@mrrlin-dev/mcp 0.3.46 → 0.3.47

package/dist/bin.cjs

@@ -36550,6 +36550,7 @@ var mrrlinAsyncJobSchema = external_exports.object({
   failureReason: mrrlinAsyncJobFailureReasonSchema.nullable().default(null),
   requestedHash: external_exports.string().nullable().default(null),
   finalReviewedHash: external_exports.string().nullable().default(null),
+  supersededAt: external_exports.string().nullable().default(null),
   attemptCount: external_exports.number().int().nonnegative(),
   leaseId: external_exports.string().nullable().default(null),
   claimedAt: external_exports.string().nullable().default(null),
@@ -36714,7 +36715,7 @@ var mrrlinDeliverablePayloadSchema = external_exports.object({
   contentHash: external_exports.string().min(1),
   format: external_exports.enum(["markdown"])
 });
-var SUPPORTED_EXECUTION_DOMAINS = ["code"];
+var SUPPORTED_EXECUTION_DOMAINS = ["code", "seo"];
 function isExecutionDomainSupported(domain2) {
   return SUPPORTED_EXECUTION_DOMAINS.includes(domain2);
 }
@@ -37510,7 +37511,8 @@ var mrrlinProjectSchedulerPolicySchema = external_exports.object({
   prodGraceSeconds: external_exports.number().int().nonnegative(),
   projectSlug: mrrlinProjectSlugSchema,
   updatedAt: external_exports.string().datetime(),
-  wakeFloor: external_exports.string().datetime().nullable()
+  wakeFloor: external_exports.string().datetime().nullable(),
+  maxConsecutiveJobFailures: external_exports.number().int().positive()
 });
 var mrrlinProjectSchedulerPolicyUpsertSchema = external_exports.object({
   autoSpecApproval: mrrlinAutoSpecApprovalSchema.optional(),
@@ -37519,7 +37521,8 @@ var mrrlinProjectSchedulerPolicyUpsertSchema = external_exports.object({
   eligibleAutonomyLevels: external_exports.array(mrrlinAutonomyLevelSchema).min(1).optional(),
   maxConcurrentAutonomousRuns: external_exports.number().int().nonnegative().optional(),
   prodGraceSeconds: external_exports.number().int().nonnegative().optional(),
-  wakeFloor: external_exports.string().datetime().nullable().optional()
+  wakeFloor: external_exports.string().datetime().nullable().optional(),
+  maxConsecutiveJobFailures: external_exports.number().int().positive().optional()
 }).refine((value) => Object.keys(value).length > 0, {
   message: "ProjectSchedulerPolicyUpsert requires at least one field"
 });
@@ -37680,7 +37683,10 @@ var mrrlinInboxItemActionPayloadSchema = external_exports.object({
   runId: mrrlinExecutionRunIdSchema,
   // Optional (additive): when true, approving the proposal creates a full code-execution run
   // (worktree + push/PR path). Scheduler proposals set it; legacy payloads omit it.
-  codeExecution: external_exports.boolean().optional()
+  codeExecution: external_exports.boolean().optional(),
+  // Optional scheduler recovery action carried by an info item.
+  retryToolName: external_exports.string().min(1).optional(),
+  retryDedupKey: external_exports.string().min(1).optional()
 });
 var mrrlinInboxQuestionOptionSchema = external_exports.object({
   answer: external_exports.string().trim().min(1),
@@ -38401,6 +38407,7 @@ function createMrrlinClient(config2) {
     },
     async listInboxItems(projectSlug, filters = {}) {
       const url2 = appendQuery(`${projectPath(projectSlug)}/inbox`, {
+        groupId: filters.groupId,
         kind: filters.kind,
         readState: filters.readState,
         status: filters.status,
@@ -39110,7 +39117,7 @@ function startCodexProbe(deps = {}) {
 }
 
 // src/_generated/version.ts
-var PKG_VERSION = "0.3.46";
+var PKG_VERSION = "0.3.47";
 
 // src/api-base-url.ts
 var DEFAULT_API_BASE_URL = "http://127.0.0.1:8787";
@@ -39269,14 +39276,14 @@ function startSchedulerTickLoop(deps) {
   let stopped = false;
   let timer = null;
   const env = deps.env ?? process.env;
-  const schedule = (delayMs) => {
+  const schedule = (delayMs2) => {
     if (stopped) return;
     timer = setTimeout(() => {
       void fireSchedulerTick(deps).then(({ nextWakeAt }) => {
         if (stopped) return;
         schedule(computeNextTickDelayMs(nextWakeAt, Date.now(), { env }));
       });
-    }, delayMs);
+    }, delayMs2);
     timer.unref?.();
   };
   schedule(computeNextTickDelayMs((/* @__PURE__ */ new Date()).toISOString(), Date.now(), { env }));
@@ -41358,12 +41365,16 @@ var EXECUTION_TURN_BUDGET_PREFIX = "Execution turn exceeded the run wall-clock b
 var EXECUTION_TURN_TIMEOUT_RETRY_CODE = "execution_turn_timeout_retry";
 var EXECUTION_TURN_TIMEOUT_RETRY_SUMMARY = "Execution turn timed out; requeued for one automatic retry.";
 var EXECUTION_TURN_TIMEOUT_RETRY_CURSOR = "execution_turn_timeout_retry:1";
+var AWAITING_MERGE_STATUS_RETRY_DELAYS_MS = [50, 100];
 function readPositiveIntEnv(name, fallback) {
   const raw = (process.env[name] ?? "").trim();
   if (!raw) return fallback;
   const parsed = Number.parseInt(raw, 10);
   return Number.isInteger(parsed) && parsed > 0 ? parsed : fallback;
 }
+function delayMs(ms) {
+  return new Promise((resolve) => setTimeout(resolve, ms));
+}
 function execTurnIdleTimeoutMs() {
   return readPositiveIntEnv("MRRLIN_DIRECTOR_BRIDGE_EXEC_TURN_IDLE_TIMEOUT_SECONDS", DEFAULT_EXEC_TURN_IDLE_TIMEOUT_SECONDS) * 1e3;
 }
@@ -42387,19 +42398,42 @@ async function reconcileTaskAwaitingMerge(deps, projectSlug, task, runId, reason
     }
   }).catch(() => {
   });
+  let lastError = null;
+  for (let attempt = 1; attempt <= AWAITING_MERGE_STATUS_RETRY_DELAYS_MS.length + 1; attempt += 1) {
+    try {
+      await deps.client.updateTask(projectSlug, task.id, { status: "handoff_needed" });
+      return;
+    } catch (error51) {
+      lastError = error51;
+      const retryDelay = AWAITING_MERGE_STATUS_RETRY_DELAYS_MS[attempt - 1];
+      if (retryDelay !== void 0) {
+        await delayMs(retryDelay);
+      }
+    }
+  }
   try {
-    await deps.client.updateTask(projectSlug, task.id, { status: "handoff_needed" });
-  } catch (error51) {
-    console.warn(
-      JSON.stringify({
-        event: "reconcile_awaiting_merge_failed",
-        projectSlug,
-        taskId: task.id,
-        runId,
-        error: error51 instanceof Error ? error51.message : String(error51)
-      })
-    );
+    await deps.client.updateExecutionRun(projectSlug, runId, { status: "handoff_queued" });
+  } catch {
+  }
+  try {
+    await deps.client.createInboxItem(projectSlug, {
+      id: `II-${import_node_crypto4.default.randomUUID()}`,
+      kind: "handoff",
+      taskId: task.id,
+      body: `Task ${task.id} (${task.title}) still needs operator follow-up: ${reason}. Mrrlin could not move the task back to Handoff needed automatically.`
+    });
+  } catch {
   }
+  console.warn(
+    JSON.stringify({
+      event: "reconcile_awaiting_merge_failed",
+      projectSlug,
+      taskId: task.id,
+      runId,
+      error: lastError instanceof Error ? lastError.message : String(lastError),
+      attempts: AWAITING_MERGE_STATUS_RETRY_DELAYS_MS.length + 1
+    })
+  );
 }
 async function notifyOperatorHandoff(deps, projectSlug, task, runId, reason) {
   try {
@@ -42560,7 +42594,12 @@ async function driveAsyncJob(deps, projectSlug, jobId) {
   let outcome = { kind: "failed", failureReason: "provider_unavailable", error: "did not run" };
   try {
     for (; ; ) {
-      const begun = await deps.client.beginAsyncJobAttempt(projectSlug, jobId, leaseId, deriveHash());
+      const begun = await deps.client.beginAsyncJobAttempt(
+        projectSlug,
+        jobId,
+        leaseId,
+        await deriveRequestedHash(deps.client, claimed.toolName, input)
+      );
       if (!begun) return;
       if (begun.attemptCount > ASYNC_JOB_MAX_ATTEMPTS) {
         outcome = { kind: "failed", failureReason: "provider_unavailable", error: "exceeded retry budget" };
@@ -42611,8 +42650,17 @@ async function driveAsyncJob(deps, projectSlug, jobId) {
   }
   if (descriptor2.wakeDirector) deps.wake(projectSlug, claimed.directorSessionId, delivery.body, delivery.kind);
 }
-function deriveHash() {
-  return `attempt-${Date.now()}`;
+async function deriveRequestedHash(client, toolName, input) {
+  if (input.requestedHash) return input.requestedHash;
+  if (toolName === "run_spec_consensus" && input.pageId) {
+    const page = await client.getWikiPage(input.projectSlug, input.pageId);
+    return page.contentHash ?? import_node_crypto5.default.createHash("sha256").update(page.markdownBody).digest("hex");
+  }
+  if (toolName === "draft_task_spec" && input.taskId) {
+    const task = await client.getTask(input.projectSlug, input.taskId);
+    return task.updatedAt ?? import_node_crypto5.default.createHash("sha256").update(JSON.stringify(task)).digest("hex");
+  }
+  return import_node_crypto5.default.createHash("sha256").update(JSON.stringify(input)).digest("hex");
 }
 async function runAsyncBackstopSweep(deps, drive = (slug, jobId) => driveAsyncJob(deps, slug, jobId)) {
   const projects = await deps.client.listProjects();
@@ -47490,7 +47538,7 @@ function buildThreadStartParams(msg, config2, desiredModel) {
     "  bullets (`- AND ...` as needed). Scenarios make the spec verifiable.",
     "- Keep behavior contracts in the spec; implementation details belong in `## Implementation Plan`.",
     "- Apply this format ONLY to Specs/{id} pages; do not retro-fit it onto other wiki page types.",
-    "- Spec consensus runs ASYNCHRONOUSLY. Call `run_spec_consensus` for a `Specs/{id}` page; it returns IMMEDIATELY. On `already_converged`, the spec is already marked ready \u2014 proceed straight to `create_execution_run`; do NOT call `mark_spec_ready` and do NOT re-run consensus. On `queued` or `already_running`, tell the user it was launched and END your turn \u2014 do NOT poll or call it again; the result arrives in the inbox (an `info` item on CONVERGED, a `handoff` item if it was blocked/failed). When a consensus `info` item says CONVERGED, proceed to `create_execution_run`. When a `handoff` says it was blocked, address the recorded issues in the spec (any edit re-arms the gate) and call `run_spec_consensus` once more. For a NON-code task (domain seo/marketing/content, ADR 0020) the spec is review-only \u2014 do NOT call `create_execution_run` (non-code execution is not yet supported and fail-closes); tell the operator the spec was reviewed.",
+    "- Spec consensus runs ASYNCHRONOUSLY. Call `run_spec_consensus` for a `Specs/{id}` page; it returns IMMEDIATELY. On `already_converged`, the spec is already marked ready \u2014 proceed straight to `create_execution_run`; do NOT call `mark_spec_ready` and do NOT re-run consensus. On `queued` or `already_running`, tell the user it was launched and END your turn \u2014 do NOT poll or call it again; the result arrives in the inbox (an `info` item on CONVERGED, a `handoff` item if it was blocked/failed). When a consensus `info` item says CONVERGED, proceed to `create_execution_run`. When a `handoff` says it was blocked, address the recorded issues in the spec (any edit re-arms the gate) and call `run_spec_consensus` once more. For a NON-code task (domain seo/marketing/content, ADR 0020): supported domains like `seo` proceed to `create_execution_run` after consensus converges; unsupported domains fail-close with `non_code_execution_unsupported`. Report the spec's review outcome to the operator.",
     '- When a task needs a browser (login flows, UI verification/clicks, scraping), pass `prerequisites:[{mcp:"mrrlin-browser",reason:"..."}]` to `create_execution_run`. If that run later fails into a handoff, the operator gets a one-click "Launch browser & resume" that brings up the dedicated browser as the turn starts.',
     "- When you actually drive a browser (calling `browser_navigate`, `browser_click`, `browser_snapshot`, `browser_type`, etc.), invoke those tools from the **`mrrlin-browser`** MCP server ONLY. Do NOT use tools from any other browser MCP (`playwright`, `chrome-devtools`, `puppeteer`, etc.) even if they are also connected on this operator's box \u2014 those use a shared/default Playwright profile that is fragile across runs. Only `mrrlin-browser` uses this project's dedicated Chromium + persistent operator profile (`~/.mrrlin/browser/profile`). If your host exposes multiple browser MCPs with overlapping tool names, prefer the `mrrlin-browser` server unconditionally.",
     "",
@@ -53658,8 +53706,8 @@ async function cachedResult(client, input) {
     }
     return { status: "already_converged", payload: { outcome: "CONVERGED", contentHash: page.contentHash } };
   }
-  if (review?.outcome === "CONSENSUS_BLOCKED") {
-    return { status: "already_reviewed", payload: { outcome: "CONSENSUS_BLOCKED", contentHash: page.contentHash } };
+  if (review?.outcome === "CONSENSUS_BLOCKED" || review?.outcome === "CONSENSUS_BLOCKED_CLARIFICATION") {
+    return { status: "already_reviewed", payload: { outcome: review.outcome, contentHash: page.contentHash } };
   }
   return null;
 }
@@ -53708,6 +53756,27 @@ async function run(ctx, input) {
   }
   if (result.outcome === "CONSENSUS_BLOCKED") {
     const clarifications = result.clarifications ?? [];
+    if (clarifications.length > 0) {
+      try {
+        const page = await ctx.client.getWikiPage(input.projectSlug, input.pageId);
+        if (page) {
+          const domain2 = await resolveSpecDomain(ctx.client, input.projectSlug, {
+            id: page.id,
+            taskId: page.taskId ?? null
+          });
+          await ctx.client.recordSpecConsensusReview(input.projectSlug, input.pageId, {
+            contentHash: domainScopedReviewHash(page.contentHash, domain2),
+            outcome: "CONSENSUS_BLOCKED_CLARIFICATION",
+            rounds: 0
+          });
+        }
+      } catch (error52) {
+        process.stderr.write(
+          `[mrrlin-mcp consensus] clarification verdict write failed for ${input.projectSlug}/${input.pageId}: ${error52 instanceof Error ? error52.message : String(error52)}
+`
+        );
+      }
+    }
     let filedQuestionCount = 0;
     if (clarifications.length > 0) {
       const page = await ctx.client.getWikiPage(input.projectSlug, input.pageId);
@@ -53731,7 +53800,7 @@ async function run(ctx, input) {
               taskId: scopeId,
               body: clarification.question,
               origin: "gate",
-              ...questionOptions ? { questionOptions } : {},
+              questionOptions,
               ...groupId ? { groupId } : {}
             });
           } catch {
@@ -53751,7 +53820,7 @@ async function run(ctx, input) {
               planId: scopeId,
               body: clarification.question,
               origin: "gate",
-              ...questionOptions ? { questionOptions } : {},
+              questionOptions,
               ...groupId ? { groupId } : {}
             });
           } catch {
@@ -54806,7 +54875,7 @@ var mcpToolDescriptions = {
   [mcpToolNames.resetDirectorThread]: "Clear a task's persisted Director execution thread handle so the next execution run starts a fresh provider thread. Records a director_thread_reset task event.",
   [mcpToolNames.updateWikiPage]: "Update a markdown Wiki page.",
   [mcpToolNames.markSpecReady]: "Mark a wiki page (folder=Specs) as ready (or not) for execution-run gating. The Director sets this true after validating the spec; any edit to the page's markdown auto-resets it to false.",
-  [mcpToolNames.runSpecConsensus]: 'Enqueue the consensus review gate over a Specs/{id} page. Returns IMMEDIATELY with one of: already_converged (a CONVERGED review already exists for the current content \u2014 proceed straight to create_execution_run, do NOT call mark_spec_ready, it is already set), already_reviewed (a terminal CONSENSUS_BLOCKED for this content), already_running (a consensus job is in flight), or queued (a new job was launched). The review runs in the background; its result/blocker arrives in the inbox (info on CONVERGED, handoff on failure) and is surfaced into the open chat. Do NOT call again for the same spec and do NOT poll. On queued/already_running, tell the user it was launched and end the turn. Pass extraHeads: ["researcher"] for specs referencing external libraries/APIs, or ["debugger"] for root-cause/investigation specs to add opt-in specialist reviewers on top of the fixed four. For a NON-code task (domain seo/marketing/content, ADR 0020) the spec is REVIEW-ONLY: do NOT call create_execution_run (it fail-closes with non_code_execution_unsupported until Phase 2) \u2014 report that the spec was reviewed and end the turn.',
+  [mcpToolNames.runSpecConsensus]: 'Enqueue the consensus review gate over a Specs/{id} page. Returns IMMEDIATELY with one of: already_converged (a CONVERGED review already exists for the current content \u2014 proceed straight to create_execution_run, do NOT call mark_spec_ready, it is already set), already_reviewed (a terminal CONSENSUS_BLOCKED for this content), already_running (a consensus job is in flight), or queued (a new job was launched). The review runs in the background; its result/blocker arrives in the inbox (info on CONVERGED, handoff on failure) and is surfaced into the open chat. Do NOT call again for the same spec and do NOT poll. On queued/already_running, tell the user it was launched and end the turn. Pass extraHeads: ["researcher"] for specs referencing external libraries/APIs, or ["debugger"] for root-cause/investigation specs to add opt-in specialist reviewers on top of the fixed four. For a NON-code task (domain seo/marketing/content, ADR 0020): supported domains like `seo` proceed to `create_execution_run` after consensus; unsupported domains fail-close with `non_code_execution_unsupported`. Report the spec\'s review outcome to the operator.',
   [mcpToolNames.runCodeReviewerGate]: `Review the current branch's code diff (vs baseRef) with a Code Reviewer gate \u2014 independent reviewers iterate review\u2192fix\u2192re-check (checkpoint, scope-limited reviser edits, typecheck/build fast-gate, rollback) until the diff converges or is blocked. Run before opening a PR. Pass extraHeads: ["security-analyst"] for sensitive diffs (auth, secrets, input handling, new endpoints).`,
   [mcpToolNames.appendExecutionArtifact]: "Append a durable artifact (plan, checkpoint, command output, test result, diff summary, self-review, error, deploy attempt, or handoff judgement) to an execution run.",
   [mcpToolNames.recordProducedUrl]: 'Call this after publishing each web page produced by the task, so the UI link opens the page itself instead of the deployed site root. Pass the page\'s full https URL (for external pages, e.g. "https://dev.to/<slug>") OR a path starting with "/" for the deploy environment (e.g. "/blog/<slug>") \u2014 paths are resolved against the run\'s deployed env URL at render time. Optional `label` (\u226440 chars) is shown next to the link; omit and the UI defaults to the page title. Call once per page; multiple calls dedupe by resolved URL in the UI.',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mrrlin-dev/mcp",
-  "version": "0.3.46",
+  "version": "0.3.47",
   "type": "module",
   "bin": {
     "mrrlin-mcp": "dist/bin.cjs"
@@ -21,17 +21,17 @@
     "@types/ws": "^8.18.1",
     "esbuild": "^0.24.0",
     "tsx": "^4.22.3",
-    "@mrrlin/client": "0.0.0",
     "@mrrlin/director-e2e": "0.0.0",
     "@mrrlin/wiki": "0.0.0",
-    "@mrrlin/schemas": "0.0.0",
+    "@mrrlin/client": "0.0.0",
     "@mrrlin/codex-client": "0.0.0",
+    "@mrrlin/schemas": "0.0.0",
     "@mrrlin/tsconfig": "0.0.0"
   },
   "dependencies": {
     "@iarna/toml": "^2.2.5",
     "@modelcontextprotocol/sdk": "^1.0.0",
-    "@mrrlin-dev/external-agents": "^0.33.1",
+    "@mrrlin-dev/external-agents": "latest",
     "@playwright/mcp": "0.0.76",
     "playwright-core": "1.61.0-alpha-1781023400000",
     "proper-lockfile": "^4.1.2",