMALIZES that grid, quantizes each
+ * a SIGNATURE_GRID x SIGNATURE_GRID grayscale grid, contrast-normalizes that grid, quantizes each
  * cell to SIGNATURE_LEVELS, and packs the cells into a hex string. Deterministic (no Date, no
  * random). Returns SIGNATURE_FALLBACK on any decode failure so two unreadable frames compare equal.
  * The hash cannot be reversed to the image, so exposing it in a trace is public-safe.
@@ -65,9 +65,9 @@ function quantizedGrid(data, srcW, srcH) {
         }
     }
     // Stretch this frame's own range across the full scale before quantizing. Without this a light UI
-    // occupies a sliver at the top of 0..255 and quantizes to a near-constant (#383).
+    // occupies a sliver at the top of 0..255 and quantizes to a near-constant.
     //
-    // The endpoints are TRIMMED rather than the raw min/max: a handful of extreme cells must not be
+    // The endpoints are trimmed rather than the raw min/max: a handful of extreme cells must not be
     // able to rescale the whole frame, or a blinking text cursor would rewrite every cell and read as
     // progress forever. Trimming K cells from each end clamps that away while still preserving real
     // structure, which occupies far more cells than K (a row of sidebar text covers dozens).
@@ -81,7 +81,7 @@ function quantizedGrid(data, srcW, srcH) {
 }
 /**
  * The frame's overall brightness, coarsely quantized. Normalization deliberately discards absolute
- * level, so this is prefixed back on for the two cases where absolute level IS the change: a uniform
+ * level, so this is prefixed back on for the two cases where absolute level is the change: a uniform
  * frame (an all-white blank vs an all-black screen normalize identically), and a whole-page dim such
  * as a modal overlay, which preserves relative structure while changing the whole frame. Quantized
  * coarsely so ordinary antialiasing noise cannot move it.

package/dist/evidence/redaction.d.ts

@@ -35,7 +35,7 @@ export interface RedactedScreenshot {
 export interface RedactScreenshotOptions {
     /**
      * Longest emitted edge in pixels. Clamped to [1, 128]: a caller may request a
-     * SMALLER (safer) thumbnail but never a larger one, so no call site can widen
+     * smaller (safer) thumbnail but never a larger one, so no call site can widen
      * the frame back into legibility. Default 96. Blur is not a caller knob: it is
      * an internal safety floor computed from the output size.
      */
@@ -52,7 +52,7 @@ export declare function digestText(text: string, length?: number): string;
 export declare function tailText(value: string, maxChars: number): string;
 /**
  * A redacted, ellipsis-prefixed tail of captured output for a (public-bound)
- * message field. Pattern-redacts the FULL text BEFORE truncating: slicing a tail
+ * message field. Pattern-redacts the full text before truncating: slicing a tail
  * first could cut through a secret's prefix (e.g. drop "sk-proj-") and defeat the
  * pattern matcher on the remainder. Callers should literal-scrub known
  * provisioned values first; this is the pattern pass. Empty output -> "(no output)".
@@ -103,7 +103,7 @@ export declare function scrubLiterals(values: readonly string[]): (text: string)
 /**
  * Coerce an unknown thrown value to its message string. Prefer this over the
  * inline `error instanceof Error ? error.message : String(error)` so error
- * stringification stays uniform. Note: this does NOT redact -- sites that emit
+ * stringification stays uniform. Note: this does not redact; sites that emit
  * to public-bound artifacts must run the result through `redactText` first.
  */
 export declare function toErrorMessage(error: unknown): string;

package/dist/evidence/redaction.js

@@ -105,9 +105,9 @@ export function publicPathForTrace(value, rootCwd) {
 // or a logged-in third-party UI. A raw frame must never reach a public artifact.
 // redactScreenshot is the fail-closed primitive that gates that surface. It
 // always returns a freshly re-encoded, downscaled, box-blurred thumbnail (or a
-// neutral placeholder), and NEVER the source pixels. The "too coarse to read"
+// neutral placeholder), and never the source pixels. The "too coarse to read"
 // invariant is enforced in code, not by defaults: the emitted width is hard-
-// capped (a caller may request a SMALLER thumbnail but never a larger one), and
+// capped (a caller may request a smaller thumbnail but never a larger one), and
 // the blur radius is computed internally with a floor, so neither a caller
 // option nor a natively small frame can widen the output back into legibility.
 // On any uncertainty (non-PNG input, an oversized or unreadable PNG, or any
@@ -241,7 +241,7 @@ export function tailText(value, maxChars) {
 }
 /**
  * A redacted, ellipsis-prefixed tail of captured output for a (public-bound)
- * message field. Pattern-redacts the FULL text BEFORE truncating: slicing a tail
+ * message field. Pattern-redacts the full text before truncating: slicing a tail
  * first could cut through a secret's prefix (e.g. drop "sk-proj-") and defeat the
  * pattern matcher on the remainder. Callers should literal-scrub known
  * provisioned values first; this is the pattern pass. Empty output -> "(no output)".
@@ -321,7 +321,7 @@ export function scrubLiterals(values) {
 /**
  * Coerce an unknown thrown value to its message string. Prefer this over the
  * inline `error instanceof Error ? error.message : String(error)` so error
- * stringification stays uniform. Note: this does NOT redact -- sites that emit
+ * stringification stays uniform. Note: this does not redact; sites that emit
  * to public-bound artifacts must run the result through `redactText` first.
  */
 export function toErrorMessage(error) {

package/dist/feedback/draft.js

@@ -49,98 +49,58 @@ export function buildDraft(bundle, bundlePath, candidateId) {
             acceptance_proof: projectFeedbackAcceptanceProof(bundle, candidate),
         };
     }
-    // #392: the fallback below is DRY-RUN-shaped, and it used to be the fallback for every bundle
-    // without a candidate — so a live run whose route filed no participant report got a draft
-    // claiming "no browser behavior was exercised" over 15 screenshots of browser behavior. A live
-    // bundle now gets a draft that describes the run that happened, built from the same review lines
-    // the stakeholder surfaces show (participants and tasks keep their denominators).
-    if (bundle.mode === "live") {
-        const participantEndings = participantOutcomeDetails(bundle.streams);
-        const review = withCuaReviewProvenance(bundle.review, bundle.streams);
-        const actualLines = [
-            review.summary,
-            ...(bundle.review.participants === undefined
-                ? []
-                : [
-                    `Participants: ${formatParticipantOutcomes(bundle.review.participants, participantEndings)}.`,
-                ]),
-            ...(bundle.review.tasks === undefined
-                ? []
-                : [`Tasks: ${formatRunTaskFunnel(bundle.review.tasks)}.`]),
-        ];
-        return {
-            schema: FEEDBACK_SCHEMA,
-            run_id: bundle.runId,
-            adapter_id: bundle.source.packageName ?? bundle.scenario.id,
-            scenario_id: bundle.scenario.id,
-            persona_id: bundle.persona.id,
-            actor: (bundle.streams ?? []).some((stream) => stream.actor?.lane === "computer-use")
-                ? "computer-use"
-                : "unknown",
-            substrate: "unknown",
-            failure_owner: "unknown",
-            summary: "Live study completed without a participant-reported finding",
-            expected: bundle.scenario.goal,
-            actual: actualLines.join(" "),
-            source_bundle: bundlePath,
-            evidence: [
-                {
-                    path: bundlePath,
-                    kind: "state",
-                    note: "Source run bundle.",
-                },
-                {
-                    path: path.join(path.dirname(bundlePath), "review.md"),
-                    kind: "review",
-                    note: "The run's review: verdict, participants, and gaps.",
-                },
-            ],
-            redaction: {
-                status: "passed",
-                notes: bundle.redaction.notes,
-            },
-            idempotency_key: `humanish:${bundle.runId}:live-run-summary`,
-            proposed_next_state: "study-quality-review",
-            acceptance_proof: [
-                feedbackProofCommands(bundle.runId).verify,
-                feedbackProofCommands(bundle.runId).watch,
-            ],
-        };
-    }
+    // A bundle without a candidate gets a draft that describes the run that happened, built from the
+    // same review lines the stakeholder surfaces show (participants and tasks keep their
+    // denominators). Feedback refuses a dry run before it gets here.
+    const participantEndings = participantOutcomeDetails(bundle.streams);
+    const review = withCuaReviewProvenance(bundle.review, bundle.streams);
+    const actualLines = [
+        review.summary,
+        ...(bundle.review.participants === undefined
+            ? []
+            : [
+                `Participants: ${formatParticipantOutcomes(bundle.review.participants, participantEndings)}.`,
+            ]),
+        ...(bundle.review.tasks === undefined
+            ? []
+            : [`Tasks: ${formatRunTaskFunnel(bundle.review.tasks)}.`]),
+    ];
     return {
         schema: FEEDBACK_SCHEMA,
         run_id: bundle.runId,
-        adapter_id: bundle.source.packageName ?? "synthetic-app",
+        adapter_id: bundle.source.packageName ?? bundle.scenario.id,
         scenario_id: bundle.scenario.id,
         persona_id: bundle.persona.id,
-        actor: "synthetic-dry-run",
-        substrate: "local-filesystem",
-        failure_owner: "harness",
-        summary: "Dry-run contract proof needs product-evidence follow-up",
-        expected: "humanish should produce verified, public-safe evidence before product claims are filed.",
-        actual: "This dry-run produced a contract-proof bundle only; no browser or product behavior was exercised.",
+        actor: (bundle.streams ?? []).some((stream) => stream.actor?.lane === "computer-use")
+            ? "computer-use"
+            : "unknown",
+        substrate: "unknown",
+        failure_owner: "unknown",
+        summary: "Live study completed without a participant-reported finding",
+        expected: bundle.scenario.goal,
+        actual: actualLines.join(" "),
         source_bundle: bundlePath,
         evidence: [
             {
                 path: bundlePath,
                 kind: "state",
-                note: "Synthetic run bundle.",
+                note: "Source run bundle.",
             },
             {
                 path: path.join(path.dirname(bundlePath), "review.md"),
                 kind: "review",
-                note: "Review skeleton labels this as contract proof only.",
+                note: "The run's review: verdict, participants, and gaps.",
             },
         ],
         redaction: {
             status: "passed",
             notes: bundle.redaction.notes,
         },
-        idempotency_key: `humanish:${bundle.runId}:dry-run-contract-proof`,
-        proposed_next_state: "watch",
+        idempotency_key: `humanish:${bundle.runId}:live-run-summary`,
+        proposed_next_state: "study-quality-review",
         acceptance_proof: [
             feedbackProofCommands(bundle.runId).verify,
-            feedbackProofCommands(bundle.runId).watch,
+            feedbackProofCommands(bundle.runId).observe,
         ],
     };
 }
@@ -232,7 +192,7 @@ export async function buildAnalysisDraft(context, options) {
         proposed_next_state: "study-quality-review",
         acceptance_proof: [
             feedbackProofCommands(bundle.runId).verify,
-            feedbackProofCommands(bundle.runId).watch,
+            feedbackProofCommands(bundle.runId).observe,
         ],
     };
 }
@@ -254,11 +214,14 @@ export function isUsableFeedbackCandidate(candidate) {
         !item.path.includes("://") &&
         !item.path.includes(".."));
 }
+/** An evidence path inside the run directory, as "run.json" or "screenshots/step-003.png". */
+function runRelative(draft, evidencePath) {
+    const relative = path.relative(path.dirname(draft.source_bundle), evidencePath);
+    return relative === "" || relative.startsWith("..") ? evidencePath : relative;
+}
 export function renderMarkdown(draft, repo) {
     return `This issue was drafted by humanish from a verified humanish run bundle.
 
-It contributes to public-safe simulation harness coverage. The feedback command did not mutate GitHub, commit code, or claim unobserved product behavior.
-
 ## Summary
 
 ${draft.summary}
@@ -273,14 +236,9 @@ ${draft.actual}
 
 ## Evidence
 
-${draft.evidence.map((item) => `- ${item.kind}: \`${item.path}\` - ${item.note}`).join("\n")}
-
-## Filing Notes
+${draft.evidence.map((item) => `- ${item.kind} ${runRelative(draft, item.path)}: ${item.note}`).join("\n")}
 
-- Repository: ${repo}
-- GitHub mutation: not performed
-- Substrate: ${draft.substrate}
-- Production data: not used
+Repository: ${repo}
 
 \`\`\`yaml
 humanish_feedback:

package/dist/feedback/export-bundle.js

@@ -1,4 +1,4 @@
-// The publish boundary for existing raw runs (#136). A derivative is an isolated
+// The publish boundary for existing raw runs. A derivative is an isolated
 // workspace, never a new run in the source history. No provider APIs are involved.
 import { createHash } from "node:crypto";
 import { lstat, mkdir, mkdtemp, readdir, realpath, rename, rm, rmdir } from "node:fs/promises";

package/dist/feedback/export.d.ts

@@ -46,5 +46,9 @@ export interface ExportDeps {
 /** The banner a --local-only export carries. Plain HTML, before the app root, so it renders with JS off. */
 export declare function localOnlyBanner(reasons: string[]): string;
 export declare function exportRun(cwdInput: string, runInput: string, options?: ExportOptions, deps?: ExportDeps): Promise<ExportResult | ExportFailure>;
-export declare function formatExportHuman(result: ExportResult | ExportFailure): string;
+/** Human output; a failure is an error the CLI prints on stderr (HumanOutput in src/cli/io.ts). */
+export declare function formatExportHuman(result: ExportResult): string;
+export declare function formatExportHuman(result: ExportResult | ExportFailure): string | {
+    error: ExportFailure["error"];
+};
 export {};

package/dist/feedback/export.js

@@ -1,4 +1,4 @@
-// `humanish export`: one file a coworker can open (#471).
+// `humanish export`: one file a coworker can open.
 //
 // A run is a directory. Sharing it meant a tunnel (`serve --expose`) or a hand-zipped bundle,
 // neither of which is "send one thing". The Observer is already a single-file artifact with the
@@ -23,6 +23,7 @@ import { analysisSharingProblems } from "../analysis/sharing.js";
 import { EVIDENCE_LIMITS, validateAnalysisEvidence } from "../analysis/evidence.js";
 import { readBoundedFile } from "../run/evidence-files.js";
 import { shellQuote } from "../substrates/shell.js";
+import { plural } from "../run/text.js";
 const EXPORT_SCHEMA = "humanish.export-result.v1";
 /** Past this the file stops being a thing you attach to an email. Declared, never silent. */
 export const DEFAULT_EXPORT_MAX_BYTES = 25 * 1024 * 1024;
@@ -40,7 +41,7 @@ const IMAGE_MIME = {
 export function localOnlyBanner(reasons) {
     const why = reasons.length === 0 ? "" : ` (${reasons.join(", ")})`;
     return (`<div id="humanish-local-only" role="alert" tabindex="0" style="position:sticky;top:0;z-index:2147483647;max-height:40vh;overflow:auto;background:#7a1f1f;color:#fff;font:600 14px/1.4 system-ui,sans-serif;padding:10px 16px;text-align:center">` +
-        `LOCAL ONLY. This export was made from a bundle that is not share-safe${why}. Do not forward it outside the team that owns the run.` +
+        `Local only. This export was made from a bundle that is not share-safe${why}. Do not forward it outside the team that owns the run.` +
         `</div>`);
 }
 async function isInside(root, candidate) {
@@ -146,9 +147,8 @@ async function readObserverSlot(cwd, runInput, runRoot, runId, verified, deps, w
         html = await readFile(observerPath, "utf8");
     }
     catch {
-        // `run` writes observer-data.json and no index.html; `watch` writes both (#597). The
-        // 0.72.0 dogfood participant hit this on its first export. Render it here, from the same
-        // artifact watch uses, so what produced the run never decides whether it can be sent.
+        // `run` writes observer-data.json and no index.html; `watch` writes both. Render it here, from
+        // the same artifact watch uses, so what produced the run never decides whether it can be sent.
         const rendered = await (deps.render ?? ((c, r) => renderObserver(c, r, { open: false })))(cwd, runInput).catch(() => ({ ok: false }));
         try {
             if (!rendered.ok)
@@ -281,7 +281,7 @@ function makePortable(inlined, verified) {
             }
         }
     }
-    // What verify said, in the file, so the chrome can agree with the result envelope (#584).
+    // What verify said, in the file, so the chrome can agree with the result envelope.
     const publicSafety = (inlined.publicSafety ?? {});
     inlined.publicSafety = {
         ...publicSafety,
@@ -358,11 +358,11 @@ function renderExportHtml(inlined, analysis, assets, verified, watermarked, omit
 }
 export function formatExportHuman(result) {
     if (!result.ok)
-        return `${result.error.code}: ${result.error.message}\n`;
+        return { error: result.error };
     if (result.format === "bundle") {
         return [
             `humanish export ${result.runId}`,
-            `workspace: ${result.path} (${(result.bytes / 1024).toFixed(0)} KB, ${result.embeddedImages} blurred image(s))`,
+            `workspace: ${result.path} (${(result.bytes / 1024).toFixed(0)} KB, ${plural(result.embeddedImages, "blurred image")})`,
             `share safety: ${result.shareSafety.status}`,
             `verify: humanish verify --cwd ${shellQuote(result.path)} --run ${shellQuote(result.runId)}`,
             `feedback: humanish feedback draft --cwd ${shellQuote(result.path)} --run ${shellQuote(result.runId)}`,
@@ -372,8 +372,8 @@ export function formatExportHuman(result) {
     }
     const lines = [
         `humanish export ${result.runId}`,
-        `file: ${result.path} (${(result.bytes / 1024).toFixed(0)} KB, ${result.embeddedImages} image(s) embedded)`,
-        `share safety: ${result.shareSafety.status}${result.watermarked ? " — WATERMARKED LOCAL ONLY" : ""}`,
+        `file: ${result.path} (${(result.bytes / 1024).toFixed(0)} KB, ${plural(result.embeddedImages, "image")} embedded)`,
+        `share safety: ${result.shareSafety.status}${result.watermarked ? " (watermarked: local only)" : ""}`,
         ...result.warnings.map((warning) => `warning: ${warning}`),
     ];
     return `${lines.join("\n")}\n`;

package/dist/feedback/feedback.d.ts

@@ -13,11 +13,11 @@ export interface FeedbackResult {
     issueUrl?: string;
     draft?: FeedbackDraft;
     /** Every usable candidate on the bundle, so a multi-participant study's second and third findings can be
-     *  chosen with `--candidate` instead of being invisible behind the first (#609). */
+     *  chosen with `--candidate` instead of being invisible behind the first. */
     candidates?: FeedbackCandidateSummary[];
     shareSafety?: VerifyResult["shareSafety"];
     error?: {
-        code: "HUMANISH_RUN_NOT_FOUND" | "HUMANISH_INVALID_RUN_BUNDLE" | "HUMANISH_INVALID_FEEDBACK_DRAFT" | "HUMANISH_FEEDBACK_SHARE_SAFETY_BLOCKED" | "HUMANISH_FEEDBACK_CANDIDATE_NOT_FOUND";
+        code: "HUMANISH_RUN_NOT_FOUND" | "HUMANISH_INVALID_RUN_BUNDLE" | "HUMANISH_INVALID_FEEDBACK_DRAFT" | "HUMANISH_FEEDBACK_SHARE_SAFETY_BLOCKED" | "HUMANISH_FEEDBACK_CANDIDATE_NOT_FOUND" | "HUMANISH_FEEDBACK_REQUIRES_LIVE_RUN";
         message: string;
     };
 }

package/dist/feedback/feedback.js

@@ -1,6 +1,7 @@
 // The feedback commands: resolve a run, draft a feedback issue from it, verify the draft, render it
 // as an issue body or URL, and list a run's candidates. What a draft contains is in draft.ts.
 import { realpath } from "node:fs/promises";
+import { runNotFoundMessage } from "../run/run-not-found.js";
 import path from "node:path";
 import { bindExistingRunArtifactPaths, isSafeRunIdSegment, resolveLatestRunDirectory, validatePreparedRunArtifactPaths, } from "../run/paths.js";
 import { bindExistingManagedHumanishOutputDirectory, readContainedRegularFile, writeContainedOutputFile, } from "../run/contained-output.js";
@@ -34,7 +35,24 @@ async function draftFeedbackBound(cwdInput, runInput, options = {}) {
                 run: runInput,
                 error: {
                     code: "HUMANISH_RUN_NOT_FOUND",
-                    message: `Run not found: ${runInput}`,
+                    message: await runNotFoundMessage(cwd, runInput),
+                },
+            },
+        };
+    }
+    // A dry run exercised no product, so it has nothing to report upstream; analyze refuses it the
+    // same way (ANALYSIS_REQUIRES_LIVE_RUN).
+    if (context.loaded.bundle.mode !== "live") {
+        return {
+            context,
+            result: {
+                schema: FEEDBACK_RESULT_SCHEMA,
+                ok: false,
+                cwd,
+                run: runInput,
+                error: {
+                    code: "HUMANISH_FEEDBACK_REQUIRES_LIVE_RUN",
+                    message: `Run ${context.storedRunId} is a dry run, which tests no product behavior. Feedback drafts need a live run.`,
                 },
             },
         };
@@ -204,7 +222,7 @@ export async function listFeedback(cwdInput, runInput) {
             run: runInput,
             error: {
                 code: "HUMANISH_RUN_NOT_FOUND",
-                message: `Run not found: ${runInput}`,
+                message: await runNotFoundMessage(cwd, runInput),
             },
         };
     }
@@ -218,7 +236,7 @@ export async function listFeedback(cwdInput, runInput) {
         run: runInput,
         ...(draft ? { draftPath, draft } : {}),
         // The choice set for `draft --candidate`: a three-participant study has up to three findings,
-        // and until #609 only the first ever reached a draft.
+        // and each can reach a draft.
         candidates: summarizeCandidates(context.loaded.bundle),
     };
 }

package/dist/feedback/proof.d.ts

@@ -2,7 +2,7 @@ import type { RunBundle, RunFeedbackCandidate } from "../run/bundle.js";
 /** Commands run from the evidence workspace using an installed humanish CLI. */
 export declare function feedbackProofCommands(runId: string): {
     verify: string;
-    watch: string;
+    observe: string;
 };
 /** Legacy command compatibility belongs in the draft, never in the source receipt. */
 export declare function projectFeedbackAcceptanceProof(bundle: RunBundle, candidate: RunFeedbackCandidate): string[];

package/dist/feedback/proof.js

@@ -5,7 +5,7 @@ export function feedbackProofCommands(runId) {
     const argument = /^[a-z0-9][a-z0-9._-]*$/i.test(runId) ? runId : shellQuote(runId);
     return {
         verify: `humanish verify --run ${argument} --json`,
-        watch: `humanish watch --run ${argument} --no-open`,
+        observe: `humanish observe --run ${argument} --no-open`,
     };
 }
 /** Legacy command compatibility belongs in the draft, never in the source receipt. */
@@ -17,7 +17,7 @@ export function projectFeedbackAcceptanceProof(bundle, candidate) {
         if (instruction === `pnpm humanish -- verify --run ${bundle.runId} --json`)
             return commands.verify;
         if (instruction === `pnpm humanish -- watch --run ${bundle.runId} --no-open`)
-            return commands.watch;
+            return commands.observe;
         return instruction;
     });
 }

package/dist/guest/bootstrap.js

@@ -1,6 +1,6 @@
 import { desktopRecordingConfigSchema, } from "../evidence/desktop-recording-types.js";
 import { validateBrowserControlIdentity, sameBrowserControlIdentity, } from "../browser-control/protocol.js";
-import { CuaExecutorError } from "../actors/computer-use/executor-error.js";
+import { ComputerUseExecutorError } from "../actors/computer-use/executor-error.js";
 import { createBrowserControlClient } from "../browser-control/client.js";
 import { guestMediaConfigSchema } from "./media-config.js";
 export const GUEST_BOOTSTRAP_VERSION = 1;
@@ -15,7 +15,7 @@ export const GUEST_BOOTSTRAP_LIMITS = Object.freeze({
     connectMs: 3000,
     port: 5251,
 });
-const refused = () => new CuaExecutorError("protocol_mismatch", "not_dispatched");
+const refused = () => new ComputerUseExecutorError("protocol_mismatch", "not_dispatched");
 /** Same entry authority as the local desktop's opaque app-port forward. */
 export function validateGuestInitialUrl(value) {
     try {
@@ -223,7 +223,7 @@ export class GuestBootstrapReader {
             return;
         this.terminal = true;
         this.release();
-        this.reject(new CuaExecutorError(code, "not_dispatched"));
+        this.reject(new ComputerUseExecutorError(code, "not_dispatched"));
         this.stream.destroy();
         this.onFailure();
     }
@@ -249,7 +249,7 @@ export async function connectGuestBootstrap(stream, identity, signal, initialUrl
     const request = encodeGuestBootstrap(identity, false, initialUrl, media, recording);
     await new Promise((resolve, reject) => {
         let bytes = Buffer.alloc(0), done = false;
-        const timer = setTimeout(() => finish(new CuaExecutorError("deadline_exceeded", "not_dispatched")), GUEST_BOOTSTRAP_LIMITS.connectMs);
+        const timer = setTimeout(() => finish(new ComputerUseExecutorError("deadline_exceeded", "not_dispatched")), GUEST_BOOTSTRAP_LIMITS.connectMs);
         const finish = (error) => {
             if (done)
                 return;
@@ -277,9 +277,9 @@ export async function connectGuestBootstrap(stream, identity, signal, initialUrl
         const end = () => {
             if (done)
                 stream.destroy();
-            finish(new CuaExecutorError("transport_failed", "not_dispatched"));
+            finish(new ComputerUseExecutorError("transport_failed", "not_dispatched"));
         };
-        const abort = () => finish(new CuaExecutorError("session_revoked", "not_dispatched"));
+        const abort = () => finish(new ComputerUseExecutorError("session_revoked", "not_dispatched"));
         const data = (chunk) => {
             if (!Buffer.isBuffer(chunk) || bytes.length + chunk.length > 64 + 1028) {
                 finish(refused());

package/dist/guest/browser-tools.js

@@ -1,5 +1,5 @@
 import { BROWSER_CONTROL_LIMITS } from "../browser-control/protocol.js";
-import { CuaExecutorError, isCuaExecutorError } from "../actors/computer-use/executor-error.js";
+import { ComputerUseExecutorError, isComputerUseExecutorError, } from "../actors/computer-use/executor-error.js";
 /** Owner-only composition for a single headed Chromium page. Never an actor API. */
 export function createGuestBrowserTools(native, content) {
     let addressBarArmed = false;
@@ -11,7 +11,7 @@ export function createGuestBrowserTools(native, content) {
             const inputGeneration = ++generation;
             await native.input(args, signal);
             if (signal.aborted || inputGeneration !== generation)
-                throw new CuaExecutorError("session_revoked", "outcome_uncertain");
+                throw new ComputerUseExecutorError("session_revoked", "outcome_uncertain");
             // Positive navigation intent, not an inference from document.hasFocus().
             addressBarArmed =
                 args.length === 3 &&
@@ -28,15 +28,15 @@ export function createGuestBrowserTools(native, content) {
             // separate qualification. Do not silently switch routes or transliterate.
             if (!/^[\x20-\x7e]+$/.test(text) ||
                 Buffer.byteLength(text) > BROWSER_CONTROL_LIMITS.textBytes) {
-                throw new CuaExecutorError("action_rejected", "not_dispatched");
+                throw new ComputerUseExecutorError("action_rejected", "not_dispatched");
             }
             const preparedGeneration = generation;
             let used = false;
             function check() {
                 if (signal.aborted)
-                    throw new CuaExecutorError("session_revoked", "not_dispatched");
+                    throw new ComputerUseExecutorError("session_revoked", "not_dispatched");
                 if (used || preparedGeneration !== generation)
-                    throw new CuaExecutorError("action_rejected", "not_dispatched");
+                    throw new ComputerUseExecutorError("action_rejected", "not_dispatched");
             }
             check();
             await content.assertReady(signal);
@@ -52,15 +52,15 @@ export function createGuestBrowserTools(native, content) {
                     try {
                         await native.input(["key", "--clearmodifiers", "ctrl+l"], signal);
                         if (signal.aborted)
-                            throw new CuaExecutorError("session_revoked", "outcome_uncertain");
+                            throw new ComputerUseExecutorError("session_revoked", "outcome_uncertain");
                         await content.assertReady(signal);
                         if (signal.aborted || preparedGeneration !== generation)
-                            throw new CuaExecutorError("session_revoked", "outcome_uncertain");
+                            throw new ComputerUseExecutorError("session_revoked", "outcome_uncertain");
                         await native.typeAscii(text, signal);
                     }
                     catch (error) {
                         // Ctrl+L itself is input. No content fallback after that dispatch.
-                        throw new CuaExecutorError(isCuaExecutorError(error) ? error.code : "execution_failed", "outcome_uncertain");
+                        throw new ComputerUseExecutorError(isComputerUseExecutorError(error) ? error.code : "execution_failed", "outcome_uncertain");
                     }
                 },
                 async close() {