omnius 1.0.340 → 1.0.341

package/dist/index.js

@@ -560736,7 +560736,7 @@ var init_critic = __esm({
 function extractSubject(errorText) {
   if (!errorText)
     return null;
-  const PATTERNS = [
+  const PATTERNS2 = [
     // Quoted module / type / symbol after recognizable phrases
     /cannot find (?:module|name|type|symbol|reference|file|namespace)\s+['"`]([^'"`\n]{1,80})['"`]/i,
     /(?:undefined|unresolved)\s+(?:reference|import|symbol)\s+(?:to\s+)?['"`]([^'"`\n]{1,80})['"`]/i,
@@ -560751,7 +560751,7 @@ function extractSubject(errorText) {
     /\bcannot resolve\s+['"`]?([^'"`\n\s]{1,120})['"`]?/i,
     /\bmodule not found:?\s+['"`]?([^'"`\n\s]{1,120})['"`]?/i
   ];
-  for (const re of PATTERNS) {
+  for (const re of PATTERNS2) {
     const m2 = errorText.match(re);
     if (m2 && m2[1]) {
       const subj = m2[1].trim();
@@ -565070,6 +565070,79 @@ var init_consolidation_runtime = __esm({
   }
 });
 
+// packages/orchestrator/dist/completion-evidence-gate.js
+function classifyCompletionClaim(text2) {
+  const t2 = String(text2 || "");
+  for (const p2 of PATTERNS) {
+    if (p2.re.test(t2)) {
+      return { category: p2.category, requiresIndependentEvidence: true, requiredCheck: p2.check };
+    }
+  }
+  return { category: null, requiresIndependentEvidence: false, requiredCheck: "" };
+}
+function detectExitCodeMisread(text2) {
+  const t2 = String(text2 || "");
+  const nonzeroExit = /\bexit(?:ed|s)?\b[^.\n]{0,20}?\b(?:code\s*)?(?:[1-9]\d*|non-?zero)\b/i;
+  const successWord = /\b(success(?:ful|fully|ed)?|succeeded|done|passed|works?|working|fine|ok|complete[d]?)\b/i;
+  if (nonzeroExit.test(t2) && successWord.test(t2))
+    return true;
+  if (/\|\|\s*true\b|\bset \+e\b|2>\/dev\/null\s*;\s*(echo|exit 0)/i.test(t2))
+    return true;
+  return false;
+}
+function auditCompletionClaims(claims) {
+  const blockers = [];
+  for (const c8 of claims || []) {
+    const cls = classifyCompletionClaim(c8.text);
+    if (!cls.requiresIndependentEvidence || !cls.category)
+      continue;
+    if (c8.status !== "supported") {
+      blockers.push({ claim: String(c8.text || "").slice(0, 160), category: cls.category, requiredCheck: cls.requiredCheck });
+    }
+  }
+  return { ok: blockers.length === 0, blockers };
+}
+function completionEvidenceDirective() {
+  return [
+    "[Evidence-gated completion — do not pretend success]",
+    "Before you claim a task is done, optimize for 'is the end state actually correct?', NOT 'did I make the change?'. Each of these claim types REQUIRES an independent check from a DIFFERENT command than the one that made the change:",
+    "- A process is 'running/started/up' → re-probe liveness after a short delay (ps/pgrep, a health endpoint, or the listening port). A PID file or a launcher log is NOT proof it stayed alive — 'started' ≠ 'running'.",
+    "- Something is 'installed/available' → verify with `command -v`/`--version`/a package query, NOT the installer's own log.",
+    "- A bug is 'fixed' → re-run the ORIGINAL failing command/test and watch it pass, not just confirm the edit landed.",
+    "- A success in simulation/mock/dry-run mode → label it as SIMULATED; never present it as the real capability.",
+    "Exit codes: a non-zero exit is NOT success; a zero exit from a wrapper / `|| true` / `set +e` is NOT proof the underlying thing worked — inspect what actually happened.",
+    "If you cannot produce the independent evidence, say so plainly and mark the item unverified or partial — never report a success state you have not verified."
+  ].join("\n");
+}
+var PATTERNS;
+var init_completion_evidence_gate = __esm({
+  "packages/orchestrator/dist/completion-evidence-gate.js"() {
+    "use strict";
+    PATTERNS = [
+      {
+        category: "process_liveness",
+        re: /\b(running|started|launched|spun up|brought up|is up|listening|serving|alive|daemon (is|now)|service (is|now)|server (is|now)|booted|online)\b/i,
+        check: "Independently probe liveness AFTER a short delay: ps/pgrep for the actual process, a health endpoint, or `ss -ltnp` for the port — NOT the launcher log or a PID file (a PID file can exist while the process is dead)."
+      },
+      {
+        category: "installation",
+        re: /\b(installed|available in PATH|on the PATH|set up|provisioned|configured and ready|now available|dependency (is )?(present|installed))\b/i,
+        check: "Independently confirm presence: `command -v <bin>` / `<bin> --version` / package manager query (e.g. `apt-cache policy`, `pip show`) — NOT the installer's own success log."
+      },
+      {
+        category: "fix_confirmation",
+        re: /\b(fixed|resolved|corrected|patched|repaired|no longer (fails|errors|crashes)|works now|issue (is )?gone)\b/i,
+        check: "Re-run the ORIGINAL failing command/test and observe it now passes — an independent reproduction, not just confirming the edit was written."
+      },
+      {
+        category: "reality",
+        re: /\b(simulat\w+|mock\w*|stub\w*|fake|dry[- ]?run|--sim|placeholder mode|fallback mode)\b/i,
+        check: "This was a SIMULATED/mock run. Label it explicitly as simulated and do NOT present it as the real capability; verify against the real system before claiming the real success."
+      }
+    ];
+  }
+});
+
 // packages/orchestrator/dist/tool-batching.js
 function isConcurrencySafe(toolName, readOnlyHints) {
   if (CONCURRENT_SAFE_TOOLS.has(toolName))
@@ -568620,6 +568693,7 @@ var init_agenticRunner = __esm({
     init_dist5();
     init_exploration_fanout();
     init_consolidation_runtime();
+    init_completion_evidence_gate();
     init_tool_batching();
     init_hooks2();
     init_todo_context_chunker();
@@ -570710,6 +570784,18 @@ ${input.answerText ?? ""}`.toLowerCase().trim();
             this._completionLedger = reconcileClaimsWithEvidence(this._completionLedger);
             this._saveCompletionLedgerSafe();
           }
+          try {
+            const _audit = auditCompletionClaims(this._completionLedger.proposedClaims.map((c8) => ({ text: c8.text, status: c8.status })));
+            if (!_audit.ok) {
+              const _checks = _audit.blockers.map((b) => `• [${b.category}] "${b.claim}" → ${b.requiredCheck}`).join("\n");
+              this._completionCaveat = [
+                this._completionCaveat || "",
+                `[UNVERIFIED SUCCESS — run these independent checks before claiming done]
+${_checks}`
+              ].filter(Boolean).join("\n\n");
+            }
+          } catch {
+          }
         }
         const optOverride = this.options.backwardPassReview;
         const raw = (process.env["OMNIUS_BACKWARD_PASS"] || "on").toLowerCase();
@@ -573943,6 +574029,12 @@ Respond with your assessment, then take action.`;
         }
         const contextComposition = await this.assembleContext(task, context2);
         let systemPrompt = contextComposition.assembled;
+        try {
+          systemPrompt = `${systemPrompt}
+
+${completionEvidenceDirective()}`;
+        } catch {
+        }
         try {
           const _cap = getMediaCapability();
           const _tier = this.options.modelTier ?? "large";
@@ -590380,6 +590472,7 @@ __export(dist_exports3, {
   appendSteeringLedgerEntry: () => appendSteeringLedgerEntry,
   appendSteeringOutcome: () => appendSteeringOutcome,
   arcSummary: () => arcSummary,
+  auditCompletionClaims: () => auditCompletionClaims,
   buildAgentNotification: () => buildAgentNotification,
   buildAgentTypeSummary: () => buildAgentTypeSummary,
   buildCompletionScenarioDecomposition: () => buildCompletionScenarioDecomposition,
@@ -590400,6 +590493,7 @@ __export(dist_exports3, {
   chooseCheapModelRoute: () => chooseCheapModelRoute,
   claimAssertion: () => claimAssertion,
   classifyBreadth: () => classifyBreadth,
+  classifyCompletionClaim: () => classifyCompletionClaim,
   classifyHandoff: () => classifyHandoff,
   classifyOllamaProcesses: () => classifyOllamaProcesses,
   cleanForStorage: () => cleanForStorage,
@@ -590408,6 +590502,7 @@ __export(dist_exports3, {
   clearTurnState: () => clearTurnState,
   combineValidatorResults: () => combineValidatorResults,
   compilePersonalityPrompt: () => compilePersonalityPrompt,
+  completionEvidenceDirective: () => completionEvidenceDirective,
   compressFindings: () => compressFindings,
   computeArc: () => computeArc,
   computeStabilityHash: () => computeStabilityHash,
@@ -590425,6 +590520,7 @@ __export(dist_exports3, {
   decomposeSpec: () => decomposeSpec,
   deleteAgentTaskSidecar: () => deleteAgentTaskSidecar,
   deriveClaimsFromProposedText: () => deriveClaimsFromProposedText,
+  detectExitCodeMisread: () => detectExitCodeMisread,
   detectExplorationIntent: () => detectExplorationIntent,
   detectPressure: () => detectPressure,
   detectTaskType: () => detectTaskType,
@@ -590638,6 +590734,7 @@ var init_dist8 = __esm({
     init_memory_consolidation();
     init_exploration_fanout();
     init_consolidation_runtime();
+    init_completion_evidence_gate();
   }
 });
 

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.340",
+  "version": "1.0.341",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.340",
+      "version": "1.0.341",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.340",
+  "version": "1.0.341",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",