omnius 1.0.461 → 1.0.462

package/dist/index.js

@@ -563781,6 +563781,15 @@ function isSuccessfulVerificationEvidence(entry) {
 function isFailedVerificationEvidence(entry) {
   return entry.role === "verification" && entry.success === false;
 }
+function isSuccessfulPostMutationObservation(entry) {
+  if (entry.success !== true)
+    return false;
+  if (entry.role === "blocker")
+    return false;
+  if (isMutationEvidence(entry))
+    return false;
+  return entry.kind === "tool_result" || entry.kind === "runtime_observation" || entry.kind === "delivery_result";
+}
 function verificationFamily(entry) {
   if (entry.role !== "verification")
     return null;
@@ -563807,6 +563816,7 @@ function finalizeCompletionLedgerTruth(ledger) {
   let unresolved = [...ledger.unresolved];
   let lastVerificationInvalidatingMutation = -1;
   let lastVerification = -1;
+  let lastCriticApprovedVerification = -1;
   const lastSuccessfulVerificationByFamily = /* @__PURE__ */ new Map();
   ledger.evidence.forEach((entry, index) => {
     if (isVerificationInvalidatingMutation(entry))
@@ -563818,6 +563828,21 @@ function finalizeCompletionLedgerTruth(ledger) {
         lastSuccessfulVerificationByFamily.set(family, index);
     }
   });
+  for (const critique2 of ledger.critiques) {
+    if (critique2.verdict !== "approve")
+      continue;
+    if (lastVerificationInvalidatingMutation >= 0 && critique2.evidenceSequence <= lastVerificationInvalidatingMutation) {
+      continue;
+    }
+    const reviewedEvidence = ledger.evidence.slice(Math.max(0, lastVerificationInvalidatingMutation + 1), Math.max(0, critique2.evidenceSequence));
+    if (reviewedEvidence.some(isSuccessfulPostMutationObservation)) {
+      lastCriticApprovedVerification = Math.max(lastCriticApprovedVerification, critique2.evidenceSequence - 1);
+    }
+  }
+  const effectiveLastVerification = Math.max(lastVerification, lastCriticApprovedVerification);
+  if (effectiveLastVerification >= 0 && effectiveLastVerification >= lastVerificationInvalidatingMutation) {
+    unresolved = unresolved.filter((item) => item.source !== "verification_missing" && item.source !== "verification_freshness");
+  }
   ledger.evidence.forEach((entry, index) => {
     if (isFailedVerificationEvidence(entry)) {
       const family = verificationFamily(entry);
@@ -563830,12 +563855,13 @@ function finalizeCompletionLedgerTruth(ledger) {
       unresolved = appendUnresolved(unresolved, `Stale edit failure remains unresolved: ${entry.summary}`, entry.id);
     }
   });
-  if (lastVerificationInvalidatingMutation >= 0 && lastVerification < 0) {
+  if (lastVerificationInvalidatingMutation >= 0 && effectiveLastVerification < 0) {
     unresolved = appendUnresolved(unresolved, "File changes occurred without any successful verification evidence.", "verification_missing");
-  } else if (lastVerification >= 0 && lastVerificationInvalidatingMutation > lastVerification) {
+  } else if (effectiveLastVerification >= 0 && lastVerificationInvalidatingMutation > effectiveLastVerification) {
     unresolved = appendUnresolved(unresolved, "File changes occurred after the last successful verification; final verification is stale.", "verification_freshness");
   }
-  const status = ledger.status === "blocked" || ledger.status === "request_changes" ? ledger.status : unresolved.length > 0 ? "incomplete_verification" : ledger.status;
+  const latestCritique = ledger.critiques.at(-1);
+  const status = ledger.status === "blocked" || ledger.status === "request_changes" ? ledger.status : unresolved.length > 0 ? "incomplete_verification" : latestCritique?.verdict === "approve" ? "approved" : ledger.status === "incomplete_verification" ? "open" : ledger.status;
   return {
     ...ledger,
     unresolved,
@@ -579170,6 +579196,15 @@ var init_agenticRunner = __esm({
       _workboardDir() {
         return this.options.workboardDir || this._workingDirectory || process.cwd();
       }
+      _todoWriteAvailable() {
+        return this.tools.has("todo_write");
+      }
+      _availablePreferredTools(candidates) {
+        const available = candidates.filter((name10) => this.tools.has(name10));
+        if (available.length > 0)
+          return available;
+        return this.tools.has("task_complete") ? ["task_complete"] : [];
+      }
       getOrCreateWorkboard() {
         const goal = this._taskState.originalGoal || this._taskState.goal || "";
         if (this._workboard) {
@@ -579425,7 +579460,7 @@ ${parts.join("\n")}
             cardId: integration.id,
             title: integration.title,
             reason: "implementation evidence exists; exercise the changed runtime path before more discovery",
-            preferredTools: ["shell"]
+            preferredTools: this._availablePreferredTools(["shell"])
           };
         }
         if (hasVerificationEvidence && verification) {
@@ -579433,7 +579468,10 @@ ${parts.join("\n")}
             cardId: verification.id,
             title: verification.title,
             reason: "verification evidence exists; reconcile outcome, todos, and blockers",
-            preferredTools: ["todo_write", "task_complete"]
+            preferredTools: this._availablePreferredTools([
+              "todo_write",
+              "task_complete"
+            ])
           };
         }
         if (discoveryEvidence > 0 && !hasImplementationEvidence && implementation2) {
@@ -579441,7 +579479,12 @@ ${parts.join("\n")}
             cardId: implementation2.id,
             title: implementation2.title,
             reason: "discovery evidence is already present; land the smallest concrete repair",
-            preferredTools: ["file_patch", "batch_edit", "file_edit", "file_write"]
+            preferredTools: this._availablePreferredTools([
+              "file_patch",
+              "batch_edit",
+              "file_edit",
+              "file_write"
+            ])
           };
         }
         const active = snapshot.cards.find((card) => card.status === "in_progress") ?? snapshot.cards.find((card) => card.status === "needs_changes") ?? snapshot.cards.find((card) => card.status === "open") ?? discovery;
@@ -579451,13 +579494,24 @@ ${parts.join("\n")}
           cardId: active.id,
           title: active.title,
           reason: active.id === "discover-current-state" ? "no authoritative target evidence is recorded yet" : "workboard card is the next unresolved card",
-          preferredTools: active.id === "discover-current-state" ? ["file_read", "grep_search", "list_directory", "shell"] : ["file_patch", "batch_edit", "file_edit", "file_write", "shell"]
+          preferredTools: active.id === "discover-current-state" ? this._availablePreferredTools([
+            "file_read",
+            "grep_search",
+            "list_directory",
+            "shell"
+          ]) : this._availablePreferredTools([
+            "file_patch",
+            "batch_edit",
+            "file_edit",
+            "file_write",
+            "shell"
+          ])
         };
       }
       _focusToolsForRequiredAction(action) {
         switch (action) {
           case "update_todos":
-            return "todo_write with a changed plan/status, or a real file edit if that is the active work";
+            return this._todoWriteAvailable() ? "todo_write with a changed plan/status, or a real file edit if that is the active work" : "todo_write is unavailable in this tool surface; continue with the intended substantive action, or task_complete with evidence if done";
           case "read_authoritative_target":
             return "use cached evidence or one authoritative file_read/read-only shell, then act";
           case "use_cached_evidence":
@@ -579511,7 +579565,9 @@ ${parts.join("\n")}
           const integration = byId.get("integrate-and-run");
           const verification = byId.get("verify-observed-outcome");
           lines.push(`workboard_current_card=${action.cardId} (${action.title})`);
-          lines.push(`workboard_preferred_tools=${action.preferredTools.join(", ")}`);
+          if (action.preferredTools.length > 0) {
+            lines.push(`workboard_preferred_tools=${action.preferredTools.join(", ")}`);
+          }
           lines.push(`workboard_reason=${action.reason}`);
           lines.push(`workboard_evidence_counts=discover:${this._workboardEvidenceCount(discovery)} implement:${this._workboardEvidenceCount(implementation2)} integrate:${this._workboardEvidenceCount(integration)} verify:${this._workboardEvidenceCount(verification)}`);
           if (this._workboardHasEditEvidence(implementation2)) {
@@ -579537,7 +579593,7 @@ ${parts.join("\n")}
         }
         if (lines.length === 2)
           return null;
-        lines.push("not_progress=repeat unchanged todo_write; repeat broad discovery already represented in this frame; keep discovery active after a real mutation");
+        lines.push(this._todoWriteAvailable() ? "not_progress=repeat unchanged todo_write; repeat broad discovery already represented in this frame; keep discovery active after a real mutation" : "not_progress=repeat broad discovery already represented in this frame; keep discovery active after a real mutation");
         return lines.join("\n");
       }
       _workboardTargetCardId(snapshot, toolName, result, meta = {}) {
@@ -580266,10 +580322,23 @@ ${parts.join("\n")}
         const pressureCue = pressureCheck(task);
         const rawPrompt = getSystemPromptForTier(this.options.modelTier);
         const basePrompt = rawPrompt + pressureCue;
+        const todoBatchGuidance = this._todoWriteAvailable() ? " Use todo_write between batches to mark progress." : "";
         const _BATCH_GUIDANCE = {
-          small: "\n\n## Response batching\n\nEmit AT MOST 2 tool calls per response. After observing their results, plan the next 2 in your following response. Smaller batches let the orchestrator deliver cache/failure/progress signals to you between actions. Tool calls beyond the cap are dropped. Use todo_write between batches to mark progress.",
-          medium: "\n\n## Response batching\n\nEmit AT MOST 4 tool calls per response. After observing their results, plan the next batch in your following response. Smaller batches let the orchestrator deliver cache/failure/progress signals to you between actions. Tool calls beyond the cap are dropped. Use todo_write between batches to mark progress.",
-          large: "\n\n## Response batching\n\nEmit AT MOST 6 tool calls per response. Smaller batches receive better feedback (cache/failure/progress signals between actions). Tool calls beyond the cap are dropped. Use todo_write between batches."
+          small: `
+
+## Response batching
+
+Emit AT MOST 2 tool calls per response. After observing their results, plan the next 2 in your following response. Smaller batches let the orchestrator deliver cache/failure/progress signals to you between actions. Tool calls beyond the cap are dropped.${todoBatchGuidance}`,
+          medium: `
+
+## Response batching
+
+Emit AT MOST 4 tool calls per response. After observing their results, plan the next batch in your following response. Smaller batches let the orchestrator deliver cache/failure/progress signals to you between actions. Tool calls beyond the cap are dropped.${todoBatchGuidance}`,
+          large: `
+
+## Response batching
+
+Emit AT MOST 6 tool calls per response. Smaller batches receive better feedback (cache/failure/progress signals between actions). Tool calls beyond the cap are dropped.${todoBatchGuidance}`
         };
         const batchGuidance = _BATCH_GUIDANCE[this.options.modelTier ?? "large"] ?? _BATCH_GUIDANCE.large;
         const shellGuidance = "\n\n## Shell command patterns\n\nWhen investigating a build/test/install failure, RUN THE COMMAND WITHOUT TRUNCATION FIRST to see the full error.\nAvoid `| tail -N` / `| head -N` / `| sed -n '1,Np'` on commands you don't yet trust — they drop the part of the output where errors usually appear.\nPipefail is on by default in this orchestrator: a pipeline's exit code reflects the FIRST failing stage, not just the last. So `<cmd> | tail -80` returns non-zero if `<cmd>` failed.\nWhen you DO want to limit context (after diagnosis), prefer `<cmd> 2>&1 | tail -300` (or larger) so you keep enough output to see the actual error block.\nIf your command exited 0 but produced suspiciously short output (~< 800 chars), the orchestrator will inject a [HINT — pipe-to-truncator detected] block telling you to re-run without truncation.";
@@ -581506,6 +581575,8 @@ ${result.output ?? ""}`;
       getOpenTodoItems() {
         if (this.options.disableTodoCompletionGuard)
           return [];
+        if (!this._todoWriteAvailable())
+          return [];
         const todos = this.readSessionTodos();
         if (!todos || todos.length === 0)
           return [];
@@ -582496,6 +582567,8 @@ ${_checks}`
        * Mutates `_lastTodoReminderTurn` when a reminder is produced.
        */
       getTodoReminderContent(currentTurn) {
+        if (!this._todoWriteAvailable())
+          return null;
         const todos = this.readSessionTodos();
         const result = computeTodoReminder({
           currentTurn,
@@ -582856,6 +582929,8 @@ ${contentPreview}
        * Recommends marking progress before further work. Returns null otherwise.
        */
       _renderProgressNudgeBlock(turn) {
+        if (!this._todoWriteAvailable())
+          return null;
         const n2 = this._writesSinceLastTodoWrite;
         if (n2 < 4)
           return null;
@@ -583469,11 +583544,14 @@ ${contentPreview}
           ].join("\n");
         }
         if (latest.tool === "todo_write") {
+          const raw2 = `${latest.error || ""}
+${latest.output || ""}`;
+          const unavailable = /Unknown tool/i.test(raw2);
           return [
             `[PLANNING TOOL RECOVERY]`,
             `Last failing tool: todo_write (turn ${latest.turn})`,
-            `A checklist write failure is bookkeeping/argument transport, not evidence that the target codebase is broken.`,
-            `Repair the todo_write payload if you still need the visible plan, or continue with the intended first substantive task using the plan content you already attempted to record.`,
+            unavailable ? `todo_write is not registered in this tool surface. Do not retry it.` : `A checklist write failure is bookkeeping/argument transport, not evidence that the target codebase is broken.`,
+            unavailable ? `Use the attempted checklist only as private state; continue with available substantive tools or task_complete with evidence.` : `Repair the todo_write payload if you still need the visible plan, or continue with the intended first substantive task using the plan content you already attempted to record.`,
             `Do not inspect or edit project files solely because todo_write rejected its arguments.`
           ].join("\n");
         }
@@ -583571,7 +583649,7 @@ ${latest.output || ""}`.trim();
             lines.push(`... +${ordered.length - visible.length} more todo node(s)`);
           }
           lines.push("decomposition_contract=Use stable ids plus parentId for subtasks. If the active item has multiple remaining actions, rewrite it as a parent with concrete child leaf todos; work leaf-first and do not complete a parent before all children are complete with evidence.");
-          lines.push(`(turn ${turn} — call todo_write ONLY to update state, never to re-read this plan)`);
+          lines.push(this._todoWriteAvailable() ? `(turn ${turn} — call todo_write ONLY to update state, never to re-read this plan)` : `(turn ${turn} — todo_write is unavailable in this tool surface; use this plan as read-only state)`);
           return lines.join("\n");
         } catch {
           return null;
@@ -583825,12 +583903,13 @@ ${sections.join("\n")}` : sections.join("\n");
        */
       _buildRepeatGateBlock(toolName, args, hits) {
         const argPreview = JSON.stringify(args ?? {}).slice(0, 200);
+        const progressStep = this._todoWriteAvailable() ? `  1. Call todo_write — decompose the remaining work into concrete, verifiable steps, mark the current step in_progress, and mark anything already done as completed.` : `  1. Re-anchor on the active evidence and name the concrete next action; do not call unavailable planning tools.`;
         return [
           `[REPEAT GATE] Exact repeat #${hits} of ${toolName}(${argPreview}).`,
           `You ALREADY have this result — re-reading identical arguments cannot produce new information, so this call was NOT executed.`,
           ``,
           `You are looping because progress tracking has drifted. Before any further redundant calls you MUST:`,
-          `  1. Call todo_write — decompose the remaining work into concrete, verifiable steps, mark the current step in_progress, and mark anything already done as completed.`,
+          progressStep,
           `  2. Then take a CONCRETE action on the evidence you already have: edit/write the implicated file, run a verification command, or call task_complete with evidence.`,
           ``,
           `Do NOT repeat this exact call. If you believe state changed, perform the concrete change first, then retry with distinct evidence.`
@@ -583847,7 +583926,7 @@ ${sections.join("\n")}` : sections.join("\n");
           "This call was not executed again, and the cached payload was not re-emitted because repeating it bloats the model context.",
           summary,
           "",
-          "Required next action: use the active evidence frame/tool cache and take a concrete next step: edit/write, run verification, update todos with changed state, or task_complete with evidence.",
+          this._todoWriteAvailable() ? "Required next action: use the active evidence frame/tool cache and take a concrete next step: edit/write, run verification, update todos with changed state, or task_complete with evidence." : "Required next action: use the active evidence frame/tool cache and take a concrete next step: edit/write, run verification, or task_complete with evidence.",
           "If you need different information, use a distinct offset/limit/query. For broad multi-file discovery, use file_explore, grep_search, fanout_explore, or full_sub_agent instead of repeating this call."
         ].join("\n");
       }
@@ -587683,7 +587762,7 @@ ${_staleSamples.join("\n")}` : ``,
             }
           }
           const turnTier = this.options.modelTier ?? "large";
-          if (turn === 0 && !this.options.disableTodoPlanningNudges && (turnTier === "small" || turnTier === "medium")) {
+          if (turn === 0 && !this.options.disableTodoPlanningNudges && this._todoWriteAvailable() && (turnTier === "small" || turnTier === "medium")) {
             const goal = this._taskState.goal || "";
             const substantiveGoal = goal.replace(/\b(?:then\s+)?call\s+task_complete\b[^.?!;]*/gi, "").replace(/\b(?:observe|report|summarize|finish|complete)\b[^.?!;]*/gi, "");
             const wordCount2 = substantiveGoal.split(/\s+/).filter(Boolean).length;
@@ -587796,7 +587875,7 @@ ${top.map((t2) => `- ${t2.name}: ${t2.desc}`).join("\n")}`);
             const isReadTask = /\bread\b|\bshow\b|\btell me\b|\bwhat is\b/i.test(taskGoal);
             const hints = [];
             if (isSimpleTask) {
-              hints.push("This is a simple task — if it needs only ONE substantive tool call, skip todo_write and call the tool directly, then task_complete. Do not count reporting, observing output, or task_complete as planning steps. If it needs 2+ substantive work steps, use todo_write to plan.");
+              hints.push(this._todoWriteAvailable() ? "This is a simple task — if it needs only ONE substantive tool call, skip todo_write and call the tool directly, then task_complete. Do not count reporting, observing output, or task_complete as planning steps. If it needs 2+ substantive work steps, use todo_write to plan." : "This is a simple task — if it needs only ONE substantive tool call, call the tool directly, then task_complete. Do not count reporting, observing output, or task_complete as planning steps.");
             }
             if (isSearchTask) {
               hints.push("SEARCH STRATEGY: Use grep_search to find what you need FIRST, THEN file_read only the specific file and lines. Do NOT read entire files hoping to find something.");
@@ -588532,7 +588611,7 @@ ${memoryLines.join("\n")}`
                   `Best practice for this model tier: emit at most ${_responseCap} tool calls per turn.`,
                   `Observe the results of the first ${_responseCap}, then plan the next batch in the NEXT turn.`,
                   `Smaller batches let cache/dedup/progress-gate/failure signals reach you BEFORE you commit to more work.`,
-                  `Use todo_write between batches to track which items are done.`
+                  this._todoWriteAvailable() ? `Use todo_write between batches to track which items are done.` : `Track progress internally between batches; do not call unavailable planning tools.`
                 ].join("\n")
               });
             }
@@ -588873,7 +588952,7 @@ Corrective action: try a different approach first: read relevant files, adjust a
                     turn,
                     timestamp: (/* @__PURE__ */ new Date()).toISOString()
                   });
-                  const budgetMsg = `[BUDGET EXHAUSTED] You have used all ${toolBudgets[tc.name]} allowed ${tc.name} calls for the current phase. You ALREADY have enough information from previous calls. DO NOT try to call ${tc.name} again — it will be blocked. If your todo list shows more phases pending: mark the current phase completed via todo_write so a new budget allowance kicks in. If all phases are done: call task_complete with your final summary.`;
+                  const budgetMsg = `[BUDGET EXHAUSTED] You have used all ${toolBudgets[tc.name]} allowed ${tc.name} calls for the current phase. You ALREADY have enough information from previous calls. DO NOT try to call ${tc.name} again — it will be blocked. ` + (this._todoWriteAvailable() ? `If your todo list shows more phases pending: mark the current phase completed via todo_write so a new budget allowance kicks in. ` : `If more phases remain, continue with a different available tool instead of repeating the exhausted one. `) + `If all phases are done: call task_complete with your final summary.`;
                   this.emit({
                     type: "tool_result",
                     toolName: tc.name,
@@ -590559,14 +590638,15 @@ Respond with EXACTLY this structure before your next tool call:
                 }
                 if (!_prior && !this._opaqueErrorHintInjected.has(_refStem) && _entry.subject == null && (categorizeError(_refErr) === "unknown" || _refErr.length > 100)) {
                   this._opaqueErrorHintInjected.add(_refStem);
+                  const todoWriteUnavailable = tc.name === "todo_write" && /Unknown tool/i.test(_entry.wentWrong);
                   const opaqueMessage = tc.name === "todo_write" ? [
                     `[PLANNING TOOL RECOVERY — todo_write failed before updating the visible checklist.]`,
                     ``,
                     `Tool: ${tc.name}`,
                     `Error: "${_entry.wentWrong}"`,
                     ``,
-                    `This is a recoverable planning-tool argument issue, not a local code or dependency failure.`,
-                    `Correct the todo_write payload if a visible checklist is still needed; otherwise continue with the first substantive task from the attempted plan.`,
+                    todoWriteUnavailable ? `todo_write is not registered in this tool surface. Do not retry it; continue using the available tools and call task_complete with evidence when done.` : `This is a recoverable planning-tool argument issue, not a local code or dependency failure.`,
+                    todoWriteUnavailable ? `Use the attempted checklist only as private state; do not let missing bookkeeping block substantive work.` : `Correct the todo_write payload if a visible checklist is still needed; otherwise continue with the first substantive task from the attempted plan.`,
                     `Do not pivot to broad shell discovery or code edits solely because todo_write failed.`,
                     ``,
                     this._renderLocalFailureNudge(turn)

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.461",
+  "version": "1.0.462",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.461",
+      "version": "1.0.462",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.461",
+  "version": "1.0.462",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",