@pentoshi/clai 3.12.1 → 3.13.1

package/dist/agent/message-slim.js

@@ -1,33 +1,23 @@
-/**
- * Cap in-memory / re-sent message bulk from native toolCalls.
- *
- * fs.write / writeMany put entire file bodies in tool-call args. Those were
- * stored verbatim in history and LoopGuard signatures, so a scaffold turn
- * could hold many multi-MB strings — memory climbed into the multi-GB range
- * and Macs heated under GC. Full file bodies remain on disk; tool results
- * still report paths/bytes.
- */
 import { createHash } from "node:crypto";
-/** Strings at or above this are replaced with a length+hash stub. */
 export const SLIM_ARG_STRING_CHARS = 400;
-/** Hard ceiling for a single string kept as-is in history (safety). */
 export const SLIM_ARG_ABSOLUTE_MAX_CHARS = 8_000;
-/** Max depth when walking args trees (writeMany files[]). */
 const SLIM_MAX_DEPTH = 6;
+const BULK_ARG_KEYS = new Set(["content", "body"]);
+const ELIDED_STUB_PATTERN = /^«\d+ chars sha256=[0-9a-f]{12}(?:\s+—.*)?»$/s;
 function shortHash(text) {
     return createHash("sha256").update(text).digest("hex").slice(0, 12);
 }
-/**
- * Replace large string values with a compact stub so history/loop-guard
- * never retains full write payloads. Small strings and structure stay intact
- * so path-based identity still works for loop detection.
- */
-export function slimValue(value, depth = 0) {
+function slimLimitForKey(key) {
+    return key !== undefined && BULK_ARG_KEYS.has(key)
+        ? SLIM_ARG_STRING_CHARS
+        : SLIM_ARG_ABSOLUTE_MAX_CHARS;
+}
+export function slimValue(value, depth = 0, key) {
     if (typeof value === "string") {
-        if (value.length < SLIM_ARG_STRING_CHARS)
+        if (value.length < slimLimitForKey(key))
             return value;
         const hash = shortHash(value);
-        return `«${value.length} chars sha256=${hash}»`;
+        return `«${value.length} chars sha256=${hash} — elided from history; never reuse this stub, regenerate the full value»`;
     }
     if (value === null || value === undefined)
         return value;
@@ -42,8 +32,8 @@ export function slimValue(value, depth = 0) {
         return value.map((entry) => slimValue(entry, depth + 1));
     }
     const out = {};
-    for (const key of Object.keys(value).sort()) {
-        out[key] = slimValue(value[key], depth + 1);
+    for (const childKey of Object.keys(value).sort()) {
+        out[childKey] = slimValue(value[childKey], depth + 1, childKey);
     }
     return out;
 }
@@ -54,7 +44,35 @@ export function slimToolArgs(args) {
     }
     return {};
 }
-/** Approximate UTF-16 code units for token budgeting (same basis as content). */
+export function findElidedStubArg(value, path = "args", depth = 0) {
+    if (typeof value === "string") {
+        return ELIDED_STUB_PATTERN.test(value) ? { key: path, value } : undefined;
+    }
+    if (value === null || value === undefined || typeof value !== "object") {
+        return undefined;
+    }
+    if (depth >= SLIM_MAX_DEPTH)
+        return undefined;
+    if (Array.isArray(value)) {
+        for (let i = 0; i < value.length; i += 1) {
+            const found = findElidedStubArg(value[i], `${path}[${i}]`, depth + 1);
+            if (found)
+                return found;
+        }
+        return undefined;
+    }
+    for (const [childKey, childValue] of Object.entries(value)) {
+        const found = findElidedStubArg(childValue, path ? `${path}.${childKey}` : childKey, depth + 1);
+        if (found)
+            return found;
+    }
+    return undefined;
+}
+export function elidedStubReuseMessage(key) {
+    return (`Tool call rejected: argument "${key}" is an elided history placeholder («N chars sha256=…»), not a real value. ` +
+        "Compressed history replaces long arguments with those stubs and the original text cannot be recovered from them. " +
+        "Re-issue the tool call with the complete literal value — never copy «…» stubs from earlier context.");
+}
 export function measureToolCallsChars(toolCalls) {
     if (!toolCalls?.length)
         return 0;

package/dist/agent/message-slim.js.map

@@ -1 +1 @@
-{"version":3,"file":"message-slim.js","sourceRoot":"","sources":["../../src/agent/message-slim.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAEzC,qEAAqE;AACrE,MAAM,CAAC,MAAM,qBAAqB,GAAG,GAAG,CAAC;AACzC,uEAAuE;AACvE,MAAM,CAAC,MAAM,2BAA2B,GAAG,KAAK,CAAC;AACjD,6DAA6D;AAC7D,MAAM,cAAc,GAAG,CAAC,CAAC;AAEzB,SAAS,SAAS,CAAC,IAAY;IAC7B,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACtE,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,SAAS,CAAC,KAAc,EAAE,KAAK,GAAG,CAAC;IACjD,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAC9B,IAAI,KAAK,CAAC,MAAM,GAAG,qBAAqB;YAAE,OAAO,KAAK,CAAC;QACvD,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;QAC9B,OAAO,IAAI,KAAK,CAAC,MAAM,iBAAiB,IAAI,GAAG,CAAC;IAClD,CAAC;IACD,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,KAAK,SAAS;QAAE,OAAO,KAAK,CAAC;IACxD,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IAC5C,IAAI,KAAK,IAAI,cAAc,EAAE,CAAC;QAC5B,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;YAAE,OAAO,KAAK,KAAK,CAAC,MAAM,SAAS,CAAC;QAC5D,OAAO,KAAK,CAAC;IACf,CAAC;IACD,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,KAAK,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC;IAC3D,CAAC;IACD,MAAM,GAAG,GAA4B,EAAE,CAAC;IACxC,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,IAAI,CAAC,KAAgC,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;QACvE,GAAG,CAAC,GAAG,CAAC,GAAG,SAAS,CAAE,KAAiC,CAAC,GAAG,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;IAC3E,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,MAAM,UAAU,YAAY,CAC1B,IAA6B;IAE7B,MAAM,OAAO,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;IAChC,IAAI,OAAO,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACtE,OAAO,OAAkC,CAAC;IAC5C,CAAC;IACD,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,qBAAqB,CACnC,SAMa;IAEb,IAAI,CAAC,SAAS,EAAE,MAAM;QAAE,OAAO,CAAC,CAAC;IACjC,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;QAC3B,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;QACvD,IAAI,EAAE,CAAC,IAAI,EAAE,CAAC;YACZ,IAAI,CAAC;gBACH,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;YACtC,CAAC;YAAC,MAAM,CAAC;gBACP,CAAC,IAAI,EAAE,CAAC;YACV,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC"}
+{"version":3,"file":"message-slim.js","sourceRoot":"","sources":["../../src/agent/message-slim.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAEzC,MAAM,CAAC,MAAM,qBAAqB,GAAG,GAAG,CAAC;AACzC,MAAM,CAAC,MAAM,2BAA2B,GAAG,KAAK,CAAC;AACjD,MAAM,cAAc,GAAG,CAAC,CAAC;AAEzB,MAAM,aAAa,GAAG,IAAI,GAAG,CAAC,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC,CAAC;AAEnD,MAAM,mBAAmB,GAAG,+CAA+C,CAAC;AAE5E,SAAS,SAAS,CAAC,IAAY;IAC7B,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACtE,CAAC;AAED,SAAS,eAAe,CAAC,GAAuB;IAC9C,OAAO,GAAG,KAAK,SAAS,IAAI,aAAa,CAAC,GAAG,CAAC,GAAG,CAAC;QAChD,CAAC,CAAC,qBAAqB;QACvB,CAAC,CAAC,2BAA2B,CAAC;AAClC,CAAC;AAED,MAAM,UAAU,SAAS,CAAC,KAAc,EAAE,KAAK,GAAG,CAAC,EAAE,GAAY;IAC/D,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAC9B,IAAI,KAAK,CAAC,MAAM,GAAG,eAAe,CAAC,GAAG,CAAC;YAAE,OAAO,KAAK,CAAC;QACtD,MAAM,IAAI,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC;QAC9B,OAAO,IAAI,KAAK,CAAC,MAAM,iBAAiB,IAAI,2EAA2E,CAAC;IAC1H,CAAC;IACD,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,KAAK,SAAS;QAAE,OAAO,KAAK,CAAC;IACxD,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IAC5C,IAAI,KAAK,IAAI,cAAc,EAAE,CAAC;QAC5B,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;YAAE,OAAO,KAAK,KAAK,CAAC,MAAM,SAAS,CAAC;QAC5D,OAAO,KAAK,CAAC;IACf,CAAC;IACD,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,KAAK,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC;IAC3D,CAAC;IACD,MAAM,GAAG,GAA4B,EAAE,CAAC;IACxC,KAAK,MAAM,QAAQ,IAAI,MAAM,CAAC,IAAI,CAAC,KAAgC,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC;QAC5E,GAAG,CAAC,QAAQ,CAAC,GAAG,SAAS,CACtB,KAAiC,CAAC,QAAQ,CAAC,EAC5C,KAAK,GAAG,CAAC,EACT,QAAQ,CACT,CAAC;IACJ,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,MAAM,UAAU,YAAY,CAC1B,IAA6B;IAE7B,MAAM,OAAO,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;IAChC,IAAI,OAAO,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACtE,OAAO,OAAkC,CAAC;IAC5C,CAAC;IACD,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,MAAM,UAAU,iBAAiB,CAC/B,KAAc,EACd,IAAI,GAAG,MAAM,EACb,KAAK,GAAG,CAAC;IAET,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QAC9B,OAAO,mBAAmB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IAC5E,CAAC;IACD,IAAI,KAAK,KAAK,IAAI,IAAI,KAAK,KAAK,SAAS,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QACvE,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,IAAI,KAAK,IAAI,cAAc;QAAE,OAAO,SAAS,CAAC;IAC9C,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QACzB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;YACzC,MAAM,KAAK,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,IAAI,CAAC,GAAG,EAAE,KAAK,GAAG,CAAC,CAAC,CAAC;YACtE,IAAI,KAAK;gBAAE,OAAO,KAAK,CAAC;QAC1B,CAAC;QACD,OAAO,SAAS,CAAC;IACnB,CAAC;IACD,KAAK,MAAM,CAAC,QAAQ,EAAE,UAAU,CAAC,IAAI,MAAM,CAAC,OAAO,CACjD,KAAgC,CACjC,EAAE,CAAC;QACF,MAAM,KAAK,GAAG,iBAAiB,CAC7B,UAAU,EACV,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,IAAI,QAAQ,EAAE,CAAC,CAAC,CAAC,QAAQ,EACvC,KAAK,GAAG,CAAC,CACV,CAAC;QACF,IAAI,KAAK;YAAE,OAAO,KAAK,CAAC;IAC1B,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,GAAW;IAChD,OAAO,CACL,iCAAiC,GAAG,6EAA6E;QACjH,mHAAmH;QACnH,qGAAqG,CACtG,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CACnC,SAMa;IAEb,IAAI,CAAC,SAAS,EAAE,MAAM;QAAE,OAAO,CAAC,CAAC;IACjC,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;QAC3B,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;QACvD,IAAI,EAAE,CAAC,IAAI,EAAE,CAAC;YACZ,IAAI,CAAC;gBACH,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC;YACtC,CAAC;YAAC,MAAM,CAAC;gBACP,CAAC,IAAI,EAAE,CAAC;YACV,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC"}

package/dist/agent/runner.js

@@ -1,6 +1,7 @@
 import chalk from "chalk";
 import { join } from "node:path";
-import { completeWithProvider, streamWithProvider } from "../llm/router.js";
+import { streamWithProvider } from "../llm/router.js";
+import { modelContextWindow } from "../llm/token-usage.js";
 import { streamAlreadyEmitted } from "../llm/stream-progress.js";
 import { classifyStreamFailure, planStreamRecovery, recordRecoveryAttempt, createStreamRecoveryState, resetStreamRecoveryState, } from "./stream-recovery.js";
 import { modelSupportsVision, resolveToolDialect } from "../llm/capabilities.js";
@@ -38,6 +39,7 @@ function safeEngagementActionsForToolCall(call) {
 }
 import { availableToolNames, normalizeToolCall, runToolCall, BATCH_SAFE_TOOLS, } from "../tools/registry.js";
 import { getToolDefinitions, getCompactToolDefinitions, RUNNER_META_TOOL_NAMES, } from "../tools/definitions.js";
+import { elidedStubReuseMessage, findElidedStubArg, } from "./message-slim.js";
 import { appendAssistantWithTools, ensureUniqueToolCallIds, toolCallIdsInHistory, appendToolResult, assertValidToolProtocol, fillMissingToolResults, repairToolProtocol, } from "./tool-history.js";
 import { formatViewportHint, registerViewport } from "../ui/output-pane.js";
 import { compactMessagesWithSummary, shouldApplyAutoCompact, COMPACTION_MEMORY_PREFIX, PLAN_IMPLEMENT_MEMORY_PREFIX, isCompactionMemoryMessage, } from "./context-manager.js";
@@ -55,7 +57,7 @@ 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, COMPACTION_MAX_COMPLETION_TOKENS, COMPACTION_MAP_MAX_COMPLETION_TOKENS, } from "./compaction-summary.js";
+import { buildCompactionRetryPrompt, compactionSinglePassInputBudget, COMPACTION_SYSTEM_PROMPT, COMPACTION_MAX_COMPLETION_TOKENS, COMPACTION_MAP_MAX_COMPLETION_TOKENS, isCompactionCompletionTruncated, looksLikeIncompleteCompactionSummary, looksLikeTranscriptReplay, normalizeCompactionSummary, } 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";
@@ -315,12 +317,21 @@ export async function runAgentTurn(prompt, options = {}) {
             process.stdout.write(chalk.dim("  ✦ Compacted Context · streaming Markdown\n\n"));
         }
     };
-    const writeCompactionDelta = (id, text) => {
-        if (!text)
+    const writeCompactionDelta = (id, text, replace = false) => {
+        if (!text && !replace)
             return;
-        emit({ type: "compaction-delta", id, text });
-        if (writesDirectly)
-            process.stdout.write(text);
+        emit({
+            type: "compaction-delta",
+            id,
+            text,
+            ...(replace ? { replace: true } : {}),
+        });
+        if (writesDirectly) {
+            if (replace)
+                process.stdout.write(chalk.dim("\n  ↻ rewriting compacted context\n\n"));
+            if (text)
+                process.stdout.write(text);
+        }
     };
     const writeCompactionCompleted = (id, summary, beforeTokens, afterTokens) => {
         emit({
@@ -1255,6 +1266,13 @@ export async function runAgentTurn(prompt, options = {}) {
                 emitToolResult(toolEventId, result, reason);
                 return { ok: false, call, result, contextOutput: reason };
             }
+            const elidedStub = findElidedStubArg(call.args);
+            if (elidedStub) {
+                const reason = elidedStubReuseMessage(elidedStub.key);
+                const result = { ok: false, output: reason, exitCode: 1 };
+                emitToolResult(toolEventId, result, reason);
+                return { ok: false, call, result, contextOutput: reason };
+            }
             if (call.name === "image.ocr" && !imageOcrEnabled) {
                 writeNotice("info", "skipped OCR because the original image is attached to the vision model", chalk.dim("  ℹ skipped OCR — inspecting the attached image directly\n"));
                 const recoveryText = "The original image is attached to this message and you can inspect it directly. " +
@@ -2773,50 +2791,105 @@ export async function runAgentTurn(prompt, options = {}) {
         const summarizeForCompaction = async (summaryPrompt, stage) => {
             const streamFinalSummary = stage?.phase !== "map";
             const compactionId = streamFinalSummary ? activeCompactionId : undefined;
-            const deltaParser = compactionId
-                ? createThinkingStreamParser((text) => writeCompactionDelta(compactionId, text), undefined, { remember: false })
+            const maxTokens = stage?.phase === "map"
+                ? COMPACTION_MAP_MAX_COMPLETION_TOKENS
+                : COMPACTION_MAX_COMPLETION_TOKENS;
+            const sourceMessages = stage?.sourceMessages;
+            const compactionTools = sourceMessages
+                ? selectToolDefs(nativeToolsActive, useCompactSystemPrompt)
                 : undefined;
             const request = {
                 provider,
                 model,
-                messages: [
-                    { role: "system", content: COMPACTION_SYSTEM_PROMPT },
-                    { role: "user", content: summaryPrompt },
-                ],
+                messages: sourceMessages
+                    ? [
+                        ...sourceMessages,
+                        { role: "user", content: summaryPrompt },
+                    ]
+                    : [
+                        { role: "system", content: COMPACTION_SYSTEM_PROMPT },
+                        { role: "user", content: summaryPrompt },
+                    ],
                 temperature: 0.1,
-                maxTokens: stage?.phase === "map"
-                    ? COMPACTION_MAP_MAX_COMPLETION_TOKENS
-                    : COMPACTION_MAX_COMPLETION_TOKENS,
+                maxTokens,
                 thinking: { enabled: false, effort: "none" },
                 signal: options.signal,
                 allowModelFallback: true,
+                ...(compactionTools?.length
+                    ? {
+                        tools: compactionTools,
+                        toolChoice: "none",
+                    }
+                    : {}),
             };
-            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)
-                return response.text;
-            const retry = await completeWithProvider({
-                ...request,
-                messages: [
-                    {
-                        role: "system",
-                        content: `${COMPACTION_SYSTEM_PROMPT}\nReturn only the continuation-memory summary. Do not include analysis, reasoning, or <think> tags.`,
-                    },
-                    { role: "user", content: summaryPrompt },
-                ],
-                temperature: 0,
-                maxTokens: stage?.phase === "map"
-                    ? COMPACTION_MAP_MAX_COMPLETION_TOKENS
-                    : COMPACTION_MAX_COMPLETION_TOKENS,
-                thinking: { enabled: false, effort: "none" },
-                allowModelFallback: true,
-            }, { maxRetries: 0 });
-            const retryVisible = stripThinking(retry.text).visible.trim();
-            if (retryVisible && compactionId) {
-                writeCompactionDelta(compactionId, retryVisible);
+            const runAttempt = async (attemptRequest, replace = false) => {
+                if (compactionId && replace) {
+                    writeCompactionDelta(compactionId, "", true);
+                }
+                const parser = createThinkingStreamParser((text) => {
+                    if (compactionId)
+                        writeCompactionDelta(compactionId, text);
+                }, undefined, { remember: false });
+                const result = await streamWithProvider(attemptRequest, (token) => parser.push(token), { onStatus: () => undefined, maxRetries: 0 });
+                parser.finish();
+                return result;
+            };
+            const first = await runAttempt(request);
+            let visible = normalizeCompactionSummary(stripThinking(first.text).visible);
+            let retryReason;
+            if (isCompactionCompletionTruncated(first, maxTokens)) {
+                retryReason = "truncated";
+            }
+            else if (!visible) {
+                retryReason = "reasoning-only";
+            }
+            else if (looksLikeTranscriptReplay(visible)) {
+                retryReason = "replayed";
+            }
+            else if (looksLikeIncompleteCompactionSummary(visible)) {
+                retryReason = "incomplete";
+            }
+            if (retryReason) {
+                const retry = await runAttempt({
+                    ...request,
+                    messages: sourceMessages
+                        ? [
+                            ...sourceMessages,
+                            {
+                                role: "user",
+                                content: buildCompactionRetryPrompt(summaryPrompt, retryReason),
+                            },
+                        ]
+                        : [
+                            {
+                                role: "system",
+                                content: `${COMPACTION_SYSTEM_PROMPT}\nReturn only a complete continuation-memory summary. Do not include analysis, reasoning, or <think> tags.`,
+                            },
+                            {
+                                role: "user",
+                                content: buildCompactionRetryPrompt(summaryPrompt, retryReason),
+                            },
+                        ],
+                    temperature: 0,
+                    maxTokens,
+                    thinking: { enabled: false, effort: "none" },
+                    allowModelFallback: true,
+                }, true);
+                if (isCompactionCompletionTruncated(retry, maxTokens)) {
+                    throw new Error("compaction failed: model hit the summary output limit twice — original context retained");
+                }
+                visible = normalizeCompactionSummary(stripThinking(retry.text).visible);
+                if (!visible) {
+                    throw new Error("compaction failed: model returned an empty summary");
+                }
+                if (looksLikeTranscriptReplay(visible)) {
+                    throw new Error("compaction failed: model replayed the transcript twice — original context retained");
+                }
+                if (looksLikeIncompleteCompactionSummary(visible)) {
+                    throw new Error("compaction failed: model returned an incomplete summary twice — original context retained");
+                }
             }
-            return retry.text;
+            return visible;
         };
         /**
          * Estimate the complete next model request, including attached native-tool
@@ -2909,9 +2982,12 @@ export async function runAgentTurn(prompt, options = {}) {
             activeCompactionId = compactionId;
             writeCompactionStarted(compactionId, beforeTokens);
             try {
+                const compactionTools = selectToolDefs(nativeToolsActive, useCompactSystemPrompt);
+                const compactionSchemaTokens = buildContextBreakdown([], compactionTools).estimatedTotalTokens;
                 const result = await compactMessagesWithSummary(messages, summarizeForCompaction, {
                     budgetTokens: 0,
                     keepRecent: AUTO_COMPACT_KEEP_RECENT,
+                    singlePassInputBudgetTokens: Math.max(0, compactionSinglePassInputBudget(contextLimitTokens ?? modelContextWindow(model, provider)) - compactionSchemaTokens),
                     ...(durableEnvelope ? { durableEnvelope } : {}),
                 });
                 const summaryBody = result.messages.find((m) => isCompactionMemoryMessage(m))?.content ??
@@ -3098,6 +3174,7 @@ export async function runAgentTurn(prompt, options = {}) {
                 }, 10_000);
                 heartbeat.unref?.();
                 const deferredToolCalls = [];
+                const streamedNativeCallNames = new Map();
                 const deltaParser = writesDirectly
                     ? undefined
                     : createThinkingStreamParser((text) => emit({ type: "assistant-delta", text }), (text) => {
@@ -3208,64 +3285,22 @@ export async function runAgentTurn(prompt, options = {}) {
                                     tools: turnTools,
                                     toolChoice: "auto",
                                     parallelToolCalls: true,
-                                    // P2-3: emit tool cards as soon as the function name arrives.
                                     onToolCallDelta: (delta) => {
                                         if (!delta.name)
                                             return;
                                         const name = fromWireName(delta.name) ?? delta.name;
-                                        const existing = deferredToolCalls[delta.index];
-                                        if (existing && existing.call.name !== "…") {
-                                            if (delta.argumentsBytes &&
-                                                delta.argumentsBytes >= 4096 &&
-                                                !writesDirectly) {
-                                                emit({
-                                                    type: "status",
-                                                    text: `${name} (${Math.round(delta.argumentsBytes / 1024)}KB args)`,
-                                                });
-                                            }
-                                            return;
-                                        }
-                                        while (deferredToolCalls.length < delta.index) {
-                                            const slot = deferredToolCalls.length;
-                                            const placeholderId = `tool-${++nextToolEventId}`;
-                                            callIds[slot] = placeholderId;
-                                            deferredToolCalls.push({
-                                                eventId: placeholderId,
-                                                call: { name: "…", args: {} },
-                                                rendered: "",
-                                                shown: false,
-                                            });
-                                        }
-                                        const call = normalizeToolCall({
-                                            name,
-                                            args: {},
-                                        });
-                                        const eventId = existing?.eventId ?? `tool-${++nextToolEventId}`;
-                                        callIds[delta.index] = eventId;
-                                        alreadyPrintedIds.add(eventId);
-                                        const toolCallLine = chalk.cyan(`  ▶ ${call.name}`) +
-                                            chalk.gray(` ${formatToolArgs(call)}`);
-                                        const entry = {
-                                            eventId,
-                                            call,
-                                            rendered: styleToolChatter(call, toolCallLine) + "\n",
-                                            shown: false,
-                                        };
-                                        if (deferredToolCalls.length === delta.index) {
-                                            deferredToolCalls.push(entry);
-                                        }
-                                        else {
-                                            deferredToolCalls[delta.index] = entry;
-                                        }
-                                        streamedCallsCount = Math.max(streamedCallsCount, deferredToolCalls.length);
+                                        streamedNativeCallNames.set(delta.index, name);
                                         if (!writesDirectly) {
-                                            writeToolCall(eventId, call, entry.rendered);
-                                            entry.shown = true;
-                                            emit({ type: "status", text: call.name });
+                                            emit({
+                                                type: "status",
+                                                text: delta.argumentsBytes && delta.argumentsBytes >= 4096
+                                                    ? `${name} (${Math.round(delta.argumentsBytes / 1024)}KB args)`
+                                                    : name,
+                                            });
                                         }
                                         else {
                                             spinner.stop();
-                                            spinner = startThinkingSpinner(`tool ${call.name}…`, options.signal);
+                                            spinner = startThinkingSpinner(`tool ${name}…`, options.signal);
                                         }
                                     },
                                 }
@@ -3283,7 +3318,7 @@ export async function runAgentTurn(prompt, options = {}) {
                                         spinner.stop();
                                     }
                                     while (streamedCallsCount < parsedCalls.length) {
-                                        const call = parsedCalls[streamedCallsCount];
+                                        const call = normalizeToolCall(parsedCalls[streamedCallsCount]);
                                         const eventId = `tool-${++nextToolEventId}`;
                                         callIds[streamedCallsCount] = eventId;
                                         alreadyPrintedIds.add(eventId);
@@ -3788,7 +3823,7 @@ export async function runAgentTurn(prompt, options = {}) {
                         continue;
                     }
                     const incompleteNativeStream = nativeToolCalls.length === 0 &&
-                        deferredToolCalls.some((entry) => entry.shown && entry.call.name !== "…");
+                        streamedNativeCallNames.size > 0;
                     if (incompleteNativeStream) {
                         const reason = "The provider began this native tool call but never completed it. Nothing ran; reissue a complete call.";
                         for (const deferred of deferredToolCalls) {
@@ -4201,13 +4236,14 @@ export async function runAgentTurn(prompt, options = {}) {
                     let parsed = parseAllToolCalls(assistantText.visible || assistantText.thinkContent);
                     if (parsed.length === 0 && call)
                         parsed = [call];
-                    bound = parsed.map((c, index) => {
+                    bound = parsed.map((rawCall, index) => {
+                        const call = normalizeToolCall(rawCall);
                         const id = syntheticToolCallId(index);
                         return {
                             index,
                             id,
-                            call: c,
-                            native: { id, name: c.name, args: c.args },
+                            call,
+                            native: { id, name: call.name, args: call.args },
                         };
                     });
                 }
@@ -4377,7 +4413,7 @@ export async function runAgentTurn(prompt, options = {}) {
                 for (const deferred of activeDeferredToolCalls.slice(0, allCalls.length)) {
                     if (!deferred.call.name || deferred.call.name === "…")
                         continue;
-                    if (!deferred.shown || !writesDirectly) {
+                    if (!deferred.shown) {
                         writeToolCall(deferred.eventId, deferred.call, deferred.rendered);
                         deferred.shown = true;
                     }