@ouro.bot/cli 0.1.0-alpha.91 → 0.1.0-alpha.93

package/changelog.json

@@ -1,6 +1,19 @@
 {
   "_note": "This changelog is maintained as part of the PR/version-bump workflow. Agent-curated, not auto-generated. Agents read this file directly via read_file to understand what changed between versions.",
   "versions": [
+    {
+      "version": "0.1.0-alpha.93",
+      "changes": [
+        "Final-answer truth checks now require fresh external-state verification (gh pr view, npm view, etc.) before allowing intent=complete when a live obligation is active, preventing stale or guessed merge/publish/deploy claims."
+      ]
+    },
+    {
+      "version": "0.1.0-alpha.92",
+      "changes": [
+        "Final-answer truth checks now reject `intent=complete` when a must-resolve handoff still has an active live-session return obligation and no newer follow-up proves the loop resumed.",
+        "This keeps visible operator loops open until the agent actually brings back the external-state update or reports a concrete blocker instead of going dark early."
+      ]
+    },
     {
       "version": "0.1.0-alpha.91",
       "changes": [

package/dist/heart/core.js

@@ -7,6 +7,7 @@ exports.getModel = getModel;
 exports.getProvider = getProvider;
 exports.createSummarize = createSummarize;
 exports.getProviderDisplayLabel = getProviderDisplayLabel;
+exports.isExternalStateQuery = isExternalStateQuery;
 exports.getFinalAnswerRetryError = getFinalAnswerRetryError;
 exports.stripLastToolCalls = stripLastToolCalls;
 exports.repairOrphanedToolCalls = repairOrphanedToolCalls;
@@ -228,7 +229,14 @@ function parseFinalAnswerPayload(argumentsText) {
         return {};
     }
 }
-function getFinalAnswerRetryError(mustResolveBeforeHandoff, intent, sawSteeringFollowUp, delegationDecision, sawSendMessageSelf, sawGoInward, sawQuerySession, innerJob) {
+/** Returns true when a tool call queries external state (GitHub, npm registry). */
+function isExternalStateQuery(toolName, args) {
+    if (toolName !== "shell")
+        return false;
+    const cmd = String(args.command ?? "");
+    return /\bgh\s+(pr|run|api|issue)\b/.test(cmd) || /\bnpm\s+(view|info|show)\b/.test(cmd);
+}
+function getFinalAnswerRetryError(mustResolveBeforeHandoff, intent, sawSteeringFollowUp, delegationDecision, sawSendMessageSelf, sawGoInward, sawQuerySession, currentObligation, innerJob, sawExternalStateQuery) {
     // 1. Delegation adherence: delegate-inward without evidence of inward action
     if (delegationDecision?.target === "delegate-inward" && !sawSendMessageSelf && !sawGoInward && !sawQuerySession) {
         (0, runtime_1.emitNervesEvent)({
@@ -254,8 +262,15 @@ function getFinalAnswerRetryError(mustResolveBeforeHandoff, intent, sawSteeringF
     if (mustResolveBeforeHandoff && intent === "direct_reply" && !sawSteeringFollowUp) {
         return "your final_answer used intent=direct_reply without a newer steering follow-up. continue the unresolved work, or call final_answer again with intent=complete or blocked when appropriate.";
     }
-    // 5. Default malformed fallback
-    return "your final_answer was incomplete or malformed. call final_answer again with your complete response.";
+    // 5. mustResolveBeforeHandoff + complete while a live return loop is still active
+    if (mustResolveBeforeHandoff && intent === "complete" && currentObligation && !sawSteeringFollowUp) {
+        return "you still owe the live session a visible return on this work. don't end the turn yet — continue until you've brought back the external-state update, or use intent=blocked with the concrete blocker.";
+    }
+    // 6. External-state grounding: obligation + complete requires fresh external verification
+    if (intent === "complete" && currentObligation && !sawExternalStateQuery && !sawSteeringFollowUp) {
+        return "you're claiming this work is complete, but the external state hasn't been verified this turn. ground your claim with a fresh check (gh pr view, npm view, gh run view, etc.) before calling final_answer.";
+    }
+    return null;
 }
 // Re-export kick utilities for backward compat
 var kicks_1 = require("./kicks");
@@ -474,6 +489,7 @@ async function runAgent(messages, callbacks, channel, signal, options) {
     let sawGoInward = false;
     let sawQuerySession = false;
     let sawBridgeManage = false;
+    let sawExternalStateQuery = false;
     // Prevent MaxListenersExceeded warning — each iteration adds a listener
     try {
         require("events").setMaxListeners(50, signal);
@@ -582,9 +598,11 @@ async function runAgent(messages, callbacks, channel, signal, options) {
                     // Extract answer from the tool call arguments.
                     // Supports: {"answer":"text","intent":"..."} or "text" (JSON string).
                     const { answer, intent } = parseFinalAnswerPayload(result.toolCalls[0].arguments);
+                    const retryError = getFinalAnswerRetryError(mustResolveBeforeHandoffActive, intent, sawSteeringFollowUp, options?.delegationDecision, sawSendMessageSelf, sawGoInward, sawQuerySession, options?.currentObligation ?? null, options?.activeWorkFrame?.inner?.job, sawExternalStateQuery);
                     const validDirectReply = mustResolveBeforeHandoffActive && intent === "direct_reply" && sawSteeringFollowUp;
                     const validTerminalIntent = intent === "complete" || intent === "blocked";
                     const validClosure = answer != null
+                        && !retryError
                         && (!mustResolveBeforeHandoffActive || validDirectReply || validTerminalIntent);
                     if (validClosure) {
                         completion = {
@@ -622,10 +640,9 @@ async function runAgent(messages, callbacks, channel, signal, options) {
                         // malformed. Clear any partial streamed text or noise, then push the
                         // assistant msg + error tool result and let the model try again.
                         callbacks.onClearText?.();
-                        const retryError = getFinalAnswerRetryError(mustResolveBeforeHandoffActive, intent, sawSteeringFollowUp, options?.delegationDecision, sawSendMessageSelf, sawGoInward);
                         messages.push(msg);
-                        messages.push({ role: "tool", tool_call_id: result.toolCalls[0].id, content: retryError });
-                        providerRuntime.appendToolOutput(result.toolCalls[0].id, retryError);
+                        messages.push({ role: "tool", tool_call_id: result.toolCalls[0].id, content: retryError ?? "your final_answer was incomplete or malformed. call final_answer again with your complete response." });
+                        providerRuntime.appendToolOutput(result.toolCalls[0].id, retryError ?? "your final_answer was incomplete or malformed. call final_answer again with your complete response.");
                     }
                     continue;
                 }
@@ -779,6 +796,9 @@ async function runAgent(messages, callbacks, channel, signal, options) {
                     /* v8 ignore next -- flag tested via truth-check integration tests @preserve */
                     if (tc.name === "bridge_manage")
                         sawBridgeManage = true;
+                    /* v8 ignore next -- flag tested via truth-check integration tests @preserve */
+                    if (isExternalStateQuery(tc.name, args))
+                        sawExternalStateQuery = true;
                     const argSummary = (0, tools_1.summarizeArgs)(tc.name, args);
                     // Confirmation check for mutate tools
                     if ((0, tools_1.isConfirmationRequired)(tc.name) && !options?.skipConfirmation) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ouro.bot/cli",
-  "version": "0.1.0-alpha.91",
+  "version": "0.1.0-alpha.93",
   "main": "dist/heart/daemon/ouro-entry.js",
   "bin": {
     "cli": "dist/heart/daemon/ouro-bot-entry.js",