@pentoshi/clai 3.11.10 → 3.11.12

package/dist/agent/runner.js

@@ -55,7 +55,7 @@ import { analyzeTask, formatTaskAnalysisHint, isNarrowExplicitNmapOperation, } f
 import { computeMaxIterations, computeStepBudget } from "./step-budget.js";
 import { isScratchOnlyWrite } from "./scratch-write.js";
 import { buildDurableEnvelope, WorkLedger, } from "./durable-envelope.js";
-import { COMPACTION_SYSTEM_PROMPT, } from "./compaction-summary.js";
+import { COMPACTION_SYSTEM_PROMPT, COMPACTION_MAX_COMPLETION_TOKENS, COMPACTION_MAP_MAX_COMPLETION_TOKENS, } from "./compaction-summary.js";
 import { maybeAppendPlanModeReminder, PLAN_REMINDER_TOAST, } from "./plan-mode-reminders.js";
 import { LoopGuard } from "./loop-guard.js";
 import { appendInterruptedReasoning, interruptedReasoningBrief, isMeaningfulResumptionYield, } from "./interrupted-reasoning.js";
@@ -557,10 +557,11 @@ export async function runAgentTurn(prompt, options = {}) {
                 imageView: visionAvailable,
                 // E6: slim native constitution when API tool schemas are attached.
                 ...(native ? { slimNative: reliability.slimNativePrompt } : {}),
-                ...(pentestPromptTurn ? { pentest: true } : {}),
             });
         };
-        const systemSections = [renderRequestEnvironmentContext()];
+        const systemSections = [
+            renderRequestEnvironmentContext({ plan: activePlan }),
+        ];
         if (projectContext) {
             systemSections.push(`Project context from .clai/context.md:\n${projectContext}`);
         }
@@ -2828,10 +2829,13 @@ export async function runAgentTurn(prompt, options = {}) {
                     { role: "user", content: summaryPrompt },
                 ],
                 temperature: 0.1,
-                maxTokens: 2_048,
+                maxTokens: stage?.phase === "map"
+                    ? COMPACTION_MAP_MAX_COMPLETION_TOKENS
+                    : COMPACTION_MAX_COMPLETION_TOKENS,
+                thinking: { enabled: false, effort: "none" },
                 signal: options.signal,
             };
-            const response = await streamWithProvider(request, (token) => deltaParser?.push(token), { onStatus: () => undefined });
+            const response = await streamWithProvider(request, (token) => deltaParser?.push(token), { onStatus: () => undefined, maxRetries: 0 });
             deltaParser?.finish();
             const parsed = stripThinking(response.text);
             if (parsed.visible.trim() || !parsed.hasThinking)
@@ -2846,9 +2850,11 @@ export async function runAgentTurn(prompt, options = {}) {
                     { role: "user", content: summaryPrompt },
                 ],
                 temperature: 0,
-                maxTokens: 8_192,
+                maxTokens: stage?.phase === "map"
+                    ? COMPACTION_MAP_MAX_COMPLETION_TOKENS
+                    : COMPACTION_MAX_COMPLETION_TOKENS,
                 thinking: { enabled: false, effort: "none" },
-            });
+            }, { maxRetries: 0 });
             const retryVisible = stripThinking(retry.text).visible.trim();
             if (retryVisible && compactionId) {
                 writeCompactionDelta(compactionId, retryVisible);
@@ -3123,7 +3129,7 @@ export async function runAgentTurn(prompt, options = {}) {
                         ? `${seconds}s`
                         : `${Math.floor(seconds / 60)}m${String(seconds % 60).padStart(2, "0")}s`;
                     if (generatedTokens > 0 && !inThinking) {
-                        return `generating response · ${generatedTokens} tokens · ${elapsed}`;
+                        return `generating response · ${elapsed}`;
                     }
                     if (sawReasoning)
                         return `thinking · ${elapsed}`;
@@ -3168,6 +3174,11 @@ export async function runAgentTurn(prompt, options = {}) {
                     const turnTools = selectToolDefs(nativeToolsActive, useCompactSystemPrompt);
                     toolsAttached = Boolean(turnTools?.length);
                     const contextBreakdown = buildContextBreakdown(messages, toolsAttached ? turnTools : undefined);
+                    emit({
+                        type: "context-estimate",
+                        estimatedTokens: contextBreakdown.estimatedTotalTokens,
+                        model,
+                    });
                     // E4: advisory only — never blocks free-tier users.
                     if (!freeTierLargeContextWarned) {
                         const notices = freeTierGuardNotices({
@@ -3723,7 +3734,10 @@ export async function runAgentTurn(prompt, options = {}) {
                                         const salvageHistoryCalls = ensureUniqueToolCallIds(nativeToolCalls, toolCallIdsInHistory(messages));
                                         const salvagedCallIndex = nativeToolCalls.indexOf(writeTc);
                                         const salvagedCallId = salvageHistoryCalls[salvagedCallIndex]?.id ?? writeTc.id;
-                                        appendAssistantWithTools(messages, assistantText.visible, salvageHistoryCalls, completion.reasoningBlock);
+                                        appendAssistantWithTools(messages, assistantText.visible, salvageHistoryCalls, completion.reasoningBlock ??
+                                            (assistantText.hasThinking && assistantText.thinkContent
+                                                ? { text: assistantText.thinkContent }
+                                                : undefined));
                                         for (const tc of salvageHistoryCalls) {
                                             appendToolResult(messages, tc.id, tc.id === salvagedCallId
                                                 ? `Tool ${tc.name} result (exit=0, ok=true):\nSalvaged partial write: ${lineCount} lines to ${salvaged.path}`
@@ -4073,6 +4087,7 @@ export async function runAgentTurn(prompt, options = {}) {
                         sawServerTail,
                         sawLocalHttpProbe,
                         sawFailedLocalHttpProbe,
+                        serverCriterionRequired: outcomeState.outcome.criteria.some((criterion) => criterion.domain === "server" && criterion.required),
                         sawActivePentestTest,
                         sawSuccessfulMutation,
                         featureAppAsk,
@@ -4256,7 +4271,9 @@ export async function runAgentTurn(prompt, options = {}) {
                 if (sequenceDecision.suppress) {
                     const reason = sequenceDecision.terminal
                         ? "The model repeated the same action sequence after it was already suppressed. No commands were run again."
-                        : "The same action sequence already ran in the previous model round. No commands were run again; reuse the existing results and choose a materially different next action or finish.";
+                        : sequenceDecision.oscillation
+                            ? "This exact action sequence already completed earlier this turn (the agent is oscillating back to finished work). No commands were run again; every one of these results is already in context — synthesize them and either advance to a genuinely new action or finish."
+                            : "The same action sequence already ran in the previous model round. No commands were run again; reuse the existing results and choose a materially different next action or finish.";
                     writeNotice("warn", reason, chalk.yellow(`  ⚠ ${reason}\n`));
                     const suppressedResults = bound.map((b) => {
                         const duplicate = runIds.has(b.id);
@@ -4286,7 +4303,10 @@ export async function runAgentTurn(prompt, options = {}) {
                         emitToolResult(eventId, result, resultReason);
                     }
                     if (historyNativeCalls.length) {
-                        appendAssistantWithTools(messages, beforeTool ?? "", historyNativeCalls, completion.reasoningBlock);
+                        appendAssistantWithTools(messages, beforeTool ?? "", historyNativeCalls, completion.reasoningBlock ??
+                            (assistantText.hasThinking && assistantText.thinkContent
+                                ? { text: assistantText.thinkContent }
+                                : undefined));
                         for (const { b, resultReason, result } of suppressedResults) {
                             appendToolResult(messages, b.id, `Tool ${b.call.name} result (exit=${result.exitCode}, ok=false):\n${resultReason}`, b.call.name, false);
                         }
@@ -4327,7 +4347,10 @@ export async function runAgentTurn(prompt, options = {}) {
                     }
                 }
                 if (historyNativeCalls.length) {
-                    appendAssistantWithTools(messages, beforeTool ?? "", historyNativeCalls, completion.reasoningBlock);
+                    appendAssistantWithTools(messages, beforeTool ?? "", historyNativeCalls, completion.reasoningBlock ??
+                        (assistantText.hasThinking && assistantText.thinkContent
+                            ? { text: assistantText.thinkContent }
+                            : undefined));
                 }
                 else {
                     const standardizedContent = (beforeTool ? beforeTool.trim() + "\n\n" : "") +