@algosuite/vo-mcp 0.2.0-beta.36 → 0.2.0-beta.37

package/dist/runner-cli.js

@@ -2922,7 +2922,7 @@ function createControlPlaneClient({
     },
     /**
      * Read the operator's dispatch-mode config (Fast→Ultracode effort setting).
-     * Returns the mode string ('fast'|'standard'|'deep'|'ultra'|'ultracode'),
+     * Returns the mode string ('fast'|'standard'|'deep'|'ultra'|'marathon'; 'ultracode' legacy),
      * defaulting to 'standard' on any error. Never throws — best-effort.
      */
     async getDispatchMode() {
@@ -3284,6 +3284,22 @@ function flushAgentStreamBuffer({
   if (trailing) deliverAgentEvent(parseEvent(trailing), { onProgress, onResult });
   return "";
 }
+function nextDeadlineDecision({
+  nowMs,
+  startMs,
+  lastActivityMs,
+  maxWallClockMs = 0,
+  stallWindowMs = 6e5,
+  legacy = false
+} = {}) {
+  if (!(maxWallClockMs > 0)) return { action: "wait", delayMs: null };
+  const elapsed = nowMs - startMs;
+  if (elapsed < maxWallClockMs) return { action: "wait", delayMs: maxWallClockMs - elapsed };
+  if (legacy) return { action: "kill", stalledForMs: null };
+  const idle = nowMs - lastActivityMs;
+  if (idle >= stallWindowMs) return { action: "kill", stalledForMs: idle };
+  return { action: "wait", delayMs: stallWindowMs - idle };
+}
 function finalizeAgentTaskResult({
   result,
   bin,
@@ -3292,6 +3308,7 @@ function finalizeAgentTaskResult({
   killed,
   cancelReason,
   maxWallClockMs,
+  stalledForMs = null,
   forcedAfterResult,
   code,
   signal,
@@ -3302,7 +3319,8 @@ function finalizeAgentTaskResult({
       ...result,
       ok: false,
       timedOut: true,
-      summary: `wall-clock timeout (${maxWallClockMs}ms)`
+      stalledForMs,
+      summary: stalledForMs != null ? `stalled: no stream activity for ${stalledForMs}ms after the wall-clock deadline (${maxWallClockMs}ms)` : `wall-clock timeout (${maxWallClockMs}ms)`
     };
   }
   if (killed) {
@@ -4123,6 +4141,8 @@ function runAgentTask({
   shouldCancel = async () => false,
   cancelPollMs = 5e3,
   maxWallClockMs = 0,
+  stallWindowMs = Number(env2?.VO_CODE_RUNNER_STALL_WINDOW_MS) > 0 ? Number(env2.VO_CODE_RUNNER_STALL_WINDOW_MS) : 6e5,
+  legacyWallClock = env2?.VO_CODE_RUNNER_LEGACY_WALLCLOCK === "1",
   postResultExitGraceMs = 1e4,
   exitDrainGraceMs = 300,
   armTerminalCleanup = armTerminalProcessCleanup,
@@ -4250,6 +4270,7 @@ function runAgentTask({
         killed,
         cancelReason,
         maxWallClockMs,
+        stalledForMs,
         forcedAfterResult,
         code,
         signal,
@@ -4268,12 +4289,31 @@ function runAgentTask({
         }
       }, 5e3);
     };
-    const wallTimer = maxWallClockMs > 0 ? setTimeout(() => {
-      timedOut = true;
-      clearInterval(poll);
-      hardKill();
-    }, maxWallClockMs) : null;
+    const deadlineStartMs = Date.now();
+    let lastActivityMs = deadlineStartMs;
+    let stalledForMs = null;
+    let wallTimer = null;
+    const armDeadline = () => {
+      const decision = nextDeadlineDecision({
+        nowMs: Date.now(),
+        startMs: deadlineStartMs,
+        lastActivityMs,
+        maxWallClockMs,
+        stallWindowMs,
+        legacy: legacyWallClock
+      });
+      if (decision.action === "kill") {
+        timedOut = true;
+        stalledForMs = decision.stalledForMs;
+        clearInterval(poll);
+        hardKill();
+        return;
+      }
+      if (decision.delayMs != null) wallTimer = setTimeout(armDeadline, decision.delayMs);
+    };
+    armDeadline();
     child.stdout.on("data", (chunk) => {
+      lastActivityMs = Date.now();
       buffer = consumeAgentStreamChunk({
         chunk,
         buffer,
@@ -4283,6 +4323,7 @@ function runAgentTask({
       });
     });
     child.stderr.on("data", (c) => {
+      lastActivityMs = Date.now();
       stderrTail = (stderrTail + c.toString()).slice(-4e3);
     });
     child.on("error", (err) => {
@@ -6007,6 +6048,7 @@ function isMaxTurnsResult(summary) {
 }
 function partialPrTitlePrefix(run = {}) {
   if (isMaxTurnsResult(run.summary)) return "\u26A0 PARTIAL (max turns reached)";
+  if (run.timedOut && run.stalledForMs != null) return "\u26A0 PARTIAL (stalled \u2014 no progress after wall clock)";
   if (run.timedOut) return "\u26A0 PARTIAL (wall-clock timeout)";
   return "\u26A0 PARTIAL (needs continuation)";
 }
@@ -7119,6 +7161,98 @@ var init_dispatch_onboarding = __esm({
   }
 });
 
+// ../../scripts/virtual-office/code-runner/methodology-composer.mjs
+function classifyTaskShape(task) {
+  const prompt = String(task?.prompt || "");
+  for (const rule of SHAPE_RULES) {
+    if (rule.matches(task, prompt)) return rule.shape;
+  }
+  return "feature";
+}
+function composeMethodologyBlock(task) {
+  const shape = classifyTaskShape(task);
+  const lines = [
+    `## Methodology (auto-composed: ${shape})`,
+    ...UNIVERSAL_DIRECTIVES.map((d) => `- ${d}`),
+    ...(SHAPE_DIRECTIVES[shape] || []).map((d) => `- ${d}`)
+  ];
+  return { shape, block: lines.join("\n") };
+}
+function withMethodology(prompt, task) {
+  const { shape, block } = composeMethodologyBlock(task);
+  return { shape, prompt: `${prompt ?? ""}
+
+${block}` };
+}
+var SHAPE_RULES, UNIVERSAL_DIRECTIVES, SHAPE_DIRECTIVES;
+var init_methodology_composer = __esm({
+  "../../scripts/virtual-office/code-runner/methodology-composer.mjs"() {
+    "use strict";
+    SHAPE_RULES = [
+      {
+        shape: "recovery",
+        matches: (task, prompt) => Boolean(task?.resumed_from) || /VO_RECOVERY_FROM_CODE_TASK/u.test(prompt) || /previous run stopped before completion/iu.test(prompt)
+      },
+      {
+        shape: "pr-repair",
+        matches: (task) => typeof task?.repair_pr_number === "number"
+      },
+      {
+        shape: "bug-fix",
+        matches: (task, prompt) => Boolean(task?.bug_id) || /^reproduce and fix\b/iu.test(prompt) || /captured by: qa sweep/iu.test(prompt)
+      },
+      {
+        shape: "roadmap-advance",
+        matches: (task, prompt) => Boolean(task?.roadmap_app || task?.roadmap_item_id || typeof task?.roadmap_phase_index === "number") || /^roadmap:/iu.test(prompt)
+      },
+      {
+        shape: "research",
+        matches: (_task, prompt) => /\b(investigate|research|root[- ]cause|audit|diagnose|find out why|explain why)\b/iu.test(prompt)
+      },
+      {
+        shape: "design",
+        matches: (_task, prompt) => /\b(design doc|architecture|architect|adr\b|lane brief|write a plan|propose (a|the) (design|plan|approach))\b/iu.test(prompt)
+      },
+      {
+        shape: "chore",
+        matches: (_task, prompt) => prompt.length < 400 && /\b(typo|rename|bump|readme|changelog|comment|reword|lint fix|formatting)\b/iu.test(prompt)
+      }
+    ];
+    UNIVERSAL_DIRECTIVES = [
+      "Verification is a stage, not a vibe: before publishing, run the tests/build your change touches and cite their actual output. A claim without execution evidence is not done.",
+      "Work to completion or end with an explicit failure reason. Do not stop because time has passed; stop when the evidence says the work is done \u2014 or state exactly what is blocking.",
+      "Default to doing the work yourself in this session. Spawn parallel subagents ONLY for pieces that are genuinely independent and independently verifiable \u2014 and verify their results yourself before integrating; never let unreviewed parallel output merge into shared files."
+    ];
+    SHAPE_DIRECTIVES = {
+      "bug-fix": [
+        "Reproduce first: write the check that fails because of this bug, prove it fails, then fix, then prove the same check passes. A fix without a failing-then-passing check is not a fix."
+      ],
+      research: [
+        "Every claim needs its source AND exact attribution \u2014 which file:line, which benchmark, which baseline, which version. Verify attribution, not just that a source exists; misattributed real facts are the dominant research failure mode.",
+        "Deliver findings as a repo artifact (docs/) with the evidence inline, not only as chat output."
+      ],
+      "roadmap-advance": [
+        "Update the roadmap doc status, regenerate the roadmap board if the doc changed, and add the roadmap-log fragment IN THIS SAME PR \u2014 a roadmap task that does not move the roadmap did not happen."
+      ],
+      design: [
+        "Produce the plan artifact (docs/lanes/ or docs/adr/ with a Related section) BEFORE writing code. State the requirements you are designing to at the top; ambiguity resolved now is rework avoided later."
+      ],
+      recovery: [
+        "Restore the preserved context first (draft PR, branch, checkpoint notes) and finish the ORIGINAL spec. Do not redo work that is already committed; verify what exists, then close the gap."
+      ],
+      "pr-repair": [
+        "Work from the exact materialized PR source. Never force-push or rebase the existing branch; publish the replacement and let the host close the original."
+      ],
+      chore: [
+        "Keep the diff minimal and mechanical. No fan-out, no speculative refactors; the verification stage is still required."
+      ],
+      feature: [
+        "Ship the tests that prove the feature works in the same change, to the output-verified standard (assert the correct answer, not that something rendered)."
+      ]
+    };
+  }
+});
+
 // ../../scripts/virtual-office/code-runner/task-prompt.mjs
 function buildMissingKnowledgeMessage(taskId, reason) {
   const id = taskId || "unknown-task";
@@ -7162,11 +7296,16 @@ function withAttachmentManifest(prompt, markdown) {
 
 ${manifest}` : prompt;
 }
+function withComposedMethodology(prompt, task, log2, taskId) {
+  const { shape, prompt: composed } = withMethodology(prompt, task);
+  log2(`task ${taskId || "unknown-task"}: methodology shape=${shape}`);
+  return composed;
+}
 async function composeCodeTaskPrompt(client, task, { log: log2 = () => {
 }, allowMissingKnowledgeContext = false, attachmentManifestMarkdown = "" } = {}) {
   const taskId = task?.code_task_id;
   if (!taskId) {
-    return composeDispatchPrompt(withAttachmentManifest(task?.prompt, attachmentManifestMarkdown), {
+    return composeDispatchPrompt(withComposedMethodology(withAttachmentManifest(task?.prompt, attachmentManifestMarkdown), task, log2, taskId), {
       repo: task?.repo,
       knowledgeContextMarkdown: handleMissingKnowledgeContext(taskId, "missing code_task_id on the claimed task", {
         allowMissingKnowledgeContext,
@@ -7175,7 +7314,7 @@ async function composeCodeTaskPrompt(client, task, { log: log2 = () => {
     });
   }
   if (typeof client?.getTaskKnowledgeContext !== "function") {
-    return composeDispatchPrompt(withAttachmentManifest(task?.prompt, attachmentManifestMarkdown), {
+    return composeDispatchPrompt(withComposedMethodology(withAttachmentManifest(task?.prompt, attachmentManifestMarkdown), task, log2, taskId), {
       repo: task?.repo,
       knowledgeContextMarkdown: handleMissingKnowledgeContext(taskId, "control-plane client cannot fetch knowledge context", {
         allowMissingKnowledgeContext,
@@ -7204,7 +7343,7 @@ async function composeCodeTaskPrompt(client, task, { log: log2 = () => {
   const prompt = operatorInstructions ? `${task?.prompt ?? ""}
 
 ${operatorInstructions}` : task?.prompt;
-  return composeDispatchPrompt(withAttachmentManifest(prompt, attachmentManifestMarkdown), {
+  return composeDispatchPrompt(withComposedMethodology(withAttachmentManifest(prompt, attachmentManifestMarkdown), task, log2, taskId), {
     repo: task?.repo,
     knowledgeContextMarkdown
   });
@@ -7214,6 +7353,7 @@ var init_task_prompt = __esm({
   "../../scripts/virtual-office/code-runner/task-prompt.mjs"() {
     "use strict";
     init_dispatch_onboarding();
+    init_methodology_composer();
     ALLOW_MISSING_KNOWLEDGE_CONTEXT_ENV = "VO_CODE_RUNNER_ALLOW_MISSING_KNOWLEDGE_CONTEXT";
   }
 });
@@ -9763,7 +9903,8 @@ var init_control_server = __esm({
 // ../../scripts/virtual-office/code-runner/effort-mode-config.mjs
 function resolveEffortMode(mode) {
   const normalized = String(mode || "").trim().toLowerCase();
-  return EFFORT_MODE_CONFIG[normalized] || EFFORT_MODE_CONFIG[DEFAULT_MODE];
+  const canonical = LEGACY_MODE_ALIASES[normalized] || normalized;
+  return EFFORT_MODE_CONFIG[canonical] || EFFORT_MODE_CONFIG[DEFAULT_MODE];
 }
 function resolveDefaultBudgetUsd(env2 = {}) {
   const raw = env2?.[DEFAULT_BUDGET_USD_ENV];
@@ -9793,7 +9934,7 @@ ${effortConfig.multiAgentInstruction}
   parts.push(String(basePrompt || "").trim());
   return parts.join("\n");
 }
-var RED_TEAM_DIRECTIVE, DEFAULT_BUDGET_USD_ENV, EFFORT_MODE_CONFIG, DEFAULT_MODE;
+var RED_TEAM_DIRECTIVE, DEFAULT_BUDGET_USD_ENV, EFFORT_MODE_CONFIG, DEFAULT_MODE, LEGACY_MODE_ALIASES;
 var init_effort_mode_config = __esm({
   "../../scripts/virtual-office/code-runner/effort-mode-config.mjs"() {
     "use strict";
@@ -9832,7 +9973,7 @@ var init_effort_mode_config = __esm({
         thinkingDirective: `Think step-by-step. Exhaustively verify every assumption against source code and documentation. ${RED_TEAM_DIRECTIVE}`,
         multiAgentInstruction: "If this task needs multiple phases (research, build, verify), propose a plan first."
       },
-      ultracode: {
+      marathon: {
         tier: "best",
         maxBudgetUsd: null,
         permissionMode: "acceptEdits",
@@ -9842,6 +9983,7 @@ var init_effort_mode_config = __esm({
       }
     };
     DEFAULT_MODE = "standard";
+    LEGACY_MODE_ALIASES = { ultracode: "marathon" };
   }
 });