@muggleai/works 5.27.0-staging.245 → 5.27.0-staging.247

package/.claude-plugin/marketplace.json

@@ -9,7 +9,7 @@
       "name": "muggleai",
       "source": "./plugin",
       "description": "Real-browser E2E acceptance for your coding agent: test the flows you changed and open the PR with screenshots. Built for flows other tools can't reach: managed test identities with live inboxes (magic links, email OTP and 2FA, verification mail, password resets), an encrypted secret store for login and test credentials, and a test graph that runs prerequisite flows first. Plain-English tests become replayable scripts on localhost or preview URLs.",
-      "version": "5.27.0-staging.245"
+      "version": "5.27.0-staging.247"
     }
   ]
 }

package/dist/{chunk-FI6VHHZN.js → chunk-6BAB7IVG.js}

@@ -5705,7 +5705,7 @@ async function checkServiceStack(params) {
   const snapshot = await readGenerationStore({ directory: ledgerDir(), emptyRecord: EMPTY_STACK_RECORD });
   const swept = sweepStackRecord({ record: snapshot.record, now: Date.now(), isAlive: isProcessAlive });
   const ownLease = swept.leases.find((lease) => lease.cwd === cwd) ?? null;
-  const base = { cwd, generation: snapshot.generation, mismatches: [], peerLease: null };
+  const base = { cwd, generation: snapshot.generation, mismatches: [], peerLease: null, servedNeeds: [] };
   const isClaimedByCaller = ownLease?.needs.some((need) => need.port === params.port) ?? false;
   const peerLease = params.port === void 0 || isClaimedByCaller ? void 0 : swept.leases.find((lease) => lease.cwd !== cwd && lease.needs.some((need) => need.port === params.port));
   if (peerLease !== void 0) {
@@ -5734,7 +5734,12 @@ async function checkServiceStack(params) {
       mismatches
     };
   }
-  return { ...base, verdict: "VALID" /* Valid */, reason: "every claimed port serves the claimed code" };
+  return {
+    ...base,
+    verdict: "VALID" /* Valid */,
+    reason: "every claimed port serves the claimed code",
+    servedNeeds: ownLease.needs
+  };
 }
 async function startStackRun(params) {
   const cwd = await resolveCheckoutRoot({ cwd: params.cwd, runGit: params.runGit ?? runGitCommand });
@@ -9596,6 +9601,20 @@ async function findRunStackRefusal(params) {
   });
   return REFUSING_VERDICTS.has(report.verdict) ? report : null;
 }
+async function readServedStack(params) {
+  if (!isLoopbackUrl(params.localUrl)) return [];
+  const report = await checkServiceStack({ cwd: params.cwd, runGit: params.runGit, readListener: params.readListener });
+  return report.servedNeeds;
+}
+function recordServedStack(params) {
+  if (params.servedStack.length === 0 || params.stackChange !== null) return;
+  const storage = getRunResultStorageService();
+  const runResult = storage.getRunResult(params.runId);
+  if (!runResult) return;
+  storage.updateRunResult(params.runId, {
+    localExecutionContext: { ...runResult.localExecutionContext, servedStack: params.servedStack }
+  });
+}
 async function runUnderStackLease(params) {
   if (!isLoopbackUrl(params.localUrl)) return await params.execute();
   const runId = randomUUID();
@@ -9985,6 +10004,7 @@ var executeTestGenerationTool = {
       logger7.warn("Test generation refused: service stack", { verdict: stackRefusal.verdict, cwd: input.cwd });
       return { content: formatRunStackRefusal(stackRefusal, input.cwd), isError: true, data: stackRefusal };
     }
+    const servedStack = await readServedStack({ cwd: input.cwd, localUrl: input.localUrl });
     try {
       const generationRun = await runUnderStackLease({
         cwd: input.cwd,
@@ -10012,6 +10032,7 @@ var executeTestGenerationTool = {
         generationRun.errorMessage ? `**Error:** ${generationRun.errorMessage}` : ""
       ].filter(Boolean).join("\n");
       const stackChange = await findStackChangedDuringRun({ cwd: input.cwd, localUrl: input.localUrl });
+      recordServedStack({ runId: generationRun.id, servedStack, stackChange });
       const reportedContent = stackChange ? content + formatStackChangedDuringRun(stackChange) : content;
       return {
         content: reportedContent,
@@ -10051,6 +10072,7 @@ var executeReplayTool = {
       logger7.warn("Test replay refused: service stack", { verdict: stackRefusal.verdict, cwd: input.cwd });
       return { content: formatRunStackRefusal(stackRefusal, input.cwd), isError: true, data: stackRefusal };
     }
+    const servedStack = await readServedStack({ cwd: input.cwd, localUrl: input.localUrl });
     try {
       const replayRun = await runUnderStackLease({
         cwd: input.cwd,
@@ -10079,6 +10101,7 @@ var executeReplayTool = {
         replayRun.errorMessage ? `**Error:** ${replayRun.errorMessage}` : ""
       ].filter(Boolean).join("\n");
       const stackChange = await findStackChangedDuringRun({ cwd: input.cwd, localUrl: input.localUrl });
+      recordServedStack({ runId: replayRun.id, servedStack, stackChange });
       const reportedContent = stackChange ? content + formatStackChangedDuringRun(stackChange) : content;
       return {
         content: reportedContent,

package/dist/{chunk-SJBYMBC2.js → chunk-B5BWBZZ5.js}

@@ -1,4 +1,4 @@
-import { __export, getLogger, getConfig, createChildLogger, buildElectronAppReleaseAssetUrl, getActiveRuntimeTarget, getAuthService, hasApiKey, getElectronAppVersion, getElectronAppDir, getPlatformKey, reconcileProjectPreferences, getDataDir2, PREFERENCES_FILE_NAME, isFirstRun, writePreferences, DEFAULT_PREFERENCES, isElectronAppInstalled, getElectronAppChecksums, getChecksumForPlatform, getElectronAppSignedFromVersion, getReleaseSignerIdentityUri, verifyFileChecksum, resolveElectronAppExecutablePath, calculateFileChecksum, initTelemetry, Surface, ServiceName, track, EventName, getQaTools, getLocalQaTools, performLogout, assertDeviceCodeClientProvisioned, performLogin, getActiveElectronAppReleaseStream, buildToolInputJsonSchema, toolRequiresAuth, getCallerCredentials, parseElectronAppDirName, getElectronAppReleaseTagPrefix, hasShownDisclosure, markDisclosureShown, getBundledElectronAppVersion, getElectronAppVersionSource, getCredentialsFilePath, ElectronAppReleaseStream, buildElectronAppChecksumsUrl, __require, getDisclosureCopy, buildOnboardingPlan, applyOnboardingAnswers, needsOnboarding, buildLlmSourcePlan, describeLlmSourceStatus, resolveLlmSourceStatus, readRunView, RunViewError, renderRunViewText, writeRunViewHtml, openBrowserUrl, evaluateCheckoutFreshness, runGitCommand, BLOCKING_FRESHNESS_VERDICTS, claimServiceStack, BUSY_EXIT_CODE, checkServiceStack, releaseServiceStack, readServiceStackStatus, resolvePreferences, OnboardingBlanketChoice, PreferenceKey, LlmSource, applyLlmSourceSelection, describeStackMismatch, getDataDir, DEFAULT_LOCAL_LLM_BASE_URL, isValidLocalBaseUrl } from './chunk-FI6VHHZN.js';
+import { __export, getLogger, getConfig, createChildLogger, buildElectronAppReleaseAssetUrl, getActiveRuntimeTarget, getAuthService, hasApiKey, getElectronAppVersion, getElectronAppDir, getPlatformKey, reconcileProjectPreferences, getDataDir2, PREFERENCES_FILE_NAME, isFirstRun, writePreferences, DEFAULT_PREFERENCES, isElectronAppInstalled, getElectronAppChecksums, getChecksumForPlatform, getElectronAppSignedFromVersion, getReleaseSignerIdentityUri, verifyFileChecksum, resolveElectronAppExecutablePath, calculateFileChecksum, initTelemetry, Surface, ServiceName, track, EventName, getQaTools, getLocalQaTools, performLogout, assertDeviceCodeClientProvisioned, performLogin, getActiveElectronAppReleaseStream, buildToolInputJsonSchema, toolRequiresAuth, getCallerCredentials, parseElectronAppDirName, getElectronAppReleaseTagPrefix, hasShownDisclosure, markDisclosureShown, getBundledElectronAppVersion, getElectronAppVersionSource, getCredentialsFilePath, ElectronAppReleaseStream, buildElectronAppChecksumsUrl, __require, getDisclosureCopy, buildOnboardingPlan, applyOnboardingAnswers, needsOnboarding, buildLlmSourcePlan, describeLlmSourceStatus, resolveLlmSourceStatus, readRunView, RunViewError, renderRunViewText, writeRunViewHtml, openBrowserUrl, evaluateCheckoutFreshness, runGitCommand, BLOCKING_FRESHNESS_VERDICTS, claimServiceStack, BUSY_EXIT_CODE, checkServiceStack, releaseServiceStack, readServiceStackStatus, resolvePreferences, OnboardingBlanketChoice, PreferenceKey, LlmSource, applyLlmSourceSelection, describeStackMismatch, getDataDir, DEFAULT_LOCAL_LLM_BASE_URL, isValidLocalBaseUrl, local_exports } from './chunk-6BAB7IVG.js';
 import { Server } from '@modelcontextprotocol/sdk/server/index.js';
 import { ListToolsRequestSchema, CallToolRequestSchema, ListResourcesRequestSchema, ReadResourceRequestSchema } from '@modelcontextprotocol/sdk/types.js';
 import { v4 } from 'uuid';
@@ -356,6 +356,16 @@ function parentIsAlive(pid) {
   }
 }
 
+// src/pr-walkthrough/constants.ts
+var REPORT_SENTINEL = "muggle-pr-section";
+var WALKTHROUGH_SLOT_MARKER = "<!-- muggle-pr-walkthrough:v1 -->";
+var WALKTHROUGH_SKIPPED_MARKER = "<!-- muggle-pr-walkthrough-status:skipped -->";
+var WALKTHROUGH_COMMENT_HEADING = "### Muggle AI \u2014 PR visual walkthrough";
+var RERUN_NOTICE_OPENING = "<!-- muggle-pr-walkthrough-rerun -->";
+var SHORT_SHA_LENGTH = 7;
+var GH_COMMENT_TIMEOUT_MS = 1e4;
+var DEPENDENCY_BOT_LOGINS = ["dependabot[bot]", "renovate[bot]"];
+
 // src/cli/pr-section/constants.ts
 var DETAIL_IMAGE_WIDTH = 720;
 var STEP_IMAGE_WIDTH = 640;
@@ -369,6 +379,84 @@ var SHORT_ID_LABEL_SEPARATOR = " \xB7 ";
 var TEST_DETAILS_ANCHOR_PREFIX = "muggle-test-";
 var STEP_PAIR_IMAGE_WIDTH = 420;
 var MAX_COMMENT_BYTES = 64e3;
+var FULL_COMMIT_SHA_PATTERN = /^[0-9a-f]{40}$/;
+var BASELINE_FAILED_STATUS = "failed";
+
+// src/cli/pr-section/baseline.ts
+var refused = (refusalReason) => ({ isCredited: false, refusalReason });
+async function resolveBaseRange(deps) {
+  const baseRef = deps.baseBranch ? `refs/remotes/origin/${deps.baseBranch}` : await deps.runGit(["symbolic-ref", "refs/remotes/origin/HEAD"], deps.cwd);
+  if (!baseRef) return null;
+  const baseTipSha = await deps.runGit(["rev-parse", "--verify", `${baseRef}^{commit}`], deps.cwd);
+  if (!baseTipSha) return null;
+  const mergeBaseSha = await deps.runGit(["merge-base", "HEAD", baseTipSha], deps.cwd);
+  return mergeBaseSha ? { mergeBaseSha, baseTipSha } : null;
+}
+async function isOnBaseSinceFork(sha, range, deps) {
+  const isAncestor = async (ancestor, descendant) => await deps.runGit(["merge-base", "--is-ancestor", ancestor, descendant], deps.cwd) !== null;
+  return await isAncestor(range.mergeBaseSha, sha) && await isAncestor(sha, range.baseTipSha);
+}
+function servedCommitsOf(baselineRun) {
+  const servedStack = baselineRun.localExecutionContext?.servedStack;
+  if (!Array.isArray(servedStack)) return [];
+  return servedStack.map((servedNeed) => servedNeed?.headSha).filter((sha) => typeof sha === "string" && FULL_COMMIT_SHA_PATTERN.test(sha));
+}
+async function judgeBaseline(test, baselineRunId, range, deps) {
+  if (range === null) return refused("the base branch could not be resolved in this checkout");
+  const baselineRun = deps.readRunResult(baselineRunId);
+  if (!baselineRun) return refused("no such run in the local run store");
+  if (baselineRun.id === test.runId) return refused("it is the failing run itself");
+  if (baselineRun.cloudTestCaseId !== test.testCaseId) return refused(`it ran test case ${baselineRun.cloudTestCaseId}`);
+  if (baselineRun.status !== BASELINE_FAILED_STATUS) return refused(`its status is ${baselineRun.status}`);
+  if (test.runSettingsId && baselineRun.cloudRunSettingsId && baselineRun.cloudRunSettingsId !== test.runSettingsId) {
+    return refused("it ran under a different runSettings record");
+  }
+  const servedCommits = servedCommitsOf(baselineRun);
+  if (servedCommits.length === 0) return refused("it recorded no stack it served");
+  for (const servedSha of servedCommits) {
+    if (!await isOnBaseSinceFork(servedSha, range, deps)) continue;
+    return {
+      isCredited: true,
+      failsOnBase: {
+        baselineRunId: baselineRun.id,
+        baselineSha: servedSha,
+        ...baselineRun.viewUrl && { viewUrl: baselineRun.viewUrl }
+      }
+    };
+  }
+  const span = `${range.mergeBaseSha.slice(0, SHORT_SHA_LENGTH)}..${range.baseTipSha.slice(0, SHORT_SHA_LENGTH)}`;
+  return refused(`none of the commits it served is on the base between ${span}`);
+}
+function withoutFailsOnBase(test) {
+  const uncredited = { ...test };
+  delete uncredited.failsOnBase;
+  return uncredited;
+}
+async function verifyBaselineFailures(report, deps) {
+  let range;
+  const tests = [];
+  for (const test of report.tests) {
+    if (test.status !== "failed") {
+      tests.push(test);
+      continue;
+    }
+    const uncredited = withoutFailsOnBase(test);
+    if (!test.baselineRunId) {
+      tests.push(uncredited);
+      continue;
+    }
+    if (range === void 0) range = await resolveBaseRange(deps);
+    const judgment = await judgeBaseline(uncredited, test.baselineRunId, range, deps);
+    if (judgment.isCredited) {
+      tests.push({ ...uncredited, failsOnBase: judgment.failsOnBase });
+      continue;
+    }
+    deps.stderrWrite(`build-pr-section: baseline ${test.baselineRunId} for "${test.name}" not credited: ${judgment.refusalReason}
+`);
+    tests.push(uncredited);
+  }
+  return { ...report, tests };
+}
 
 // src/e2e-result/types.ts
 var UnpassedCaseStatus = /* @__PURE__ */ ((UnpassedCaseStatus2) => {
@@ -400,8 +488,12 @@ function describeUnpassedCase(unpassed) {
   const outcome = unpassed.status === "failed" /* Failed */ && unpassed.failedStep !== void 0 ? `failed at step ${unpassed.failedStep}` : unpassed.status;
   return `${outcome}: ${escapeHtml(unpassed.detail)}`;
 }
+function describeFailsOnBase(failsOnBase) {
+  const baselineLink = failsOnBase.viewUrl ? ` ([baseline run](${failsOnBase.viewUrl}))` : "";
+  return ` \xB7 also fails on base \`${failsOnBase.baselineSha.slice(0, SHORT_SHA_LENGTH)}\`${baselineLink}`;
+}
 function renderUnpassedCaseBullet(unpassed, linkText, linkTarget) {
-  return `- ${UNPASSED_STATUS_EMOJI[unpassed.status]} **${escapeHtml(withResolutionLabel(unpassed.name, unpassed.resolutionLabel))}** \u2014 ${describeUnpassedCase(unpassed)} \xB7 [${linkText}](${linkTarget})`;
+  return `- ${UNPASSED_STATUS_EMOJI[unpassed.status]} **${escapeHtml(withResolutionLabel(unpassed.name, unpassed.resolutionLabel))}** \u2014 ${describeUnpassedCase(unpassed)} \xB7 [${linkText}](${linkTarget})` + (unpassed.failsOnBase ? describeFailsOnBase(unpassed.failsOnBase) : "");
 }
 var UnpassedCaseSchema = z.object({
   name: z.string(),
@@ -411,7 +503,8 @@ var UnpassedCaseSchema = z.object({
   status: z.enum(UnpassedCaseStatus),
   failedStep: z.number().int().positive().optional(),
   detail: z.string(),
-  viewUrl: z.string()
+  viewUrl: z.string(),
+  failsOnBase: z.object({ baselineRunId: z.string(), baselineSha: z.string(), viewUrl: z.string().optional() }).optional()
 });
 var E2eResultRecordSchema = z.object({
   total: z.number().int().nonnegative(),
@@ -467,7 +560,8 @@ function toUnpassedCase(test) {
       status: "failed" /* Failed */,
       failedStep: failureOrdinal(test, test.failureStepIndex),
       detail: oneLineDetail(test.error),
-      viewUrl: test.viewUrl
+      viewUrl: test.viewUrl,
+      ...test.failsOnBase && { failsOnBase: test.failsOnBase }
     };
   }
   if (test.status === "inconclusive") {
@@ -532,6 +626,9 @@ var FailedTestSchema = z.object({
   steps: z.array(StepSchema),
   failureStepIndex: z.number().int().nonnegative(),
   error: z.string().min(1),
+  // A run of the same test case on the PR's base the author says failed too; build-pr-section verifies it before
+  // the walkthrough credits it.
+  baselineRunId: z.string().min(1).optional(),
   artifactsDir: z.string().min(1).optional(),
   useCaseName: z.string().min(1).optional(),
   description: z.string().min(1).optional(),
@@ -1054,7 +1151,7 @@ async function resolveGsScreenshotUrls(report, opts) {
     if (gsUrls.length === 0) {
       return report;
     }
-    const mcps = await import('./src-33XQK7NM.js');
+    const mcps = await import('./src-MCJFH7AB.js');
     const credentials = await mcps.getCallerCredentialsAsync();
     if (!credentials.bearerToken && !credentials.apiKey) {
       stderrWrite(
@@ -1148,7 +1245,15 @@ ${err.issues.map((i) => `  - ${i.path.join(".")}: ${i.message}`).join("\n")}
     }
     return 1;
   }
-  const resolvedReport = await resolveGsScreenshotUrls(report, { stderrWrite: opts.stderrWrite });
+  const verifyBaselines = opts.verifyBaselines ?? ((parsedReport) => verifyBaselineFailures(parsedReport, {
+    cwd: process.cwd(),
+    baseBranch: opts.baseBranch,
+    runGit: runGitCommand,
+    readRunResult: (runId) => local_exports.getRunResultStorageService().getRunResult(runId),
+    stderrWrite: opts.stderrWrite
+  }));
+  const verifiedReport = await verifyBaselines(report);
+  const resolvedReport = await resolveGsScreenshotUrls(verifiedReport, { stderrWrite: opts.stderrWrite });
   const sentinelCost = Buffer.byteLength(`${REPORT_SECTION_SENTINEL}
 `, "utf-8");
   const renderedSection = buildPrSection(resolvedReport, {
@@ -1174,7 +1279,8 @@ async function buildPrSectionCommand(options) {
     stdin: process.stdin,
     stdoutWrite: (s) => process.stdout.write(s),
     stderrWrite: (s) => process.stderr.write(s),
-    maxBodyBytes
+    maxBodyBytes,
+    baseBranch: options.base
   });
   if (code !== 0) {
     process.exitCode = code;
@@ -2278,7 +2384,6 @@ var WATCH_STATUS_FILE = "status.json";
 var WATCH_STATE_FILE = "state.json";
 var WATCH_EVENTS_FILE = "events.jsonl";
 var WATCH_EVENTS_ROTATE_BYTES = 5 * 1024 * 1024;
-var MONITOR_STDOUT_CAP_MS = 30 * 6e4;
 var WATCHER_REARM_GRACE_MS = 5 * 6e4;
 var CODEX_ACTIVE_WINDOW_MS = 10 * 6e4;
 var CODEX_ROLLOUT_DAY_DIRS_SCANNED = 7;
@@ -3192,16 +3297,6 @@ var NON_APP_PATH_PATTERNS = [
   /^\.gitlab-ci\.yml$/
 ];
 
-// src/pr-walkthrough/constants.ts
-var REPORT_SENTINEL = "muggle-pr-section";
-var WALKTHROUGH_SLOT_MARKER = "<!-- muggle-pr-walkthrough:v1 -->";
-var WALKTHROUGH_SKIPPED_MARKER = "<!-- muggle-pr-walkthrough-status:skipped -->";
-var WALKTHROUGH_COMMENT_HEADING = "### Muggle AI \u2014 PR visual walkthrough";
-var RERUN_NOTICE_OPENING = "<!-- muggle-pr-walkthrough-rerun -->";
-var SHORT_SHA_LENGTH = 7;
-var GH_COMMENT_TIMEOUT_MS = 1e4;
-var DEPENDENCY_BOT_LOGINS = ["dependabot[bot]", "renovate[bot]"];
-
 // src/pr-walkthrough/comment.ts
 function renderReservedComment() {
   return `${WALKTHROUGH_SLOT_MARKER}
@@ -3411,6 +3506,7 @@ var CHECK_SUMMARY_MAX_CASES = 50;
 var OUTCOME_CONCLUSIONS = {
   ["settled" /* Settled */]: "success" /* Success */,
   ["e2e-not-passing" /* E2eNotPassing */]: "failure" /* Failure */,
+  ["e2e-fails-on-base" /* E2eFailsOnBase */]: "neutral" /* Neutral */,
   ["rerun-in-progress" /* RerunInProgress */]: null,
   ["unsettled" /* Unsettled */]: "failure" /* Failure */,
   ["unreadable" /* Unreadable */]: "neutral" /* Neutral */
@@ -3427,15 +3523,32 @@ function countLabel(count, status) {
   return count > 0 ? `${count} ${status}` : null;
 }
 function renderNotPassingSummary(record) {
+  return renderCaseSummary(
+    record,
+    `E2E acceptance did not pass \u2014 ${unpassedTally(record)} of ${testCaseCount(record)}. Fix them, re-run the whole suite, and update the walkthrough comment; editing it re-runs this check.`
+  );
+}
+function renderFailsOnBaseSummary(record) {
+  return renderCaseSummary(
+    record,
+    `E2E acceptance: ${unpassedTally(record)} of ${testCaseCount(record)}, each verified to fail on the base too. This PR did not cause them; fix them on the base branch.`
+  );
+}
+function testCaseCount(record) {
+  return `${record.total} test ${record.total === 1 ? "case" : "cases"}`;
+}
+function unpassedTally(record) {
   const failed2 = record.unpassed.filter((c) => c.status === "failed" /* Failed */).length;
-  const tally = [
+  return [
     countLabel(failed2, "failed" /* Failed */),
     countLabel(record.unpassed.length - failed2, "inconclusive" /* Inconclusive */)
   ].filter((part) => part !== null).join(", ");
+}
+function renderCaseSummary(record, headline) {
   const listed = record.unpassed.slice(0, CHECK_SUMMARY_MAX_CASES).map((unpassed) => renderUnpassedCaseBullet(unpassed, "open run", unpassed.viewUrl));
   const hidden = record.unpassed.length - listed.length;
   return [
-    `E2E acceptance did not pass \u2014 ${tally} of ${record.total} test ${record.total === 1 ? "case" : "cases"}. Fix them, re-run the whole suite, and update the walkthrough comment; editing it re-runs this check.`,
+    headline,
     "",
     ...listed,
     ...hidden > 0 ? [`- \u2026and ${hidden} more \u2014 see the walkthrough comment.`] : []
@@ -3469,7 +3582,10 @@ function judgeComments(pr, comments, run) {
   }
   const record = latestE2eResult(bodies);
   if (record && record.unpassed.length > 0) {
-    return { outcome: "e2e-not-passing" /* E2eNotPassing */, summary: renderNotPassingSummary(record) };
+    const isEveryUnpassedOnBase = record.unpassed.every(
+      (unpassed) => unpassed.status === "failed" /* Failed */ && unpassed.failsOnBase !== void 0
+    );
+    return isEveryUnpassedOnBase ? { outcome: "e2e-fails-on-base" /* E2eFailsOnBase */, summary: renderFailsOnBaseSummary(record) } : { outcome: "e2e-not-passing" /* E2eNotPassing */, summary: renderNotPassingSummary(record) };
   }
   const outcome = verdictNow === "satisfied" /* Satisfied */ ? "settled" /* Settled */ : "unsettled" /* Unsettled */;
   return { outcome, summary: FIXED_SUMMARIES[outcome] };
@@ -4873,8 +4989,7 @@ function resolveWatcherLiveness(loop, heartbeatAgeMs, isOwnerSessionActive, isPr
     return isHeartbeatWithin(WATCH_HEARTBEAT_STALE_MS) ? "polling" /* Polling */ : "dead" /* Dead */;
   }
   if (loop) {
-    const hasOutlivedMonitor = (loop.ageMs ?? 0) > MONITOR_STDOUT_CAP_MS;
-    return hasOutlivedMonitor || isOwnerSessionActive === false ? "deaf" /* Deaf */ : "polling" /* Polling */;
+    return isOwnerSessionActive === false ? "deaf" /* Deaf */ : "polling" /* Polling */;
   }
   return isHeartbeatWithin(WATCHER_REARM_GRACE_MS) ? "rearming" /* Rearming */ : "dead" /* Dead */;
 }
@@ -4941,12 +5056,11 @@ function watcherFinding(assessment, subject) {
     );
   }
   if (assessment.watcher === "deaf" /* Deaf */) {
-    const reason = assessment.isOwnerSessionActive === false ? `its owner session ${shortSessionId(assessment.ownerSessionId ?? "")} has ended` : `it has run ${formatDuration(assessment.loopAgeMs ?? 0)}, past the 30-minute Monitor cap`;
     return finding(
       "WATCHER_DEAF" /* WatcherDeaf */,
       [assessment.pr.url, assessment.loopPid],
       subject,
-      `watch loop pid ${assessment.loopPid} reports to no session: ${reason}`
+      `watch loop pid ${assessment.loopPid} reports to no session: its owner session ${shortSessionId(assessment.ownerSessionId ?? "")} has ended`
     );
   }
   return void 0;
@@ -6058,8 +6172,8 @@ function renderReloadReport(report, isDryRun) {
   const lines = report.decisions.map((decision) => {
     const label = decision.name ?? decision.shortId ?? decision.sessionId;
     const failed2 = report.failedSessionIds.includes(decision.sessionId) ? " (restart failed)" : "";
-    const refused = report.untrustedSessionIds.includes(decision.sessionId) ? ` (left running: ${decision.cwd ?? "its workspace"} refuses a background restart \u2014 run /reload-plugins in it)` : "";
-    return `  ${decision.action.padEnd(16)} ${label}  ${decision.reason}${failed2}${refused}`;
+    const refused2 = report.untrustedSessionIds.includes(decision.sessionId) ? ` (left running: ${decision.cwd ?? "its workspace"} refuses a background restart \u2014 run /reload-plugins in it)` : "";
+    return `  ${decision.action.padEnd(16)} ${label}  ${decision.reason}${failed2}${refused2}`;
   });
   const isRestart = (action) => action === "respawn" /* Respawn */ || action === "resume-reconcile" /* ResumeReconcile */;
   const respawnVerb = isDryRun ? "would restart" : "restarted";
@@ -6137,7 +6251,7 @@ function createProgram() {
   program.command("login").description("Authenticate with Muggle AI (uses device code flow)").option("--key-name <name>", "Name for the API key").option("--key-expiry <expiry>", "API key expiry: 30d, 90d, 1y, never", "90d").action(loginCommand);
   program.command("logout").description("Clear stored credentials").action(logoutCommand);
   program.command("status").description("Show authentication status").action(statusCommand);
-  program.command("build-pr-section").description("Render a muggle-do PR body evidence block from an e2e report on stdin").option("--max-body-bytes <n>", "Max UTF-8 byte budget for the PR body (default 60000)").action(buildPrSectionCommand);
+  program.command("build-pr-section").description("Render a muggle-do PR body evidence block from an e2e report on stdin").option("--max-body-bytes <n>", "Max UTF-8 byte budget for the PR body (default 60000)").option("--base <branch>", "Branch the PR merges into, for verifying cited baseline runs (default: the remote's default branch)").action(buildPrSectionCommand);
   program.command("run-view").argument("[runId]", "Run id, or a unique prefix of one (default: the newest run)").description("Show a local run's steps beside the frames they produced").option("--html", "Render a self-contained HTML page instead of terminal text").option("--open", "Open the rendered page in the default browser (implies --html)").option("--out-dir <dir>", "Directory to write the page into (default: current directory)").action(runViewCommand);
   program.command("pr-walkthrough-check").description("Check that a PR's Muggle AI walkthrough comment carries a result or a stated skip").option("--repo <owner/repo>", "Repository (default: $GITHUB_REPOSITORY)").option("--pr <number>", "Pull request number (default: the Actions event payload)").option("--head-sha <sha>", "Commit the check run attaches to (default: the PR head)").option("--check-run", "Publish the verdict as a check run instead of failing the process").action(prWalkthroughCheckCommand);
   program.command("checkout-freshness").description("Fetch the base branch and check that a checkout is built on its latest commit").option("--cwd <dir>", "Directory inside the checkout (default: the current directory)").option("--base <branch>", "Branch the change merges into (default: origin's default branch)").option("--no-fetch", "Check against the last fetched base instead of fetching").option("--json", "Print the full report as JSON").action(checkoutFreshnessCommand);

package/dist/cli.js

@@ -1,6 +1,6 @@
 #!/usr/bin/env node
-import { runCli } from './chunk-SJBYMBC2.js';
-import './chunk-FI6VHHZN.js';
+import { runCli } from './chunk-B5BWBZZ5.js';
+import './chunk-6BAB7IVG.js';
 
 // src/cli/main.ts
 runCli().catch((error) => {

package/dist/index.js

@@ -1,2 +1,2 @@
-export { src_exports as commands, createUnifiedMcpServer, server_exports as server } from './chunk-SJBYMBC2.js';
-export { createChildLogger, e2e_exports as e2e, getConfig, getLocalQaTools, getLogger, getQaTools, local_exports as localQa, mcp_exports as mcp, e2e_exports as qa, src_exports as shared } from './chunk-FI6VHHZN.js';
+export { src_exports as commands, createUnifiedMcpServer, server_exports as server } from './chunk-B5BWBZZ5.js';
+export { createChildLogger, e2e_exports as e2e, getConfig, getLocalQaTools, getLogger, getQaTools, local_exports as localQa, mcp_exports as mcp, e2e_exports as qa, src_exports as shared } from './chunk-6BAB7IVG.js';

package/dist/plugin/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "muggle",
   "description": "Real-browser E2E acceptance for your coding agent: test the flows you changed and open the PR with screenshots. Built for flows other tools can't reach: managed test identities with live inboxes (magic links, email OTP and 2FA, verification mail, password resets), an encrypted secret store for login and test credentials, and a test graph that runs prerequisite flows first. Plain-English tests become replayable scripts on localhost or preview URLs.",
-  "version": "5.27.0-staging.245",
+  "version": "5.27.0-staging.247",
   "author": {
     "name": "Muggle AI",
     "email": "support@muggle-ai.com"

package/dist/plugin/.codex-mcp.json

@@ -4,7 +4,7 @@
       "command": "npx",
       "args": [
         "-y",
-        "@muggleai/works@5.27.0-staging.245",
+        "@muggleai/works@5.27.0-staging.247",
         "serve"
       ],
       "env": {

package/dist/plugin/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "muggle",
-  "version": "5.27.0-staging.245",
+  "version": "5.27.0-staging.247",
   "description": "Real-browser E2E acceptance for your coding agent: test the flows you changed and open the PR with screenshots. Built for flows other tools can't reach: managed test identities with live inboxes (magic links, email OTP and 2FA, verification mail, password resets), an encrypted secret store for login and test credentials, and a test graph that runs prerequisite flows first. Plain-English tests become replayable scripts on localhost or preview URLs.",
   "author": {
     "name": "Muggle AI",

package/dist/plugin/.cursor-plugin/plugin.json

@@ -2,7 +2,7 @@
   "name": "muggle",
   "displayName": "Muggle AI",
   "description": "Real-browser E2E acceptance for your coding agent: test the flows you changed and open the PR with screenshots. Built for flows other tools can't reach: managed test identities with live inboxes (magic links, email OTP and 2FA, verification mail, password resets), an encrypted secret store for login and test credentials, and a test graph that runs prerequisite flows first. Plain-English tests become replayable scripts on localhost or preview URLs.",
-  "version": "5.27.0-staging.245",
+  "version": "5.27.0-staging.247",
   "author": {
     "name": "Muggle AI",
     "email": "support@muggle-ai.com"

package/dist/plugin/.mcp.json

@@ -4,7 +4,7 @@
       "command": "npx",
       "args": [
         "-y",
-        "@muggleai/works@5.27.0-staging.245",
+        "@muggleai/works@5.27.0-staging.247",
         "serve"
       ]
     }

package/dist/plugin/hooks/README.md

@@ -77,9 +77,9 @@ Codex runs a plugin's hooks only once the user trusts them: it asks at startup (
 
 `SessionStart` (`scripts/reconcile-stale-watchers.sh`) — a standalone advisory, not part of the `guardrails.mjs` decision tree above.
 
-`muggle-pr-followup` watchers are session-only (a monitor or `/loop` cron); they die with their session, leaving open PRs with no live poller. This is by design — a review is addressed only inside a session that carries the context to address it. The skill's [`reconcile`](../skills/muggle-pr-followup/reconcile.md) procedure recovers them at the next session start — finalizes slots whose PR went terminal, sweeps orphan crons, re-arms silently-stopped open watchers — but re-arming needs Claude tools a shell hook can't call. So this hook nudges rather than acts.
+`muggle-pr-followup` watchers are session-only (a background watch loop or `/loop` cron); they die with their session, leaving open PRs with no live poller. This is by design — a review is addressed only inside a session that carries the context to address it. The skill's [`reconcile`](../skills/muggle-pr-followup/reconcile.md) procedure recovers them at the next session start — finalizes slots whose PR went terminal, sweeps orphan crons, re-arms silently-stopped open watchers — but re-arming needs Claude tools a shell hook can't call. So this hook nudges rather than acts.
 
-The nudge counts only slots **this session owns**, and reports the rest as orphans it will not act on — ownership and the recovery rules it gates are defined in [`reconcile`](../skills/muggle-pr-followup/reconcile.md). Nothing owned and nothing orphaned → it emits nothing. A pure directory scan (no `gh`, no writes), so it's cheap enough for every session start.
+The nudge counts only slots **this session owns** — ownership and the recovery rules it gates are defined in [`reconcile`](../skills/muggle-pr-followup/reconcile.md). A session that owns none hears nothing: the hook speaks into every session start, and slots owned elsewhere are no work of this one's — `muggle watch status` reports them. A pure directory scan (no `gh`, no writes), so it's cheap enough for every session start.
 
 ## Session-start upkeep
 

package/dist/plugin/scripts/pr-watch-arm.sh

@@ -8,9 +8,10 @@
 # and never reaching its terminal check. Every health signal said fine. Doing
 # the three steps here, in one command, is what makes that unskippable.
 #
-# The floors are taken from the drain's own read rather than from a fresh fetch
-# afterwards: anything that lands between the two would otherwise be marked seen
-# without ever being reported.
+# A first arm takes its floors from the drain's own read rather than from a fresh
+# fetch afterwards: anything that lands between the two would otherwise be marked
+# seen without ever being reported. A re-arm keeps the review, comment and thread
+# floors the session already holds, for the same reason.
 #
 # A stale branch or a red head is floored only when the slot proves a cycle ran
 # for it. Flooring whatever was outstanding at arm time let a session that never
@@ -38,7 +39,7 @@ if [ -z "$slot" ] || [ -z "$repo" ] || [ -z "$pr_number" ] || [ -z "$base_branch
     exit 2
 fi
 
-for lib in pr-watch-guards.sh pr-watch-events.sh pr-watch-fetch.sh pr-watch-due.sh; do
+for lib in pr-watch-guards.sh pr-watch-events.sh pr-watch-fetch.sh pr-watch-due.sh pr-watch-floors.sh; do
     if [ ! -f "${script_dir}/${lib}" ]; then
         echo "pr-watch-arm: missing ${lib} beside this script" >&2
         exit 2
@@ -135,9 +136,10 @@ recorded_rebase_due_key="$rebase_due_key"
 thread_count=0
 [ -n "$unresolved_threads" ] && thread_count=$(printf '%s' "$unresolved_threads" | tr ';' '\n' | grep -c .)
 
-# Printed, not swallowed: these are the things the arming session owes action on
-# before the monitor takes over. Unattempted work is left unfloored, so the
-# loop's first pass reports it again as an event.
+# A record of the PR as armed, read from the watch's output when its loop ends.
+# Printing wakes nobody: the session hears the watch only when the loop exits.
+# Unattempted work is left unfloored, so the loop's first poll reports it and
+# exits — that report is the wake.
 echo "DRAIN pr=$pr_number state=$pr_state head=${head_sha:0:8}"
 echo "DRAIN unresolved-threads=$thread_count latest-review=${latest_review:-0} latest-comment=${latest_comment:-0}"
 echo "DRAIN checks pending=${pending_checks:-0} failed=${failed_checks:-0} behind-base=${behind_count:-unknown} mergeable=${mergeable:-UNKNOWN}"
@@ -158,10 +160,25 @@ if [ -n "$rebase_due_key" ]; then
     fi
 fi
 
+# A re-arm carries the review, comment and thread floors forward. The session
+# advanced them past what it handled, so anything newer — a review landing
+# between its last tick's read and this one — is above them, and the loop's
+# first poll reports it. Seeding them from this read would floor that review
+# unseen. Only a first arm, with nothing handled yet, takes them from its read.
+review_floor="${latest_review:-0}"
+comment_floor="${latest_comment:-0}"
+threads_floor="$unresolved_threads"
+if [ -f "${slot}/watch-watermark.env" ]; then
+    review_floor="$(watch_read_slot_value "$slot" watch-watermark.env REV)"
+    comment_floor="$(watch_read_slot_value "$slot" watch-watermark.env COM)"
+    threads_floor="$(watch_threads_still_unresolved \
+        "$(watch_read_slot_value "$slot" watch-watermark.env THREADS)" "$unresolved_threads")"
+fi
+
 {
-    printf 'REV=%s\n' "${latest_review:-0}"
-    printf 'COM=%s\n' "${latest_comment:-0}"
-    printf 'THREADS="%s"\n' "$unresolved_threads"
+    printf 'REV=%s\n' "${review_floor:-0}"
+    printf 'COM=%s\n' "${comment_floor:-0}"
+    printf 'THREADS="%s"\n' "$threads_floor"
     printf 'CIRED="%s"\n' "$ci_red_floor"
     printf 'REBASED="%s"\n' "$rebase_floor"
     printf 'BLOCKED_CIDIGEST="%s"\n' "$blocked_digest_floor"

package/dist/plugin/scripts/pr-watch-floors.sh

@@ -0,0 +1,62 @@
+#!/usr/bin/env bash
+
+# The floors a watch compares each poll against, as they persist between loops.
+# Sourced by pr-watch-arm.sh and pr-watch-loop.sh.
+#
+# A loop exits on every report, because the session that owns it hears it only
+# when it exits. So no floor can live in a loop's memory: whatever stops one
+# occurrence firing twice has to be on disk for the next loop to read. Two files
+# hold them — watch-watermark.env, which the session advances past what it
+# handled, and watch-reported.env, which the loop advances past what it told
+# the session — and a poll fires only on what is above both.
+
+MUGGLE_PR_WATCH_REPORTED_FILE="watch-reported.env"
+
+# Prints the value of `key` in a slot file of `KEY="value"` lines, quotes
+# stripped, the last match winning; nothing when the file or key is absent.
+# Read with builtins, never sourced: THREADS holds bare semicolons, which a
+# sourced unquoted line would split, and a CI digest carries check names an
+# author controls.
+watch_read_slot_value() {
+    local slot="$1" file_name="$2" key="$3" line value=""
+    [ -f "${slot}/${file_name}" ] || return 0
+    while IFS= read -r line || [ -n "$line" ]; do
+        line="${line%$'\r'}"
+        case "$line" in
+            "${key}="*)
+                value="${line#"${key}"=}"
+                value="${value%\"}"
+                value="${value#\"}"
+                ;;
+        esac
+    done < "${slot}/${file_name}"
+    printf '%s' "$value"
+}
+
+# Writes <slot>/watch-reported.env: the newest review and comment ids and the
+# unresolved threads a loop's last poll saw — every one of them either reported
+# or already floored — and the red head, rebase key and blocked digest it
+# reported, with the block (`blocked_for`, that block's watermark digest) the
+# digest was reported under.
+watch_record_reported() {
+    local slot="$1" review="$2" comment="$3" threads="$4" ci_red="$5" rebase="$6" blocked_digest="$7" blocked_for="$8"
+    printf 'REV=%s\nCOM=%s\nTHREADS="%s"\nCIRED="%s"\nREBASED="%s"\nBLOCKED_CIDIGEST="%s"\nBLOCKED_FOR="%s"\n' \
+        "$review" "$comment" "$threads" "$ci_red" "$rebase" "$blocked_digest" "$blocked_for" \
+        > "${slot}/${MUGGLE_PR_WATCH_REPORTED_FILE}"
+}
+
+# Prints the threads of `known` (semicolon-joined) that are still in
+# `unresolved`, semicolon-joined. A thread resolved since it was floored drops
+# out, so its unresolving again is news.
+watch_threads_still_unresolved() {
+    local known="$1" unresolved="$2" thread_id kept=""
+    local -a thread_ids
+    IFS=';' read -ra thread_ids <<< "$known"
+    for thread_id in "${thread_ids[@]}"; do
+        [ -n "$thread_id" ] || continue
+        case ";${unresolved};" in
+            *";${thread_id};"*) kept="${kept:+${kept};}${thread_id}" ;;
+        esac
+    done
+    printf '%s' "$kept"
+}