denpex-mcp 0.2.32 → 0.2.33

package/dist/index.js

@@ -7989,7 +7989,7 @@ function ul(e) {
   let t = { "flight-recorder": "Collective flight recorder", "memory-snapshot": "CUDA memory snapshot", nsight: "Nsight kernel trace", checkpoints: "Checkpoint discovery", stragglers: "Aligned rank step timings" }, r = { "flight-recorder": "Capture watchdog timeout evidence and aligned dumps from every process-group member. No collective deadlock or culprit can be established from healthy or incomplete snapshots alone.", "memory-snapshot": "Compare the OOM request and available device memory on the affected device. Inspect allocation stacks and awaiting-free memory; test one batch-size, lifetime or allocator change against the same workload. The largest allocation alone does not establish a cause.", nsight: "Inspect the longest kernels and captured device/stream gaps alongside CPU, transfer and communication events. A stream gap alone does not establish GPU-wide idleness.", checkpoints: "Use candidates for read-only investigation. Verify complete DCP shard coverage and successfully load the checkpoint before resuming. Metadata presence or a manifest match alone does not authorize automated rollback.", stragglers: "Compare repeatedly slow ranks with workload shape, CPU timings, clocks and fabric deltas from the same steps. A slow rank alone does not prove defective hardware." };
   return { tier: 0, failureType: e.status === "INCOMPLETE" ? "ARTIFACT_EVIDENCE_INCOMPLETE" : "GPU_ARTIFACT_ANALYSIS", confidence: e.status === "INCOMPLETE" ? 0 : 70, summary: `${t[e.module]}: ${e.status === "COMPLETE" ? "local inspection completed" : "inspection incomplete"}. ${e.findings.length} reported finding(s). This summary does not establish a root cause or certify hardware health.`, action: e.status === "INCOMPLETE" ? "Resolve the reported inspection limitations and submit a fresh bounded JSON summary. No healthy verdict or safe resume point can be established from incomplete inspection." : r[e.module], fallbackAction: "Keep the original artifact locally and collect the crash log plus measurements from the same incident window.", emergencyAction: null, resumeFrom: null, sourceType: "artifact-evidence", evidence: [...e.findings.map((i) => `${i.code}: ${i.detail}`), `Measured artifact summary: ${JSON.stringify(e.metrics).slice(0, 16e3)}`], advisory: e.limitations, forensicReport: { diagnosticArtifacts: [e] } };
 }
-var rr = "2026.10.10+verified-recovery-boundaries-r32-native-artifacts-r2";
+var rr = "2026.10.11+verified-recovery-boundaries-r32-usability-r3";
 var Qb = Object.freeze(["worker/artifact-evidence.ts", "worker/flight-recorder.ts", "worker/hardware-failure-corpus.ts", "worker/hardware-counters.ts", "worker/data/nvidia-xid-catalog.json", "worker/data/nvidia-xid-catalog.generated.ts", "agent/denpex_artifacts.py", "agent/denpex_fabric.py", "agent/denpex_preflight.py", "worker/workload-instruction-safety.ts", "worker/remedy-applicability.ts", "worker/follow-up-context.ts", "worker/diagnosis-worker-optimized.ts", "worker/control-arm-falsification.ts", "worker/diagnosis-governance.ts", "worker/telemetry-parse.ts", "worker/keyword-match-tier.ts", "worker/detectors.ts", "worker/config-fault-detectors.ts", "worker/host-config-detectors.ts", "worker/environment-fault-detectors.ts", "worker/fleet-config-detectors.ts", "worker/infra-fault-detectors.ts", "worker/scheduler-network-detectors.ts", "worker/access-control-detectors.ts", "worker/fabric-config-detectors.ts", "worker/host-runtime-detectors.ts", "worker/pipeline-runtime-detectors.ts", "worker/host-capacity-detectors.ts", "worker/orchestration-precision-detectors.ts", "worker/capacity-transport-detectors.ts", "worker/throughput-runtime-detectors.ts", "worker/policy-configuration-detectors.ts", "worker/adversarial-causal-detectors.ts", "worker/detector-precedence.ts", "worker/incident-fact-ledger.ts", "worker/deep-reasoning-contract.ts", "worker/long-tail-causal-detectors.ts", "worker/precision-causal-detectors.ts", "worker/generalization-causal-detectors.ts", "worker/causal-timeline.ts", "worker/generalization-cycle-3-detectors.ts", "worker/generalization-cycle-4-detectors.ts", "worker/generalization-cycle-4-revealed-regressions.ts", "worker/generalization-cycle-5-detectors.ts", "worker/generalization-cycle-6-detectors.ts", "worker/generalization-cycle-7-revealed-regressions.ts", "worker/agent-challenge-causal-detectors.ts", "worker/clarifying-questions.ts", "worker/investigation-loop.ts", "worker/remediation-engine.ts", "worker/rma-generator.ts", "worker/negated-evidence.ts", "worker/healthy-baseline.ts", "worker/host-memory-evidence.ts", "worker/patterns/registry.ts", "worker/patterns/broad_taxonomy_patterns.ts", "worker/patterns/family-contracts.ts", "worker/patterns/seo_gap_patterns.ts", "worker/patterns/audit_gap_patterns.ts", "worker/patterns/cycle9_gap_patterns.ts", "worker/patterns/master_improvements.ts", "worker/patterns/remediation_gap_patterns.ts", "worker/patterns/silent_failures.ts", "worker/patterns/checkpoint_errors.ts", "worker/patterns/ddp_errors.ts", "worker/patterns/enterprise_core.ts", "worker/patterns/fallen_off_bus.ts", "worker/patterns/flash_attention.ts", "worker/patterns/fs_errors.ts", "worker/patterns/gpu_reset.ts", "worker/patterns/hung_rank.ts", "worker/patterns/hw_slowdown.ts", "worker/patterns/ib_errors.ts", "worker/patterns/index.ts", "worker/patterns/inference_serving.ts", "worker/patterns/init_errors.ts", "worker/patterns/jax_flax.ts", "worker/patterns/kernel_deadlock.ts", "worker/patterns/kmsg_gpu_errors.ts", "worker/patterns/nccl_hang_detector.ts", "worker/patterns/nccl_slow_node.ts", "worker/patterns/nvlink_check.ts", "worker/patterns/pcie_degradation.ts", "worker/patterns/prefilter.ts", "worker/patterns/straggler_detector.ts", "worker/patterns/sxid_detector.ts", "worker/patterns/thermal_check.ts", "worker/patterns/xid_core.ts", "worker/clock-skew.ts", "worker/serving-slo.ts", "worker/workload-class.ts", "worker/encyclopedia-name-evidence.ts", "worker/encyclopedia-negated-claim.ts", "worker/torchelastic-exitcode.ts", "worker/input-starvation.ts", "worker/cuda-driver-compat.ts", "worker/loss-scale-thrash.ts", "worker/transient-recovered.ts", "worker/throughput-paste-detector.ts", "worker/throughput-log-parse.ts", "worker/throughput-regression.ts", "worker/contracts.ts", "worker/finalize-diagnosis.ts", "worker/checkpoint-analyzer.ts", "worker/goodput.ts", "worker/straggler-detector.ts", "worker/inference-analysis.ts", "worker/fleet-risk.ts", "worker/remediation-scripts.ts", "worker/diagnostic-evidence-requests.ts", "worker/operation-evidence-contracts.ts", "worker/ownership.ts", "worker/mre-generator.ts", "worker/rma-ticket.ts", "worker/log-negation.ts", "worker/diagnostic-input-quality.ts", "worker/diagnostic-request-admission.ts", "worker/exoneration.ts", "worker/evidence-hierarchy.ts", "worker/failure-priors.ts", "worker/confidence-calibration.ts", "worker/encyclopedia-match.ts", "worker/bayesian-network.ts", "src/data/failures-matcher.ts", "src/lib/log-extract.ts", "src/lib/what-to-do.ts"]);
 var Vn = "b4b0b07f7f2277a828466d6a337a43559b9ca5984fb6fa3c3f9aba892a622b4a";
 var Ii = { sha: "0000000000000000000000000000000000000000", shortSha: "offline", branch: "offline-distribution", subject: "Canonical zero-egress diagnosis engine", committedAt: "1970-01-01T00:00:00.000Z", engineVersion: rr };
@@ -13513,7 +13513,7 @@ function registerPrompts(server) {
 
 // src/version.ts
 var SERVER_NAME = "denpex";
-var SERVER_VERSION = "0.2.32";
+var SERVER_VERSION = "0.2.33";
 
 // src/server.ts
 function createServer() {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "denpex-mcp",
-  "version": "0.2.32",
+  "version": "0.2.33",
   "description": "Denpex MCP server, diagnose GPU/ML training failures (CUDA OOM, NCCL timeouts, Xid faults, NaN loss) from Claude Desktop, Cursor, or Claude Code, with optional explicit zero-egress mode.",
   "type": "module",
   "license": "MIT",