@pushpalsdev/cli 1.1.23 → 1.1.24

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pushpalsdev/cli",
-  "version": "1.1.23",
+  "version": "1.1.24",
   "description": "PushPals terminal CLI for LocalBuddy -> RemoteBuddy orchestration",
   "license": "MIT",
   "repository": {

package/runtime/sandbox/.pushpals-remotebuddy-fallback.js

@@ -8285,6 +8285,7 @@ function buildWorkerSpawnCommand(options) {
 }
 
 // apps/remotebuddy/src/remotebuddy_main.ts
+var AUTONOMY_TASK_DEDUPE_COOLDOWN_MS = 6 * 60 * 60 * 1000;
 var CONFIG = loadPushPalsConfig();
 function parseArgs() {
   const args = process.argv.slice(2);
@@ -8464,6 +8465,11 @@ function buildTaskExecuteDedupeKey(sessionId, params) {
   }
   return `task.execute:${normalizedOrigin}:${normalizedSessionId}:${uniqueTargets.join("|")}`.toLowerCase();
 }
+function resolveTaskExecuteDedupeCooldownMs(params, dedupeKey) {
+  if (!dedupeKey)
+    return 0;
+  return params.origin === "autonomy" ? AUTONOMY_TASK_DEDUPE_COOLDOWN_MS : 0;
+}
 function parseAutonomyRequestMetadata(value) {
   let root = asObject2(value);
   if (!root && typeof value === "string") {
@@ -9509,6 +9515,9 @@ Please reply with the missing details and I will enqueue a follow-up request.` :
       const dedupeKey = buildTaskExecuteDedupeKey(sessionId, params);
       if (dedupeKey)
         payload.dedupeKey = dedupeKey;
+      const dedupeCooldownMs = resolveTaskExecuteDedupeCooldownMs(params, dedupeKey);
+      if (dedupeCooldownMs > 0)
+        payload.dedupeCooldownMs = dedupeCooldownMs;
       if (targetWorkerId)
         payload.targetWorkerId = targetWorkerId;
       const res = await this.fetchImpl(`${this.server}/jobs/enqueue`, {
@@ -10603,6 +10612,7 @@ if (import.meta.main) {
   });
 }
 export {
+  resolveTaskExecuteDedupeCooldownMs,
   extractRequiredValidationStepsFromVisionMarkdown,
   buildTaskExecuteDedupeKey,
   RemoteBuddyOrchestrator

package/runtime/sandbox/apps/workerpals/src/backends/openai_codex/openai_codex_executor.py

@@ -109,11 +109,13 @@ _MAX_CREDIBLE_WRAPPER_LOOP_TOP_LEVELS = 4
 _MAX_NO_EDIT_RECOVERY_ATTEMPTS = 1
 _MAX_ROLLOUT_RECOVERY_ATTEMPTS = 1
 _DEFAULT_NO_EDIT_WATCHDOG_S = 480
-_SMALL_TASK_NO_EDIT_WATCHDOG_S = 360
+_SMALL_TASK_NO_EDIT_WATCHDOG_S = 240
+_NARROW_TEST_TASK_NO_EDIT_WATCHDOG_S = 180
 _WEB_REVIEW_NO_EDIT_WATCHDOG_S = 240
 _DEFAULT_NO_EDIT_RECHECK_S = 120
 _DEFAULT_ROLLOUT_WATCHDOG_S = 300
 _SMALL_TASK_ROLLOUT_WATCHDOG_S = 240
+_NARROW_TEST_TASK_ROLLOUT_WATCHDOG_S = 150
 _WEB_REVIEW_ROLLOUT_WATCHDOG_S = 180
 
 
@@ -591,6 +593,21 @@ def _looks_like_small_task_prompt(prompt: str) -> bool:
         "browser smoke",
         "web delivery",
         "navigation trustworthy",
+        "test-only",
+        "test only",
+        "contract test",
+        "contract coverage",
+        "ranking contract",
+        "focused scenario",
+        "targeted test",
+        "one-file",
+        "one file",
+        "single-file",
+        "single file",
+        "max_files_to_edit: 1",
+        "max_files_to_edit=1",
+        "maxfilestoedit: 1",
+        "maxfilestoedit=1",
     )
     heavy_markers = (
         "merge-conflict",
@@ -607,6 +624,34 @@ def _looks_like_small_task_prompt(prompt: str) -> bool:
     )
 
 
+def _looks_like_narrow_test_task_prompt(prompt: str) -> bool:
+    text = str(prompt or "").lower()
+    if not text:
+        return False
+    narrow_markers = (
+        "contract test",
+        "contract coverage",
+        "ranking contract",
+        "test-only",
+        "test only",
+        "targeted test",
+        "focused scenario",
+    )
+    if not any(marker in text for marker in narrow_markers):
+        return False
+    broad_markers = (
+        "full render harness",
+        "full-surface",
+        "full surface",
+        "e2e",
+        "browser validation",
+        "browser smoke",
+        "migration",
+        "broad refactor",
+    )
+    return not any(marker in text for marker in broad_markers)
+
+
 def _resolve_task_reasoning_effort(
     configured_effort: str,
     prompt: str,
@@ -652,7 +697,9 @@ def _resolve_no_edit_watchdog_seconds(
         return None
 
     prompt_text = str(prompt or "").lower()
-    if "repo-native web review" in prompt_text or "web review path" in prompt_text:
+    if _looks_like_narrow_test_task_prompt(prompt):
+        default_s = _NARROW_TEST_TASK_NO_EDIT_WATCHDOG_S
+    elif "repo-native web review" in prompt_text or "web review path" in prompt_text:
         default_s = _WEB_REVIEW_NO_EDIT_WATCHDOG_S
     else:
         default_s = (
@@ -703,7 +750,9 @@ def _resolve_rollout_watchdog_seconds(
         else:
             return max(1, min(parsed, max(1, communicate_timeout_s - 1)))
 
-    if _looks_like_web_review_prompt(prompt):
+    if _looks_like_narrow_test_task_prompt(prompt):
+        default_s = _NARROW_TEST_TASK_ROLLOUT_WATCHDOG_S
+    elif _looks_like_web_review_prompt(prompt):
         default_s = _WEB_REVIEW_ROLLOUT_WATCHDOG_S
     elif _looks_like_small_task_prompt(prompt):
         default_s = _SMALL_TASK_ROLLOUT_WATCHDOG_S
@@ -766,6 +815,8 @@ def _describe_publishable_paths(paths: List[str]) -> str:
 def _build_no_edit_recovery_guidance(trace_excerpt: str, artifact_only_paths: str = "") -> str:
     lines = [
         "No-edit watchdog recovery: the previous Codex attempt spent too much of the execution budget without producing publishable file changes.",
+        "This recovery attempt has a patch-first contract: make one publishable edit before any further broad discovery. If you need one narrow read of the hinted file to place the edit, do that once, then patch immediately.",
+        "Do not repeat the same read/search sequence from the previous attempt. Re-reading the target without editing is a failed recovery.",
         "Start from the already inspected context. Do not re-read broad repo topology, route wrappers, or missing test infrastructure unless that is the blocker.",
         "Runtime/dependency artifacts such as node_modules, outputs, .worktrees, .codex, dist, build, and coverage do not count as progress.",
         "Within the first response/action, edit the smallest behavior-owning file that satisfies the task. If the hinted file is a thin wrapper, patch the owner you already identified.",

package/runtime/sandbox/apps/workerpals/src/backends/openai_codex/test_openai_codex_runtime_config.py

@@ -1386,6 +1386,16 @@ class OpenAICodexRuntimeConfigTests(unittest.TestCase):
 
         self.assertEqual(watchdog_s, 240)
 
+    def test_narrow_contract_tests_use_fast_no_edit_watchdog(self) -> None:
+        prompt = (
+            "Update app/__tests__/opportunity-graph.contract.test.ts to tighten the "
+            "ranking contract test. Keep this test-only and preserve existing behavior."
+        )
+        with mock.patch.dict(os.environ, {"WORKERPALS_OPENAI_CODEX_NO_EDIT_WATCHDOG_S": ""}, clear=False):
+            watchdog_s = _resolve_no_edit_watchdog_seconds(prompt, 1200)
+
+        self.assertEqual(watchdog_s, 180)
+
     def test_no_edit_recovery_guidance_warns_against_artifact_only_progress(self) -> None:
         guidance = _build_no_edit_recovery_guidance(
             "item.completed | still inspecting",
@@ -1393,6 +1403,8 @@ class OpenAICodexRuntimeConfigTests(unittest.TestCase):
         )
 
         self.assertIn("node_modules", guidance)
+        self.assertIn("patch-first contract", guidance)
+        self.assertIn("Re-reading the target without editing is a failed recovery", guidance)
         self.assertIn("do not invent PushPals/autonomy-specific files", guidance)
         self.assertIn("Previous Codex event trace excerpt", guidance)
 
@@ -1411,6 +1423,15 @@ class OpenAICodexRuntimeConfigTests(unittest.TestCase):
         self.assertEqual(no_edit_s, 240)
         self.assertEqual(rollout_s, 180)
 
+    def test_narrow_contract_rollout_watchdog_is_earlier_than_no_edit_watchdog(self) -> None:
+        prompt = "Tighten the focused contract test for one ranking behavior."
+        with mock.patch.dict(os.environ, {"WORKERPALS_OPENAI_CODEX_ROLLOUT_WATCHDOG_S": ""}, clear=False):
+            no_edit_s = _resolve_no_edit_watchdog_seconds(prompt, 1200)
+            rollout_s = _resolve_rollout_watchdog_seconds(prompt, 1200, no_edit_s)
+
+        self.assertEqual(no_edit_s, 180)
+        self.assertEqual(rollout_s, 120)
+
     def test_offtrack_rollout_detects_missing_path_and_harness_drift(self) -> None:
         trace = {
             "summaries": [

package/runtime/sandbox/apps/workerpals/src/workerpals_main.ts

@@ -1785,9 +1785,10 @@ async function workerLoop(
                   terminalStage: currentJobPhase ?? (result.ok ? "completed" : "worker"),
                   executorBackend: resolveExecutor(CONFIG),
                   summary: result.summary,
-                  watchdogFired: /timed out|timeout|signal 15|terminated|exit 143|exit 137/i.test(
-                    `${result.summary}\n${result.stderr ?? ""}`,
-                  ),
+                  watchdogFired:
+                    /watchdog|rollout coach|timed out|timeout|signal 15|terminated|exit 143|exit 137/i.test(
+                      `${result.summary}\n${result.stderr ?? ""}\n${result.stdout ?? ""}`,
+                    ),
                   metadata: {
                     workerId: opts.workerId,
                     docker: Boolean(dockerExecutor),