denpex-mcp 0.2.23 → 0.2.24

package/dist/index.js

@@ -12741,9 +12741,9 @@ function registerFleetTools(server, ctx) {
   server.registerTool(
     "validate_checkpoint",
     {
-      title: "Find the newest checkpoint that is safe to resume from",
+      title: "Inspect supplied checkpoint integrity evidence",
       annotations: { readOnlyHint: true, idempotentHint: true, openWorldHint: true },
-      description: "Before you resume a crashed job, check that the checkpoint you are about to resume FROM is not the one the crash corrupted. Give it `ls -la` of the checkpoint directory, and optionally `sha256sum -c` output and any load errors. Returns which checkpoints are valid, which are corrupt, and the newest safe resume point. Resuming from a mid-write checkpoint costs days and is the single most expensive avoidable mistake after a crash.",
+      description: "Inspect directory listings, checksum output and load errors for signs of checkpoint corruption. Supplied metadata does not verify full shard coverage, compatible optimizer state or successful workload recovery. Validate actual loading and intended progress before resuming.",
       inputSchema: {
         directoryListing: z9.string().optional().describe("Output of `ls -la` (or `ls -la --time-style=full-iso`) on the checkpoint directory."),
         sha256Output: z9.string().optional().describe("Output of `sha256sum -c MANIFEST.sha256`, if you keep a manifest."),
@@ -12751,6 +12751,9 @@ function registerFleetTools(server, ctx) {
         failureType: z9.string().optional().describe("The diagnosed failure type, if you have one, it sharpens the verdict.")
       },
       outputSchema: {
+        candidateArtifact: z9.string().nullable(),
+        resumeVerified: z9.literal(false),
+        workloadRecoveryVerified: z9.literal(false),
         latestValidCheckpoint: z9.string().nullable(),
         validCheckpoints: z9.array(z9.string()),
         corruptedCheckpoints: z9.array(z9.string()),
@@ -12776,13 +12779,17 @@ function registerFleetTools(server, ctx) {
         const latest = str2(r.latestValidCheckpoint);
         const recommendation = str2(r.recommendation) ?? "";
         const lines = [
-          latest ? `## Safe resume point: \`${latest}\`` : `## No verified-safe checkpoint found`,
-          ...bullets("Corrupted", corrupt),
-          ...bullets("Valid", valid.slice(0, 10)),
-          ...bullets("Issues", issues)
+          "## Checkpoint evidence assessment",
+          ...bullets("Corruption signals", corrupt),
+          ...bullets("Checksum-passing entries supplied", valid.slice(0, 10)),
+          ...bullets("Issues", issues),
+          "Resume is not verified. Check the complete manifest, actual load, configuration compatibility and workload progress."
         ];
         if (recommendation) lines.push("", `**Recommendation:** ${recommendation}`);
         return toolResult(lines.join("\n"), {
+          candidateArtifact: str2(r.candidateArtifact),
+          resumeVerified: false,
+          workloadRecoveryVerified: false,
           latestValidCheckpoint: latest,
           validCheckpoints: valid,
           corruptedCheckpoints: corrupt,
@@ -12968,7 +12975,7 @@ function registerPrompts(server) {
 
 // src/version.ts
 var SERVER_NAME = "denpex";
-var SERVER_VERSION = "0.2.23";
+var SERVER_VERSION = "0.2.24";
 
 // src/server.ts
 function createServer() {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "denpex-mcp",
-  "version": "0.2.23",
+  "version": "0.2.24",
   "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",