omnius 1.0.367 → 1.0.368

package/dist/index.js

@@ -558608,6 +558608,19 @@ function isFailedVerificationEvidence(entry) {
   const text2 = `${entry.toolName ?? ""} ${entry.summary}`.toLowerCase();
   return /\b(test|tests|verify|verification|build|compile|tsc|vitest|jest|pytest|go test|cargo test)\b/.test(text2);
 }
+function verificationFamily(entry) {
+  const text2 = `${entry.toolName ?? ""} ${entry.summary}`.toLowerCase();
+  if (/\b(vitest|jest|pytest|go test|cargo test|npm test|pnpm test|yarn test|node\s+tests?|test|tests|spec)\b/.test(text2)) {
+    return "test";
+  }
+  if (/\b(typecheck|tsc)\b/.test(text2))
+    return "typecheck";
+  if (/\b(build|built|compile|compiled)\b/.test(text2))
+    return "build";
+  if (/\b(verify|verified|verification)\b/.test(text2))
+    return "verify";
+  return null;
+}
 function isStaleEditEvidence(entry) {
   if (entry.success !== false)
     return false;
@@ -558618,13 +558631,24 @@ function finalizeCompletionLedgerTruth(ledger) {
   let unresolved = [...ledger.unresolved];
   let lastMutation = -1;
   let lastVerification = -1;
+  const lastSuccessfulVerificationByFamily = /* @__PURE__ */ new Map();
   ledger.evidence.forEach((entry, index) => {
     if (isMutationEvidence(entry))
       lastMutation = index;
-    if (isSuccessfulVerificationEvidence(entry))
+    if (isSuccessfulVerificationEvidence(entry)) {
       lastVerification = index;
+      const family = verificationFamily(entry);
+      if (family)
+        lastSuccessfulVerificationByFamily.set(family, index);
+    }
+  });
+  ledger.evidence.forEach((entry, index) => {
     if (isFailedVerificationEvidence(entry)) {
-      unresolved = appendUnresolved(unresolved, `Verification failed or did not prove success: ${entry.summary}`, entry.id);
+      const family = verificationFamily(entry);
+      const resolvedByLaterSuccess = family !== null && (lastSuccessfulVerificationByFamily.get(family) ?? -1) > index;
+      if (!resolvedByLaterSuccess) {
+        unresolved = appendUnresolved(unresolved, `Verification failed or did not prove success: ${entry.summary}`, entry.id);
+      }
     }
     if (isStaleEditEvidence(entry)) {
       unresolved = appendUnresolved(unresolved, `Stale edit failure remains unresolved: ${entry.summary}`, entry.id);
@@ -568816,6 +568840,16 @@ function classifyShellIntent(cmd) {
   } else {
     verbToken = first2;
   }
+  if (isRunner && verbToken === "run" && tokens.length >= 3) {
+    const scriptToken = tokens[2].toLowerCase().replace(/^["']|["']$/g, "");
+    const scriptRow = VERB_TABLE[scriptToken];
+    if (scriptRow) {
+      return { klass: scriptRow.klass, verb: scriptRow.canonical, tool: first2 };
+    }
+    if (/\b(test|spec|vitest|jest|pytest|check|verify|lint|typecheck)\b/i.test(scriptToken)) {
+      return { klass: "verify", verb: "test", tool: first2 };
+    }
+  }
   const row = VERB_TABLE[verbToken];
   if (!row)
     return { klass: "other", verb: verbToken, tool: first2 };
@@ -568831,7 +568865,7 @@ function verifyShellOutcome(command, ctx3, result) {
   if (!result.success) {
     return { trustworthy: true, intentBucket: baseBucket, outcomeClass: "broken" };
   }
-  if (intent.klass === "read") {
+  if (intent.klass === "read" || intent.klass === "verify") {
     return { trustworthy: true, intentBucket: baseBucket, outcomeClass: "verified" };
   }
   const mtimeCheck = checkMutateMtimeDelta(intent, cwd4, ctx3, result);
@@ -569169,13 +569203,17 @@ var init_postActionVerifier = __esm({
       build: { canonical: "build", klass: "mutate" },
       compile: { canonical: "build", klass: "mutate" },
       make: { canonical: "build", klass: "mutate" },
-      // ── test family (mutate — typically writes coverage/snapshots) ────────────
-      test: { canonical: "test", klass: "mutate" },
-      t: { canonical: "test", klass: "mutate" },
-      spec: { canonical: "test", klass: "mutate" },
-      vitest: { canonical: "test", klass: "mutate" },
-      jest: { canonical: "test", klass: "mutate" },
-      pytest: { canonical: "test", klass: "mutate" },
+      // ── verification family (read-only by contract for the verifier) ──────────
+      test: { canonical: "test", klass: "verify" },
+      t: { canonical: "test", klass: "verify" },
+      spec: { canonical: "test", klass: "verify" },
+      vitest: { canonical: "test", klass: "verify" },
+      jest: { canonical: "test", klass: "verify" },
+      pytest: { canonical: "test", klass: "verify" },
+      check: { canonical: "check", klass: "verify" },
+      verify: { canonical: "verify", klass: "verify" },
+      lint: { canonical: "lint", klass: "verify" },
+      typecheck: { canonical: "typecheck", klass: "verify" },
       // ── publish / scaffold / generate (mutate) ────────────────────────────────
       publish: { canonical: "publish", klass: "mutate" },
       init: { canonical: "init", klass: "mutate" },
@@ -571912,6 +571950,37 @@ Pick the SMALLEST concrete deliverable from the spec — typically the project e
           return false;
         return true;
       }
+      _toolEvidencePreview(result, displayOutput, max = 500) {
+        const modelVisible = result.llmContent ?? result.output ?? displayOutput ?? "";
+        const failurePrefix = result.success === false && result.error ? `Error: ${result.error}` : "";
+        const combined = failurePrefix && modelVisible && !String(modelVisible).startsWith(failurePrefix) ? `${failurePrefix}
+${modelVisible}` : modelVisible || result.error || displayOutput || "";
+        return String(combined).slice(0, max);
+      }
+      _recordCompletionToolEvidenceFromResult(input) {
+        if (!this._completionLedger || input.toolName === "task_complete")
+          return;
+        const realFileMutation = input.realFileMutation ?? this._isRealProjectMutation(input.toolName, input.result);
+        const realMutationPaths = input.realMutationPaths ?? (realFileMutation ? this._extractToolTargetPaths(input.toolName, input.args, input.result) : []);
+        this._completionLedger = recordToolEvidence(this._completionLedger, {
+          name: input.toolName,
+          success: input.result.success,
+          outputPreview: (input.outputPreview ?? this._toolEvidencePreview(input.result)).toString().slice(0, 500),
+          argsKey: input.argsKey.slice(0, 300)
+        });
+        if (realFileMutation && realMutationPaths.length > 0) {
+          for (const filePath of realMutationPaths) {
+            this._completionLedger = recordCompletionEvidence(this._completionLedger, {
+              kind: "file_change",
+              toolName: input.toolName,
+              success: true,
+              summary: `file change: ${filePath}`,
+              rawRef: `file://${filePath}`
+            });
+          }
+        }
+        this._saveCompletionLedgerSafe();
+      }
       _isAtomicBatchEditAbort(toolName, result) {
         if (toolName !== "batch_edit" || !result || result.success !== false) {
           return false;
@@ -572605,10 +572674,20 @@ ${_checks}`
         const maxCycles = parseInt(process.env["OMNIUS_BACKWARD_PASS_MAX_CYCLES"] || "2", 10) || 2;
         if (this._backwardPassCyclesUsed >= maxCycles) {
           const lastCritique2 = this._lastBackwardPassCritique;
-          const concern = lastCritique2?.rationale?.trim() || this._lastBackwardPassVerdict || "unspecified";
+          const evidenceAfterCritique = lastCritique2 ? toolCallLog.slice(lastCritique2.toolLogLength) : [];
+          const freshVerification = evidenceAfterCritique.filter((entry) => {
+            if (entry.success !== true)
+              return false;
+            const text2 = `${entry.name} ${entry.argsKey} ${entry.outputPreview ?? ""}`.toLowerCase();
+            return /\b(test|tests|vitest|jest|pytest|go test|cargo test|npm test|pnpm test|yarn test|typecheck|tsc|build|verify|verification)\b/.test(text2) && /\b(pass|passed|success|exit:?\s*0|ok|0 failures|0 failed)\b/.test(text2);
+          }).slice(-3);
+          const concern = freshVerification.length > 0 ? [
+            "Prior reviewer concern may be stale; fresh verification evidence was recorded after that critique.",
+            ...freshVerification.map((entry) => `${entry.name}: ${(entry.outputPreview ?? entry.argsKey).slice(0, 220)}`)
+          ].join(" ") : lastCritique2?.rationale?.trim() || this._lastBackwardPassVerdict || "unspecified";
           this._completionCaveat = [
             `[COMPLETION CAVEAT] Backward-pass review did not fully approve after ${this._backwardPassCyclesUsed}/${maxCycles} cycle(s).`,
-            `Unresolved reviewer concern: ${concern.slice(0, 600)}`,
+            `${freshVerification.length > 0 ? "Reviewer reconciliation note" : "Unresolved reviewer concern"}: ${concern.slice(0, 600)}`,
             "The work was completed with deliverables present; treat the above as follow-up rather than a blocker."
           ].join("\n");
           this.emit({
@@ -580161,7 +580240,7 @@ ${bookkeepingGuidance}` : bookkeepingGuidance;
                 currentLogEntry.success = result.success;
                 currentLogEntry.mutated = realFileMutation;
                 currentLogEntry.mutatedFiles = realMutationPaths;
-                currentLogEntry.outputPreview = (result.success ? result.llmContent ?? result.output ?? output : result.error ?? result.output ?? output).toString().slice(0, 500);
+                currentLogEntry.outputPreview = this._toolEvidencePreview(result, output);
               }
               this._toolEvents.push({
                 name: tc.name,
@@ -580502,26 +580581,15 @@ Then use file_read on individual FILES inside it.`);
                   lastFailureHandoffTurn = turn;
                 }
               }
-              if (this._completionLedger && tc.name !== "task_complete") {
-                this._completionLedger = recordToolEvidence(this._completionLedger, {
-                  name: tc.name,
-                  success: result.success,
-                  outputPreview: (result.success ? result.llmContent ?? result.output ?? output : result.error ?? result.output ?? output).toString().slice(0, 500),
-                  argsKey: tc.arguments ? JSON.stringify(tc.arguments).slice(0, 300) : ""
-                });
-                if (realFileMutation && realMutationPaths.length > 0) {
-                  for (const filePath2 of realMutationPaths) {
-                    this._completionLedger = recordCompletionEvidence(this._completionLedger, {
-                      kind: "file_change",
-                      toolName: tc.name,
-                      success: true,
-                      summary: `file change: ${filePath2}`,
-                      rawRef: `file://${filePath2}`
-                    });
-                  }
-                }
-                this._saveCompletionLedgerSafe();
-              }
+              this._recordCompletionToolEvidenceFromResult({
+                toolName: tc.name,
+                argsKey: tc.arguments ? JSON.stringify(tc.arguments) : "",
+                args: tc.arguments,
+                result,
+                outputPreview: this._toolEvidencePreview(result, output),
+                realFileMutation,
+                realMutationPaths
+              });
               this._onTypedEvent?.({
                 type: "tool_call_finished",
                 runId: this._sessionId ?? "unknown",
@@ -581617,6 +581685,15 @@ Full content available via: repl_exec(code="data = retrieve('${handleId}')")  or
                 } else {
                   output = modelContent2;
                 }
+                const bfRealFileMutation = this._isRealProjectMutation(tc.name, result);
+                const bfRealMutationPaths = bfRealFileMutation ? this._extractToolTargetPaths(tc.name, tc.arguments, result) : [];
+                const bfLogEntry = toolCallLog[toolCallLog.length - 1];
+                if (bfLogEntry) {
+                  bfLogEntry.success = result.success;
+                  bfLogEntry.mutated = bfRealFileMutation;
+                  bfLogEntry.mutatedFiles = bfRealMutationPaths;
+                  bfLogEntry.outputPreview = this._toolEvidencePreview(result, output);
+                }
                 this.emit({
                   type: "tool_result",
                   toolName: tc.name,
@@ -581626,14 +581703,15 @@ Full content available via: repl_exec(code="data = retrieve('${handleId}')")  or
                   timestamp: (/* @__PURE__ */ new Date()).toISOString()
                 });
                 this._trackDecomp2(tc, result, turn);
-                if (this._completionLedger && tc.name !== "task_complete") {
-                  this._completionLedger = recordToolEvidence(this._completionLedger, {
-                    name: tc.name,
-                    success: result.success,
-                    outputPreview: output.slice(0, 500),
-                    argsKey: bfArgsKey.slice(0, 300)
-                  });
-                }
+                this._recordCompletionToolEvidenceFromResult({
+                  toolName: tc.name,
+                  argsKey: bfArgsKey,
+                  args: tc.arguments,
+                  result,
+                  outputPreview: this._toolEvidencePreview(result, output),
+                  realFileMutation: bfRealFileMutation,
+                  realMutationPaths: bfRealMutationPaths
+                });
                 const enoentTools2 = /* @__PURE__ */ new Set([
                   "file_read",
                   "list_directory",

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.367",
+  "version": "1.0.368",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.367",
+      "version": "1.0.368",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.367",
+  "version": "1.0.368",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",