@muggleai/works 5.31.0-staging.276 → 5.31.0-staging.278

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": "5.31.0-staging.276"
+      "version": "5.31.0-staging.278"
     }
   ]
 }

package/README.md

@@ -40,12 +40,13 @@ Works in Claude Code, Cursor, Codex, Windsurf, and any MCP client. Powered by [M
 | 3 | Build | Delegates real design surface to a design → plan → subagent build, then commits (signed) | Conventional commits |
 | 4 | Impact analysis | Maps the diff to the user flows it can break | Affected-flow list |
 | 5 | Unit tests | Runs the suite authored in stage 3 | Exit code |
-| 6 | Acceptance | Drives a real browser through the affected flows | Verdict + `runId` + per-step screenshots |
-| 7 | Pull request | Opens the PR/MR with the evidence block in the body | PR URL |
+| 5.5 | Pull request | Opens the PR/MR before the browser run, its walkthrough check red until stage 7 | PR URL |
+| 6 | Acceptance | Drives a real browser through the affected flows, its progress showing on the PR | Verdict + `runId` + per-step screenshots |
+| 7 | Walkthrough | Settles the PR's walkthrough comment with the run's evidence | Walkthrough comment |
 | 7.5 | Repair | Investigates and fixes acceptance failures, up to 3 iterations | Repair log |
 | 8 | Watcher | Polls reviews, CI, and base-branch drift; re-enters the cycle on each | `followup.log` |
 
-Stage 7 refuses to run until the Definition of Done holds. Stage 8 is dispatched only once stage 7.5 clears — a cycle with unrepaired failures never reaches a watcher.
+Stage 5.5 refuses to open the PR until build and unit tests meet the Definition of Done, and stage 7 refuses to settle its walkthrough without an E2E verdict. Stage 8 is dispatched only once stage 7.5 clears — a cycle with unrepaired failures never reaches a watcher.
 
 ---
 

package/dist/{chunk-HK7XCCNT.js → chunk-CGKARU5D.js}

@@ -1,4 +1,4 @@
-import { __export, getLogger, TelemetryEnvironment, 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-7J7DS4SA.js';
+import { __export, getLogger, TelemetryEnvironment, 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-NTEBYUCA.js';
 import { Server } from '@modelcontextprotocol/sdk/server/index.js';
 import { ListToolsRequestSchema, CallToolRequestSchema, ListResourcesRequestSchema, ReadResourceRequestSchema } from '@modelcontextprotocol/sdk/types.js';
 import { v4 } from 'uuid';
@@ -1165,7 +1165,7 @@ async function resolveGsScreenshotUrls(report, opts) {
     if (gsUrls.length === 0) {
       return report;
     }
-    const mcps = await import('./src-QA7TWQ2M.js');
+    const mcps = await import('./src-LYOBYPFZ.js');
     const credentials = await mcps.getCallerCredentialsAsync();
     if (!credentials.bearerToken && !credentials.apiKey) {
       stderrWrite(
@@ -2937,6 +2937,11 @@ var MUGGLE_MCP_SERVER_PATTERN = /@muggleai[\\/]works\b.*\bserve\b/;
 var PR_WATCH_LOOP_PATTERN = /pr-watch-loop\.sh\b.*?--slot\s+(?:"([^"]+)"|(\S+))/;
 var CLAUDE_SESSION_REGISTRY_DIR_NAME = "sessions";
 var CLAUDE_PROJECTS_DIR_NAME = "projects";
+var CLAUDE_JOBS_DIR_NAME = "jobs";
+var CLAUDE_JOB_STATE_FILE_NAME = "state.json";
+var SESSION_NAME_FLAG = "--name";
+var CLAUDE_PROMPT_NAME_MAX_LENGTH = 200;
+var CUSTOM_TITLE_ENTRY_TYPE = "custom-title";
 var CLAUDE_INSTALLED_PLUGINS_RELATIVE_PATH = ["plugins", "installed_plugins.json"];
 var CLAUDE_GLOBAL_CONFIG_FILE_NAME = ".claude.json";
 var CLAUDE_SETTINGS_FILE_NAME = "settings.json";
@@ -3004,6 +3009,7 @@ function resolveWatchPaths(env = process.env) {
     slotsDir: join(muggleDataDir, ...PR_WATCH_SLOTS_RELATIVE_PATH),
     claudeSessionRegistryDir: join(claudeConfigDir, CLAUDE_SESSION_REGISTRY_DIR_NAME),
     claudeProjectsDir: join(claudeConfigDir, CLAUDE_PROJECTS_DIR_NAME),
+    claudeJobsDir: join(claudeConfigDir, CLAUDE_JOBS_DIR_NAME),
     claudeInstalledPluginsFile: join(claudeConfigDir, ...CLAUDE_INSTALLED_PLUGINS_RELATIVE_PATH),
     claudeGlobalConfigFile: join(env.CLAUDE_CONFIG_DIR || homedir(), CLAUDE_GLOBAL_CONFIG_FILE_NAME),
     claudeSettingsFile: join(claudeConfigDir, CLAUDE_SETTINGS_FILE_NAME),
@@ -5167,6 +5173,44 @@ function readCodexSessions(sessionsDir, nowMs) {
   }
   return sessions;
 }
+function jobStatePath(jobsDir, shortId) {
+  return join(jobsDir, shortId, CLAUDE_JOB_STATE_FILE_NAME);
+}
+function readJobState(statePath) {
+  try {
+    const parsed = JSON.parse(readFileSync(statePath, "utf-8"));
+    return typeof parsed === "object" && parsed !== null && !Array.isArray(parsed) ? parsed : void 0;
+  } catch {
+    return void 0;
+  }
+}
+function respawnFlagsOf(state) {
+  return Array.isArray(state.respawnFlags) ? state.respawnFlags.filter((flag) => typeof flag === "string") : [];
+}
+function nameFlagOf(respawnFlags) {
+  const flagIndex = respawnFlags.lastIndexOf(SESSION_NAME_FLAG);
+  return flagIndex === -1 ? void 0 : optionalString(respawnFlags[flagIndex + 1]);
+}
+function readClaudeJob(jobsDir, shortId) {
+  const state = readJobState(jobStatePath(jobsDir, shortId));
+  if (state === void 0) return void 0;
+  const intent = optionalString(state.intent);
+  return { respawnFlags: respawnFlagsOf(state), ...intent === void 0 ? {} : { intent } };
+}
+function pinClaudeJobName(jobsDir, shortId, name) {
+  const statePath = jobStatePath(jobsDir, shortId);
+  const state = readJobState(statePath);
+  if (state === void 0) return false;
+  const respawnFlags = respawnFlagsOf(state);
+  if (nameFlagOf(respawnFlags) === name) return true;
+  const otherFlags = respawnFlags.filter((flag, index) => flag !== SESSION_NAME_FLAG && respawnFlags[index - 1] !== SESSION_NAME_FLAG);
+  try {
+    writeFileSync(statePath, JSON.stringify({ ...state, respawnFlags: [...otherFlags, SESSION_NAME_FLAG, name] }, null, 2));
+    return true;
+  } catch {
+    return false;
+  }
+}
 function findTranscriptPath(projectsDir, sessionId) {
   let projectDirNames;
   try {
@@ -5211,6 +5255,23 @@ function readLimitStop(projectsDir, sessionId) {
   }
   return void 0;
 }
+function readCustomTitle(projectsDir, sessionId) {
+  const transcriptPath = findTranscriptPath(projectsDir, sessionId);
+  if (transcriptPath === void 0) return void 0;
+  let transcript;
+  try {
+    transcript = readFileSync(transcriptPath, "utf-8");
+  } catch {
+    return void 0;
+  }
+  let customTitle;
+  for (const line of transcript.split("\n")) {
+    if (!line.includes(CUSTOM_TITLE_ENTRY_TYPE)) continue;
+    const entry = parseTranscriptLine(line);
+    if (entry?.type === CUSTOM_TITLE_ENTRY_TYPE) customTitle = optionalString(entry.customTitle) ?? customTitle;
+  }
+  return customTitle;
+}
 function findClaudeSessionCwd(projectsDir, sessionId) {
   const transcriptPath = findTranscriptPath(projectsDir, sessionId);
   if (transcriptPath === void 0) return void 0;
@@ -5430,7 +5491,7 @@ function startServiceLoops(assessments, nowMs, isPollingDisabled, ownerSessionOf
 }
 
 // src/watch/session-restart.ts
-function resumeSession(target, runCommand, waitForExit, isSessionRunning2) {
+function resumeSession(target, runCommand, waitForExit, isSessionRunning2, pinSessionName) {
   const runningSessionId = target.runningSessionId ?? target.sessionId;
   if (target.runningShortId !== void 0) {
     runCommand("claude", ["stop", target.runningShortId]);
@@ -5438,6 +5499,7 @@ function resumeSession(target, runCommand, waitForExit, isSessionRunning2) {
   } else if (isSessionRunning2(runningSessionId)) {
     return { outcome: "still-running" /* StillRunning */ };
   }
+  pinSessionName(target.sessionId, target.runningShortId ?? shortSessionId(target.sessionId));
   const resumed = runCommand("claude", ["--resume", target.sessionId, "--bg", target.prompt], { cwd: target.cwd, isStderrCaptured: true });
   if (resumed?.exitCode !== 0) return { outcome: "failed" /* Failed */ };
   const backgroundedShortId = resumed.stdout.match(BACKGROUNDED_SESSION_ID_PATTERN)?.[1];
@@ -5530,7 +5592,8 @@ function executeWatcherRecoveries(planned, attemptsByOwner, deps) {
       { sessionId: recovery.ownerSessionId, cwd, prompt, runningShortId: recovery.ownerShortId, runningSessionId: recovery.ownerRunningSessionId },
       deps.runCommand,
       deps.waitForExit,
-      deps.isSessionRunning
+      deps.isSessionRunning,
+      deps.pinSessionName
     );
     if (resumed.outcome === "continued" /* Continued */) {
       markServiceLoopReportDelivered(recovery.slotDir);
@@ -5889,7 +5952,8 @@ function executeLimitResumes(planned, attemptsBySession, deps) {
         { sessionId: decision.sessionId, cwd, prompt: LIMIT_RESUME_PROMPT, runningShortId: decision.runningShortId },
         deps.runCommand,
         deps.waitForExit,
-        deps.isSessionRunning
+        deps.isSessionRunning,
+        deps.pinSessionName
       );
       outcome = continuedOutcomeOf(resumed.outcome);
       message = continueMessageOf(resumed, decision.reason);
@@ -6159,7 +6223,7 @@ function planReloadFanOut(sessions, pending, watchOwnerSessionIds, maxRestarts =
   }
   return decisions;
 }
-function executeReloadFanOut(decisions, runCommand, canRestartIn) {
+function executeReloadFanOut(decisions, runCommand, canRestartIn, pinSessionName) {
   const restartableByCwd = /* @__PURE__ */ new Map();
   const isRestartable = (cwd) => {
     if (!restartableByCwd.has(cwd)) restartableByCwd.set(cwd, canRestartIn(cwd));
@@ -6172,6 +6236,7 @@ function executeReloadFanOut(decisions, runCommand, canRestartIn) {
       report.untrustedSessionIds.push(decision.sessionId);
       continue;
     }
+    pinSessionName(decision.sessionId, decision.shortId);
     const isRestarted = runCommand("claude", ["respawn", decision.shortId])?.exitCode === 0;
     (isRestarted ? report.respawnedSessionIds : report.failedSessionIds).push(decision.sessionId);
   }
@@ -6372,6 +6437,16 @@ function stampSessionLoads(sessions, table, tracked, installs, nowMs) {
   });
 }
 
+// src/watch/session-names.ts
+function isPlaceholderSessionName(name, shortId) {
+  return name === void 0 || name === shortId || name === LIMIT_RESUME_PROMPT || name.startsWith("/");
+}
+function resolveSessionName(sources) {
+  const promptName = sources.job?.intent?.replace(/\s+/g, " ").trim().slice(0, CLAUDE_PROMPT_NAME_MAX_LENGTH);
+  const candidates = [sources.customTitle, sources.job && nameFlagOf(sources.job.respawnFlags), sources.registeredName, promptName];
+  return candidates.find((candidate) => !isPlaceholderSessionName(candidate, sources.shortId));
+}
+
 // src/watch/terminal-pr-cache.ts
 function isRecheckDue(record, nowMs) {
   if (!record) return true;
@@ -6677,12 +6752,12 @@ function reloadRecordFor(state, targetAtMs) {
 }
 function runReloadPass(record, claudeSessions, watchOwnerSessionIds, deps, at) {
   const decisions = planReloadFanOut(claudeSessions, record, watchOwnerSessionIds);
-  const report = executeReloadFanOut(decisions, deps.runCommand, canRestartVia(deps));
+  const report = executeReloadFanOut(decisions, deps.runCommand, canRestartVia(deps), pinSessionNameVia(deps, claudeSessions));
   return { report, pendingReload: advancePendingReload(record, report), events: reloadEvents(report, at) };
 }
 function runRemoteControlPass(login, claudeSessions, restartsBySession, maxRestarts, deps, at) {
   const plan = planRemoteControlRestarts(claudeSessions, login, restartsBySession, deps.nowMs, maxRestarts);
-  const executed = executeReloadFanOut(plan.decisions, deps.runCommand, canRestartVia(deps));
+  const executed = executeReloadFanOut(plan.decisions, deps.runCommand, canRestartVia(deps), pinSessionNameVia(deps, claudeSessions));
   return {
     report: { ...executed, connectedSessionIds: plan.connectedSessionIds },
     restartsBySession: recordRemoteControlRestarts(restartsBySession, executed, deps.nowMs),
@@ -6698,14 +6773,26 @@ function waitForExitVia(deps) {
 function isSessionRunningVia(deps) {
   return (sessionId) => isSessionRunning(deps.runCommand, deps.paths.claudeSessionRegistryDir, sessionId) !== false;
 }
-function resumeDepsOf(deps) {
+function pinSessionNameVia(deps, claudeSessions) {
+  return (sessionId, shortId) => {
+    const name = resolveSessionName({
+      shortId,
+      customTitle: readCustomTitle(deps.paths.claudeProjectsDir, sessionId),
+      job: readClaudeJob(deps.paths.claudeJobsDir, shortId),
+      registeredName: claudeSessions?.find((session) => session.sessionId === sessionId)?.name
+    });
+    if (name !== void 0) pinClaudeJobName(deps.paths.claudeJobsDir, shortId, name);
+  };
+}
+function resumeDepsOf(deps, claudeSessions) {
   return {
     runCommand: deps.runCommand,
     claudeProjectsDir: deps.paths.claudeProjectsDir,
     nowMs: deps.nowMs,
     canRestartIn: canRestartVia(deps),
     waitForExit: waitForExitVia(deps),
-    isSessionRunning: isSessionRunningVia(deps)
+    isSessionRunning: isSessionRunningVia(deps),
+    pinSessionName: pinSessionNameVia(deps, claudeSessions)
   };
 }
 function runLimitResumePass(claudeSessions, attemptsBySession, maxContinues, deps) {
@@ -6722,7 +6809,7 @@ function runLimitResumePass(claudeSessions, attemptsBySession, maxContinues, dep
     isClaudeUsageLimitInForce(deps.paths.claudeGlobalConfigFile, deps.nowMs),
     maxContinues
   );
-  const executed = executeLimitResumes(planned, attemptsBySession, resumeDepsOf(deps));
+  const executed = executeLimitResumes(planned, attemptsBySession, resumeDepsOf(deps, claudeSessions));
   return { report: { decisions: executed.decisions }, attemptsBySession: executed.attemptsBySession, events: executed.events };
 }
 function reloadTargetAtMs(deps, installs) {
@@ -6787,7 +6874,7 @@ function runTickUnlocked(deps) {
       sessionsById,
       isClaudeUsageLimitInForce(deps.paths.claudeGlobalConfigFile, deps.nowMs)
     );
-    const executed = executeWatcherRecoveries(planned, recoveryAttemptsByOwner, resumeDepsOf(deps));
+    const executed = executeWatcherRecoveries(planned, recoveryAttemptsByOwner, resumeDepsOf(deps, claudeSessions));
     recoveries = executed.recoveries;
     recoveryAttemptsByOwner = executed.attemptsByOwner;
     events.push(...executed.events);

package/dist/{chunk-7J7DS4SA.js → chunk-NTEBYUCA.js}

@@ -2858,7 +2858,7 @@ var PREFERENCES_SCHEMA = {
     description: "When a prior session left an E2E validation context for this working tree, reuse it instead of re-asking the validation questions"
   },
   ["autoRouteBuildToMuggleDo" /* AutoRouteBuildToMuggleDo */]: {
-    description: "When a prompt looks like a build/implement/fix request, route it through the /muggle-do pipeline (requirements \u2192 build \u2192 impact \u2192 tests \u2192 E2E \u2192 PR \u2192 watcher)"
+    description: "When a prompt looks like a build/implement/fix request, route it through the /muggle-do pipeline (requirements \u2192 build \u2192 impact \u2192 tests \u2192 PR \u2192 E2E \u2192 watcher)"
   },
   ["autoWatchPR" /* AutoWatchPR */]: {
     description: "After a test run opens a PR, start a muggle-pr-followup watcher that polls for new reviews and hands them to /muggle-do"

package/dist/cli.js

@@ -1,6 +1,6 @@
 #!/usr/bin/env node
-import { runCli } from './chunk-HK7XCCNT.js';
-import './chunk-7J7DS4SA.js';
+import { runCli } from './chunk-CGKARU5D.js';
+import './chunk-NTEBYUCA.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-HK7XCCNT.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-7J7DS4SA.js';
+export { src_exports as commands, createUnifiedMcpServer, server_exports as server } from './chunk-CGKARU5D.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-NTEBYUCA.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": "5.31.0-staging.276",
+  "version": "5.31.0-staging.278",
   "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@5.31.0-staging.276",
+        "@muggleai/works@5.31.0-staging.278",
         "serve"
       ],
       "env": {

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

@@ -1,6 +1,6 @@
 {
   "name": "muggle",
-  "version": "5.31.0-staging.276",
+  "version": "5.31.0-staging.278",
   "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": "5.31.0-staging.276",
+  "version": "5.31.0-staging.278",
   "author": {
     "name": "Muggle AI",
     "email": "support@muggle-ai.com"

package/dist/plugin/.mcp.json

@@ -4,7 +4,7 @@
       "command": "npx",
       "args": [
         "-y",
-        "@muggleai/works@5.31.0-staging.276",
+        "@muggleai/works@5.31.0-staging.278",
         "serve"
       ]
     }

package/dist/plugin/scripts/guardrails.mjs

@@ -137,6 +137,7 @@ var RECONCILE_REPLAY_TIMEOUT_MS = 15e3;
 var GUARDRAIL_SUBCOMMAND_ASSIGNMENT = /^GUARDRAIL_SUBCOMMAND="([\w-]+)"/m;
 var HOOK_COMMAND_SCRIPTS = /^set -- ([^;]+);/;
 var GUARDRAIL_DISPATCH_SCRIPT = "guardrail-dispatch.sh";
+var MUGGLE_DO_STAGES_AFTER_BUILD = "impact \u2192 unit tests \u2192 PR \u2192 E2E \u2192 watcher";
 
 // src/guardrails/store/fileLock.ts
 function isProcessAlive(pid) {
@@ -3305,7 +3306,8 @@ function watchGate() {
   if (state.watchSkipped !== true && prsAwaitingStateScan(unwatchedPrUrls, state, prStateScan).length > 0) {
     startBackgroundJob("scan-watch-pr-states" /* ScanWatchPrStates */, [sessionId]);
   }
-  const decision = watchGateDecision(state, resolveOwedWatchPrs(unwatchedPrUrls, state, prStateScan));
+  const owedWatchPrUrls = resolveOwedWatchPrs(withoutDelegatedPrs(unwatchedPrUrls, state, Date.now()), state, prStateScan);
+  const decision = watchGateDecision(state, owedWatchPrUrls);
   if (decision.action === "release" /* Release */) return releaseGate("watchReleased");
   if (decision.action === "none" /* None */) return "{}";
   recordGateBlock("watchBlockCount", decision.blockCount);
@@ -3359,7 +3361,7 @@ function buildFollowthroughGate() {
   if (decision.action === "release" /* Release */) return releaseGate("buildSkipped");
   if (decision.action === "none" /* None */) return "{}";
   recordGateBlock("buildBlockCount", decision.blockCount);
-  const reason = decision.blockCount === 1 ? `Do not end the turn yet. This session took a build/implement/fix request but no PR was opened. A root cause written into the transcript ships nothing \u2014 once the session ends it is gone, and the watcher gate never fires because it only looks at PRs that exist. Carry the work to a PR via ${muggleSkill("/muggle-do")}, which runs requirements \u2192 build \u2192 impact \u2192 unit tests \u2192 E2E \u2192 PR \u2192 watcher. If no PR is owed here (the user changed their mind, the fix landed in another repo, the answer was advice rather than a change), tell the user why and run \`echo "MUGGLE_BUILD_SKIP: <reason>"\` \u2014 that records the skip and keeps this gate quiet for the rest of the session.` : `Build request still unanswered \u2014 no PR opened (reminder ${decision.blockCount}/${MAX_BUILD_BLOCKS}): carry it to a PR via ${muggleSkill("/muggle-do")}, or record a legitimate skip via \`echo "MUGGLE_BUILD_SKIP: <reason>"\`.`;
+  const reason = decision.blockCount === 1 ? `Do not end the turn yet. This session took a build/implement/fix request but no PR was opened. A root cause written into the transcript ships nothing \u2014 once the session ends it is gone, and the watcher gate never fires because it only looks at PRs that exist. Carry the work to a PR via ${muggleSkill("/muggle-do")}, which runs requirements \u2192 build \u2192 ${MUGGLE_DO_STAGES_AFTER_BUILD}. If no PR is owed here (the user changed their mind, the fix landed in another repo, the answer was advice rather than a change), tell the user why and run \`echo "MUGGLE_BUILD_SKIP: <reason>"\` \u2014 that records the skip and keeps this gate quiet for the rest of the session.` : `Build request still unanswered \u2014 no PR opened (reminder ${decision.blockCount}/${MAX_BUILD_BLOCKS}): carry it to a PR via ${muggleSkill("/muggle-do")}, or record a legitimate skip via \`echo "MUGGLE_BUILD_SKIP: <reason>"\`.`;
   return blockStop(reason, host);
 }
 function walkthroughGate() {
@@ -3482,7 +3484,7 @@ function buildRouter() {
   if (readState(sessionId).buildIntentRouted) return "{}";
   commitStateMutation(sessionId, { kind: "fields" /* Fields */, fields: { buildIntentRouted: true } });
   if (isMuggleDoInvoked) return "{}";
-  const ctx = `This looks like a build/implement/fix request. Per the autoRouteBuildToMuggleDo preference, route it through ${muggleSkill("/muggle-do")} \u2014 which runs requirements \u2192 build (delegated to superpowers' design\u2192plan\u2192review) \u2192 impact \u2192 unit tests \u2192 E2E \u2192 PR \u2192 watcher. If autoRouteBuildToMuggleDo=always, enter that flow; if =ask, offer it; if =never, proceed normally.`;
+  const ctx = `This looks like a build/implement/fix request. Per the autoRouteBuildToMuggleDo preference, route it through ${muggleSkill("/muggle-do")} \u2014 which runs requirements \u2192 build (delegated to superpowers' design\u2192plan\u2192review) \u2192 ${MUGGLE_DO_STAGES_AFTER_BUILD}. If autoRouteBuildToMuggleDo=always, enter that flow; if =ask, offer it; if =never, proceed normally.`;
   return envelope("UserPromptSubmit", ctx, host);
 }
 function closingAssistantText() {

package/dist/plugin/skills/do/background-e2e.md

@@ -34,7 +34,7 @@ The agent knows only its brief. Pass every value itself, never a pointer to the
 
 - The change's slug and absolute session directory.
 - Repo, worktree path, branch, base branch, and the head SHA under test — for a wave, one entry per member, from its `Wave members` block.
-- The PR URL when the PR already exists (a repair re-run, or a PR opened outside `muggle-do`) — for a wave, every member's. Muggle reads PR URLs from the brief to know which PR's walkthrough is waiting on this run, so leave none out and add none that belong to another change.
+- The change's PR URL — for a wave, every member's. Stage 5.5 opened it before this run; only an `autoCreatePR` skip leaves none. Muggle reads PR URLs from the brief to know which PR's walkthrough is waiting on this run, so leave none out and add none that belong to another change.
 - `Validation`, `localUrl`, `projectId`, and test-credential status from `state.md`.
 - The `e2eScope` from impact analysis, verbatim; a Stage 7.5 re-run passes only its affected test cases.
 - The acceptance criteria from `requirements.md`.
@@ -63,7 +63,7 @@ On the agent's completion notification, the main session:
 
 1. **Checks freshness.** Every `Tested:` line must name its branch's current head. A branch that moved since dispatch — any member's, for a wave — makes the report stale: queue the change again and post nothing.
 2. **Records it** — `E2E dispatch: reported — …` in `state.md`.
-3. **Continues that change from Stage 7**, in its worktree. [`open-prs/forward.md`](open-prs/forward.md) opens its PR with this report — every member's PR for a wave, in the wave order of [`wave-e2e.md`](wave-e2e.md) Stage 7; when a PR already exists, the report settles that PR's walkthrough through Mode A with its number and repo passed explicitly — never from the current branch, since the session's cwd may be another change's worktree.
+3. **Continues that change from Stage 7**, in its worktree. [`open-prs/forward.md`](open-prs/forward.md) settles the walkthrough of the PR Stage 5.5 opened with this report — every member's PR for a wave, per [`wave-e2e.md`](wave-e2e.md) Stage 7 — through Mode A with its number and repo passed explicitly, never from the current branch, since the session's cwd may be another change's worktree.
 4. **Runs Stage 7.5** inline. A repair pass's re-run returns to this lane, behind any local run in flight.
 5. **Arms the change's watcher** once 7.5 clears (Stage 8).
 6. **Dispatches the next queued change.**
@@ -76,7 +76,7 @@ A session that fixed bugs by hand and opened each PR itself follows the same lan
 
 ## Enforcement
 
-- Dispatching a background acceptance-tester holds the E2E and debug-path Stop gates off for a bounded window, and for that window the walkthrough gate sets aside the PRs its brief names — the session is never steered into a second browser run alongside the agent's.
+- Dispatching a background acceptance-tester holds the E2E and debug-path Stop gates off for a bounded window, and for that window the walkthrough and watch gates set aside the PRs its brief names — the session is never steered into a second browser run alongside the agent's, nor into arming a watch on a PR whose run has not come back.
 - A walkthrough posted to one PR settles that PR only; every other PR the session opened is still checked for its own.
 - A skip declaration settles only the PRs of the change it is declared in — the working tree's repo and branch: the PRs opened for that branch and its branch's PR, never another change's, wherever the session's directory sat when a PR was opened.
 - Every PR's walkthrough check stays red until its slot is settled, so a change whose run never came back cannot pass as tested.

package/dist/plugin/skills/do/build.md

@@ -33,7 +33,7 @@ For each affected repo:
 
 ## Delegation
 
-For a non-trivial change — multiple files, real design surface, or anything you would otherwise brainstorm before coding — run the implementation through superpowers' design → plan → subagent-driven build, then return to this stage's Output. That is a runtime hand-off (an action), not a doc dependency; do not encode superpowers' internals here. Routing a build request into this pipeline (the `autoRouteBuildToMuggleDo` front-door guardrail) exists to combine superpowers' design rigor with this pipeline's impact analysis, E2E, PR, and watcher — neither delivers both alone.
+For a non-trivial change — multiple files, real design surface, or anything you would otherwise brainstorm before coding — run the implementation through superpowers' design → plan → subagent-driven build, then return to this stage's Output. That is a runtime hand-off (an action), not a doc dependency; do not encode superpowers' internals here. Routing a build request into this pipeline (the `autoRouteBuildToMuggleDo` front-door guardrail) exists to combine superpowers' design rigor with this pipeline's impact analysis, PR, E2E, and watcher — neither delivers both alone.
 
 ## Output
 

package/dist/plugin/skills/do/e2e-acceptance.md

@@ -50,6 +50,7 @@ You receive everything from `state.md` already — pre-flight resolved it:
 - The list of changed repos, files, and a summary of changes
 - The requirements goal
 - The `e2eScope` from impact analysis — the impacted, retired, and new use cases and test cases this stage runs
+- In a forward run, the PR Stage 5.5 opened from the working tree — the run's [PR progress](../_shared/dev-loop/run.md#pr-progress) shows on it, and Stage 7 settles its walkthrough
 
 ## Your Job
 
@@ -79,9 +80,9 @@ A scope with deliverable checks and no browser case runs none of these and goes
 2. **Auth** — `muggle-remote-auth-status` must be `authenticated`; else escalate.
 3. **Identity tenant/domain match** — if test credentials were marked `existing`, confirm the repo's configured identity tenant/domain matches the recorded tenant/domain. Mismatch → halt.
 
-**Readiness halt.** The cycle planned a browser run and still owes it, so an app that cannot be brought up halts the cycle and is never a `SKIPPED` exit: declare no skip code, write no waiver, run nothing downstream. Report the URL probed, what `muggle-test-prepare` tried (the start command and how it failed), and that no PR was opened or updated; then end the turn and wait for the user. This holds for unattended runs and for a request that asked for the PR. A forward cycle opens no PR, posts no walkthrough, runs no Stage 7.5 and arms no watcher; a watcher cycle pushes nothing and escalates through its own path, which keeps the watcher running. A run that failed because the app never answered (connection refused, page never loaded) returns to item 1, and still down is this halt. A Stop-hook reminder that E2E is still owed does not lift it.
+**Readiness halt.** The cycle planned a browser run and still owes it, so an app that cannot be brought up halts the cycle and is never a `SKIPPED` exit: declare no skip code, write no waiver, run nothing downstream. Report the URL probed, what `muggle-test-prepare` tried (the start command and how it failed), and that the PR's walkthrough is unsettled; then end the turn and wait for the user. This holds for unattended runs and for a request that asked for the PR. A forward cycle leaves its PR open with the walkthrough unsettled, so its walkthrough check stays red: it posts no walkthrough, runs no Stage 7.5 and arms no watcher. Nothing more is owed that PR until the app answers, so run `echo "MUGGLE_WALKTHROUGH_SKIP: readiness halt — <PR URL>"` and `echo "MUGGLE_WATCH_SKIP: readiness halt — <PR URL>"`, which keep the Stop gates from pressing for a walkthrough or watch the halt withholds. A watcher cycle pushes nothing and escalates through its own path, which keeps the watcher running. A run that failed because the app never answered (connection refused, page never loaded) returns to item 1, and still down is this halt. A Stop-hook reminder that E2E is still owed does not lift it.
 
-Only the app not answering is this halt. A run the runner itself lost while readiness passes — an Electron or MCP crash, a Studio abort before the first step — is a failing run, however often it repeats: record it in the Failed block and go on to Stage 7, which opens the PR `[E2E FAILING]` for Stage 7.5 to classify.
+Only the app not answering is this halt. A run the runner itself lost while readiness passes — an Electron or MCP crash, a Studio abort before the first step — is a failing run, however often it repeats: record it in the Failed block and go on to Stage 7, which retitles the PR `[E2E FAILING]` for Stage 7.5 to classify.
 
 ### Step 1: Authentication already verified