@selesai/code 0.10.2 → 0.12.0

package/CHANGELOG.md

@@ -2,6 +2,40 @@
 
 All notable changes to `@selesai/code` will be documented in this file.
 
+## [0.12.0] - 2026-08-29
+
+### Added
+- **`/tokenin usage` subcommand.** The `tokenin-onboarding` extension now queries the LiteLLM `/key/info` endpoint with the active Token-In token and shows a boxed spend/budget/remaining/reset display with a progress bar. Run `/tokenin usage` to check quota without leaving the session.
+- **Terminal capability overrides.** New `terminal.images` (`"kitty"`, `"iterm2"`, `"auto"`, or `false`), `terminal.trueColor`, and `terminal.hyperlinks` settings let you pin terminal capabilities instead of relying on auto-detection.
+- **`fullscreenCopyOnSelect` setting.** New setting (default `true`) and `/settings` toggle: automatically copy selected text in fullscreen mode; disable to copy selections with Ctrl+X.
+- **`ui_prompt_start` / `ui_prompt_end` extension events.** Extensions can now observe when the agent starts and stops waiting on a blocking user-facing UI prompt (`select`, `confirm`, `input`, `editor`, `custom`).
+- **llama.cpp router autoload.** Unloaded preset models are now selectable when the llama.cpp router has `models_autoload` enabled, so they load on first use instead of being hidden.
+- **RPC `clear_queue` command.** RPC clients can clear the pending steering/follow-up queue and get back the dropped items.
+
+### Changed
+- **Custom message ordering during streaming.** Context-only custom messages queued while the agent is streaming are now flushed at turn end (after tool results are in) instead of being appended between a tool call and its result, which some providers reject on replay.
+- **Auto-compaction before the next assistant response.** The session now checks the context threshold before preparing the next turn and compacts when needed, so long conversations stay within the model window.
+- **Persisted default model joins scoped model lists.** When a scoped model list (`--models`) is active, the persisted default model is added to the scope and to `enabledModels` so it stays reachable.
+- **Model catalog updates.** Celestial, DeepSeek, and GPT models now declare `compat.requiresReasoningContentOnAssistantMessages` where required.
+- **Windows process cleanup.** `taskkill` is invoked via the trusted System32 path instead of relying on `PATH`, and failed spawns no longer crash Node.
+
+### Fixed
+- **Footer provider count after model switch.** The interactive footer's available-provider count now refreshes immediately after switching models in the model selector.
+- **Summarization length-stop guard.** A summarization response that hits the token cap (partial text) is now treated as a failure and never persisted as a session checkpoint.
+
+## [0.11.0] - 2026-08-29
+
+### Added
+- **Auto session naming.** The new `auto-session-name` extension names the session on every user message using the currently selected model. It sends only user-message history (no tool results, assistant replies, or thinking) with a 10-token output cap and a 5-word name limit, and runs in the background so it never delays your message. Thinking is explicitly disabled via `reasoning_effort: "none"` for gateways that ignore the standard `thinking: {type:"disabled"}` field.
+- **Cost reconciliation.** The new `cost-reconcile` extension records the real billed cost reported by gateways (LiteLLM `x-litellm-response-cost` header, OpenRouter `usage.cost`/`total_cost`) as session entries; the zentui footer prefers these authoritative amounts over rate-card estimates.
+- **Tool display.** The `pi-tool-display` extension is now bundled: compact OpenCode-style tool rendering, diff visualization, output truncation, a native user message box, and the `/tool-display` command with presets.
+
+### Changed
+- **Read-only tool bootstrap.** `grep`, `find`, and `ls` are registered only when not already owned by another extension (e.g. `pi-tool-display`), so boxed renderers are no longer discarded on startup.
+- **Zentui user messages.** The user-message markdown theme now uses the `accent` color instead of `userMessageText`.
+- **Zentui cost totals.** The footer refreshes once ~12s after settle so late gateway cost reconciliation is picked up.
+- **Bundled extensions.** `pi-tool-display`'s user-owned `config.json` is excluded from the shipped package; the default `packages` list no longer installs it from git.
+
 ## [0.10.2] - 2026-08-28
 
 ### Added

package/dist/cli/credential-print.js

@@ -7,10 +7,10 @@ export function isCredentialPrintHelp(args) {
     return (args[0] === "auth" && (args[1] === undefined || args[1] === "help" || args[1] === "--help" || args[1] === "-h"));
 }
 export function printCredentialPrintHelp() {
-    console.log(`Usage:
-  ${APP_NAME} auth print-api-key --model <model> [--provider <provider>]
-  ${APP_NAME} auth print-bearer-token --model <model> [--provider <provider>] [--min-expiry <duration>]
-
+    console.log(`Usage:
+  ${APP_NAME} auth print-api-key --model <model> [--provider <provider>]
+  ${APP_NAME} auth print-bearer-token --model <model> [--provider <provider>] [--min-expiry <duration>]
+
 Prints the configured credential alone on stdout. Provider inference uses configured credentials; specify --provider to select explicitly. Bearer tokens have a 30-minute minimum expiry by default. --min-expiry accepts ms, s, m, or h (for example, 30m).`);
 }
 /** Parse the auth command surface before normal startup. */

package/dist/cli/startup-ui.js

@@ -1,4 +1,4 @@
-import { ProcessTerminal, setKeybindings, TuiMainScreen } from "@earendil-works/pi-tui";
+import { ProcessTerminal, setCapabilityOverrides, setKeybindings, TuiMainScreen, } from "@earendil-works/pi-tui";
 import { existsSync } from "fs";
 import { ENV_AGENT_DIR, getAgentDir, getFirstRunMarkerPath, markFirstRunComplete } from "../config.js";
 import { KeybindingsManager } from "../core/keybindings.js";
@@ -43,6 +43,7 @@ async function loadStartupThemes(settingsManager) {
     return loadThemes(resolvedPaths.themes);
 }
 export async function createStartupTui(settingsManager) {
+    setCapabilityOverrides(settingsManager.getTerminalCapabilityOverrides());
     setRegisteredThemes(await loadStartupThemes(settingsManager));
     const terminalTheme = detectTerminalBackgroundFromEnv().theme;
     initTheme(resolveThemeSetting(settingsManager.getThemeSetting(), terminalTheme) ?? terminalTheme);

package/dist/core/agent-session.d.ts

@@ -218,6 +218,8 @@ export declare class AgentSession {
     private _followUpMessages;
     /** Messages queued to be included with the next user prompt as context ("asides"). */
     private _pendingNextTurnMessages;
+    /** Context-only custom messages queued during a run, flushed once the current turn's tool results are in. */
+    private _pendingCustomMessages;
     private _compactionAbortController;
     private _autoCompactionAbortController;
     private _overflowRecoveryAttempted;
@@ -266,6 +268,7 @@ export declare class AgentSession {
      * happens here instead of in wrappers.
      */
     private _installAgentToolHooks;
+    private _compactBeforeNextAssistantResponse;
     private _installAgentNextTurnRefresh;
     /** Emit an event to all listeners */
     private _emit;
@@ -412,8 +415,9 @@ export declare class AgentSession {
     /**
      * Send a custom message to the session. Creates a CustomMessageEntry.
      *
-     * Handles three cases:
+     * Handles four cases:
      * - Streaming: queues message, processed when loop pulls from queue
+     * - Streaming + triggerTurn false: appended to state/session once the current turn ends
      * - Not streaming + triggerTurn: appends to state/session, starts new turn
      * - Not streaming + no trigger: appends to state/session, no turn
      *
@@ -425,6 +429,12 @@ export declare class AgentSession {
         triggerTurn?: boolean;
         deliverAs?: "steer" | "followUp" | "nextTurn";
     }): Promise<void>;
+    private _appendCustomMessage;
+    /**
+     * Append custom messages queued while the agent was running.
+     * Called once the current turn's tool results are in agent state and session history.
+     */
+    private _flushPendingCustomMessages;
     private _captionImagesForCurrentModel;
     /**
      * Build a small, text-only slice of the most recent conversation to give the
@@ -485,6 +495,7 @@ export declare class AgentSession {
      * @throws Error if no auth is configured for the model
      */
     setModel(model: Model<any>, options?: ModelMutationOptions): Promise<void>;
+    private _addPersistedDefaultToNonEmptyScope;
     /**
      * Cycle to next/previous model.
      * Uses scoped models (from --models flag) if available, otherwise all available models.

package/dist/core/agent-session.js

@@ -102,6 +102,8 @@ export class AgentSession {
     _followUpMessages = [];
     /** Messages queued to be included with the next user prompt as context ("asides"). */
     _pendingNextTurnMessages = [];
+    /** Context-only custom messages queued during a run, flushed once the current turn's tool results are in. */
+    _pendingCustomMessages = [];
     // Compaction state
     _compactionAbortController = undefined;
     _autoCompactionAbortController = undefined;
@@ -269,18 +271,33 @@ export class AgentSession {
             };
         };
     }
+    async _compactBeforeNextAssistantResponse(context) {
+        const model = this.model;
+        const settings = this.settingsManager.getCompactionSettings();
+        if (!model ||
+            model.contextWindow <= 0 ||
+            !shouldCompact(estimateContextTokens(context.messages).tokens, model.contextWindow, settings)) {
+            return context;
+        }
+        await this._runAutoCompaction("threshold", false);
+        return {
+            ...context,
+            messages: this.agent.state.messages.slice(),
+        };
+    }
     _installAgentNextTurnRefresh() {
         const previousPrepareNextTurnWithContext = this.agent.prepareNextTurnWithContext ??
             (this.agent.prepareNextTurn
                 ? async (_turn, signal) => await this.agent.prepareNextTurn?.(signal)
                 : undefined);
         this.agent.prepareNextTurnWithContext = async (turn, signal) => {
-            const previousSnapshot = await previousPrepareNextTurnWithContext?.(turn, signal);
-            const previousContext = previousSnapshot?.context ?? turn.context;
+            const context = await this._compactBeforeNextAssistantResponse(turn.context);
+            const previousSnapshot = await previousPrepareNextTurnWithContext?.({ ...turn, context }, signal);
+            const nextContext = previousSnapshot?.context ?? context;
             return {
                 ...previousSnapshot,
                 context: {
-                    ...previousContext,
+                    ...nextContext,
                     systemPrompt: this._systemPromptOverride ?? this._baseSystemPrompt,
                     tools: this.agent.state.tools.slice(),
                 },
@@ -436,6 +453,14 @@ export class AgentSession {
                 }
             }
         }
+        // A turn ends after its assistant message and every tool result has been appended,
+        // so this is the first point in the run where a context-only custom message can be
+        // inserted without landing between a tool call and its result. Flushing after the
+        // extension and listener dispatch above also picks up messages that turn_end
+        // handlers queued.
+        if (event.type === "turn_end") {
+            this._flushPendingCustomMessages();
+        }
     };
     _willRetryAfterAgentEnd(event) {
         const settings = this.settingsManager.getRetrySettings();
@@ -813,6 +838,7 @@ export class AgentSession {
         finally {
             this._systemPromptOverride = undefined;
             this._flushPendingBashMessages();
+            this._flushPendingCustomMessages();
             await this._emitAgentSettled();
         }
     }
@@ -908,8 +934,9 @@ export class AgentSession {
                 preflightResult?.(true);
                 return;
             }
-            // Flush any pending bash messages before the new prompt
+            // Flush any pending bash and custom messages before the new prompt
             this._flushPendingBashMessages();
+            this._flushPendingCustomMessages();
             // Validate model
             if (!this.model) {
                 throw new Error(formatNoModelSelectedMessage());
@@ -1138,8 +1165,9 @@ export class AgentSession {
     /**
      * Send a custom message to the session. Creates a CustomMessageEntry.
      *
-     * Handles three cases:
+     * Handles four cases:
      * - Streaming: queues message, processed when loop pulls from queue
+     * - Streaming + triggerTurn false: appended to state/session once the current turn ends
      * - Not streaming + triggerTurn: appends to state/session, starts new turn
      * - Not streaming + no trigger: appends to state/session, no turn
      *
@@ -1171,11 +1199,34 @@ export class AgentSession {
         else if (options?.triggerTurn) {
             await this._runAgentPrompt(appMessage);
         }
+        else if (this.isStreaming) {
+            // Appending now would put the message between an assistant tool call and its
+            // result, which providers that validate message order reject on replay. Defer
+            // to the end of the turn. Nothing is emitted yet: message events must not
+            // describe messages the session tree does not contain.
+            this._pendingCustomMessages.push(appMessage);
+        }
         else {
-            this.agent.state.messages.push(appMessage);
-            this.sessionManager.appendCustomMessageEntry(message.customType, message.content, message.display, message.details);
-            this._emit({ type: "message_start", message: appMessage });
-            this._emit({ type: "message_end", message: appMessage });
+            this._appendCustomMessage(appMessage);
+        }
+    }
+    _appendCustomMessage(appMessage) {
+        this.agent.state.messages.push(appMessage);
+        this.sessionManager.appendCustomMessageEntry(appMessage.customType, appMessage.content, appMessage.display, appMessage.details);
+        this._emit({ type: "message_start", message: appMessage });
+        this._emit({ type: "message_end", message: appMessage });
+    }
+    /**
+     * Append custom messages queued while the agent was running.
+     * Called once the current turn's tool results are in agent state and session history.
+     */
+    _flushPendingCustomMessages() {
+        if (this._pendingCustomMessages.length === 0)
+            return;
+        const pending = this._pendingCustomMessages;
+        this._pendingCustomMessages = [];
+        for (const appMessage of pending) {
+            this._appendCustomMessage(appMessage);
         }
     }
     /**
@@ -1448,10 +1499,25 @@ export class AgentSession {
         this.sessionManager.appendModelChange(model.provider, model.id);
         if (options.persist) {
             this.settingsManager.setDefaultModelAndProvider(model.provider, model.id);
+            this._addPersistedDefaultToNonEmptyScope(model);
         }
         this.setThinkingLevel(thinkingLevel);
         await this._emitModelSelect(model, previousModel, "set");
     }
+    _addPersistedDefaultToNonEmptyScope(model) {
+        if (this._scopedModels.length === 0)
+            return;
+        if (this._scopedModels.some((scoped) => modelsAreEqual(scoped.model, model)))
+            return;
+        this._scopedModels = [...this._scopedModels, { model }];
+        const enabledModels = this.settingsManager.getEnabledModels();
+        if (!enabledModels?.length)
+            return;
+        const modelReference = `${model.provider}/${model.id}`;
+        if (enabledModels.some((pattern) => pattern.toLowerCase() === modelReference.toLowerCase()))
+            return;
+        this.settingsManager.setEnabledModels([...enabledModels, modelReference]);
+    }
     /**
      * Cycle to next/previous model.
      * Uses scoped models (from --models flag) if available, otherwise all available models.
@@ -1485,6 +1551,7 @@ export class AgentSession {
         this.sessionManager.appendModelChange(next.model.provider, next.model.id);
         if (options.persist) {
             this.settingsManager.setDefaultModelAndProvider(next.model.provider, next.model.id);
+            this._addPersistedDefaultToNonEmptyScope(next.model);
         }
         // Apply thinking level.
         // - Explicit scoped model thinking level overrides current session level
@@ -1510,6 +1577,7 @@ export class AgentSession {
         this.sessionManager.appendModelChange(nextModel.provider, nextModel.id);
         if (options.persist) {
             this.settingsManager.setDefaultModelAndProvider(nextModel.provider, nextModel.id);
+            this._addPersistedDefaultToNonEmptyScope(nextModel);
         }
         // Re-clamp thinking level for new model's capabilities
         this.setThinkingLevel(thinkingLevel);

package/dist/core/compaction/branch-summarization.js

@@ -6,7 +6,7 @@
  */
 import { contentText } from "@earendil-works/pi-ai";
 import { convertToLlm, createBranchSummaryMessage, createCompactionSummaryMessage, createCustomMessage, } from "../messages.js";
-import { completeSummarization, estimateTokens } from "./compaction.js";
+import { completeSummarization, estimateTokens, getSummarizationFailure } from "./compaction.js";
 import { computeFileLists, createFileOps, extractFileOpsFromMessage, formatFileOperations, SUMMARIZATION_SYSTEM_PROMPT, serializeConversation, } from "./utils.js";
 // ============================================================================
 // Entry Collection
@@ -146,37 +146,37 @@ export function prepareBranchEntries(entries, tokenBudget = 0) {
 // ============================================================================
 // Summary Generation
 // ============================================================================
-const BRANCH_SUMMARY_PREAMBLE = `The user explored a different conversation branch before returning here.
-Summary of that exploration:
-
+const BRANCH_SUMMARY_PREAMBLE = `The user explored a different conversation branch before returning here.
+Summary of that exploration:
+
 `;
-const BRANCH_SUMMARY_PROMPT = `Create a structured summary of this conversation branch for context when returning later.
-
-Use this EXACT format:
-
-## Goal
-[What was the user trying to accomplish in this branch?]
-
-## Constraints & Preferences
-- [Any constraints, preferences, or requirements mentioned]
-- [Or "(none)" if none were mentioned]
-
-## Progress
-### Done
-- [x] [Completed tasks/changes]
-
-### In Progress
-- [ ] [Work that was started but not finished]
-
-### Blocked
-- [Issues preventing progress, if any]
-
-## Key Decisions
-- **[Decision]**: [Brief rationale]
-
-## Next Steps
-1. [What should happen next to continue this work]
-
+const BRANCH_SUMMARY_PROMPT = `Create a structured summary of this conversation branch for context when returning later.
+
+Use this EXACT format:
+
+## Goal
+[What was the user trying to accomplish in this branch?]
+
+## Constraints & Preferences
+- [Any constraints, preferences, or requirements mentioned]
+- [Or "(none)" if none were mentioned]
+
+## Progress
+### Done
+- [x] [Completed tasks/changes]
+
+### In Progress
+- [ ] [Work that was started but not finished]
+
+### Blocked
+- [Issues preventing progress, if any]
+
+## Key Decisions
+- **[Decision]**: [Brief rationale]
+
+## Next Steps
+1. [What should happen next to continue this work]
+
 Keep each section concise. Preserve exact file paths, function names, and error messages.`;
 /**
  * Generate a summary of abandoned branch entries.
@@ -227,8 +227,9 @@ export async function generateBranchSummary(entries, options) {
     if (response.stopReason === "aborted") {
         return { aborted: true };
     }
-    if (response.stopReason === "error") {
-        return { error: response.errorMessage || "Summarization failed" };
+    const failure = getSummarizationFailure(response, "Branch summarization");
+    if (failure) {
+        return { error: failure };
     }
     let summary = contentText(response.content);
     // Prepend preamble to provide context about the branch summary

package/dist/core/compaction/compaction.d.ts

@@ -90,6 +90,11 @@ export interface CutPointResult {
  * Only considers entries between `startIndex` and `endIndex` (exclusive).
  */
 export declare function findCutPoint(entries: SessionEntry[], startIndex: number, endIndex: number, keepRecentTokens: number): CutPointResult;
+/**
+ * Returns an error message when a summarization response cannot safely be persisted.
+ * A length stop contains partial text and must not become a session checkpoint.
+ */
+export declare function getSummarizationFailure(response: AssistantMessage, label: string): string | undefined;
 /**
  * Shared choke point for every compaction/branch-summary summarization call. Wraps the
  * single LLM call in {@link retryAssistantCall} so transient stream drops (e.g.

package/dist/core/compaction/compaction.js

@@ -353,76 +353,89 @@ export function findCutPoint(entries, startIndex, endIndex, keepRecentTokens) {
 // ============================================================================
 // Summarization
 // ============================================================================
-const SUMMARIZATION_PROMPT = `The messages above are a conversation to summarize. Create a structured context checkpoint summary that another LLM will use to continue the work.
-
-Use this EXACT format:
-
-## Goal
-[What is the user trying to accomplish? Can be multiple items if the session covers different tasks.]
-
-## Constraints & Preferences
-- [Any constraints, preferences, or requirements mentioned by user]
-- [Or "(none)" if none were mentioned]
-
-## Progress
-### Done
-- [x] [Completed tasks/changes]
-
-### In Progress
-- [ ] [Current work]
-
-### Blocked
-- [Issues preventing progress, if any]
-
-## Key Decisions
-- **[Decision]**: [Brief rationale]
-
-## Next Steps
-1. [Ordered list of what should happen next]
-
-## Critical Context
-- [Any data, examples, or references needed to continue]
-- [Or "(none)" if not applicable]
-
+const SUMMARIZATION_PROMPT = `The messages above are a conversation to summarize. Create a structured context checkpoint summary that another LLM will use to continue the work.
+
+Use this EXACT format:
+
+## Goal
+[What is the user trying to accomplish? Can be multiple items if the session covers different tasks.]
+
+## Constraints & Preferences
+- [Any constraints, preferences, or requirements mentioned by user]
+- [Or "(none)" if none were mentioned]
+
+## Progress
+### Done
+- [x] [Completed tasks/changes]
+
+### In Progress
+- [ ] [Current work]
+
+### Blocked
+- [Issues preventing progress, if any]
+
+## Key Decisions
+- **[Decision]**: [Brief rationale]
+
+## Next Steps
+1. [Ordered list of what should happen next]
+
+## Critical Context
+- [Any data, examples, or references needed to continue]
+- [Or "(none)" if not applicable]
+
 Keep each section concise. Preserve exact file paths, function names, and error messages.`;
-const UPDATE_SUMMARIZATION_PROMPT = `The messages above are NEW conversation messages to incorporate into the existing summary provided in <previous-summary> tags.
-
-Update the existing structured summary with new information. RULES:
-- PRESERVE all existing information from the previous summary
-- ADD new progress, decisions, and context from the new messages
-- UPDATE the Progress section: move items from "In Progress" to "Done" when completed
-- UPDATE "Next Steps" based on what was accomplished
-- PRESERVE exact file paths, function names, and error messages
-- If something is no longer relevant, you may remove it
-
-Use this EXACT format:
-
-## Goal
-[Preserve existing goals, add new ones if the task expanded]
-
-## Constraints & Preferences
-- [Preserve existing, add new ones discovered]
-
-## Progress
-### Done
-- [x] [Include previously done items AND newly completed items]
-
-### In Progress
-- [ ] [Current work - update based on progress]
-
-### Blocked
-- [Current blockers - remove if resolved]
-
-## Key Decisions
-- **[Decision]**: [Brief rationale] (preserve all previous, add new)
-
-## Next Steps
-1. [Update based on current state]
-
-## Critical Context
-- [Preserve important context, add new if needed]
-
+const UPDATE_SUMMARIZATION_PROMPT = `The messages above are NEW conversation messages to incorporate into the existing summary provided in <previous-summary> tags.
+
+Update the existing structured summary with new information. RULES:
+- PRESERVE all existing information from the previous summary
+- ADD new progress, decisions, and context from the new messages
+- UPDATE the Progress section: move items from "In Progress" to "Done" when completed
+- UPDATE "Next Steps" based on what was accomplished
+- PRESERVE exact file paths, function names, and error messages
+- If something is no longer relevant, you may remove it
+
+Use this EXACT format:
+
+## Goal
+[Preserve existing goals, add new ones if the task expanded]
+
+## Constraints & Preferences
+- [Preserve existing, add new ones discovered]
+
+## Progress
+### Done
+- [x] [Include previously done items AND newly completed items]
+
+### In Progress
+- [ ] [Current work - update based on progress]
+
+### Blocked
+- [Current blockers - remove if resolved]
+
+## Key Decisions
+- **[Decision]**: [Brief rationale] (preserve all previous, add new)
+
+## Next Steps
+1. [Update based on current state]
+
+## Critical Context
+- [Preserve important context, add new if needed]
+
 Keep each section concise. Preserve exact file paths, function names, and error messages.`;
+/**
+ * Returns an error message when a summarization response cannot safely be persisted.
+ * A length stop contains partial text and must not become a session checkpoint.
+ */
+export function getSummarizationFailure(response, label) {
+    if (response.stopReason === "error") {
+        return `${label} failed: ${response.errorMessage || "Unknown error"}`;
+    }
+    if (response.stopReason === "length") {
+        return `${label} failed: generation hit the token cap and the summary is incomplete`;
+    }
+    return undefined;
+}
 function createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel) {
     const options = { maxTokens, signal, apiKey, headers, env };
     if (model.reasoning && thinkingLevel && thinkingLevel !== "off") {
@@ -483,8 +496,9 @@ export async function generateSummaryWithUsage(currentMessages, model, reserveTo
     ];
     const completionOptions = createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel);
     const response = await completeSummarization(model, { systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages }, completionOptions, streamFn, retry, callbacks);
-    if (response.stopReason === "error") {
-        throw new Error(`Summarization failed: ${response.errorMessage || "Unknown error"}`);
+    const failure = getSummarizationFailure(response, "Summarization");
+    if (failure) {
+        throw new Error(failure);
     }
     const textContent = contentText(response.content);
     return { text: textContent, usage: response.usage };
@@ -559,19 +573,19 @@ export function prepareCompaction(pathEntries, settings) {
 // ============================================================================
 // Main compaction function
 // ============================================================================
-const TURN_PREFIX_SUMMARIZATION_PROMPT = `This is the PREFIX of a turn that was too large to keep. The SUFFIX (recent work) is retained.
-
-Summarize the prefix to provide context for the retained suffix:
-
-## Original Request
-[What did the user ask for in this turn?]
-
-## Early Progress
-- [Key decisions and work done in the prefix]
-
-## Context for Suffix
-- [Information needed to understand the retained recent work]
-
+const TURN_PREFIX_SUMMARIZATION_PROMPT = `This is the PREFIX of a turn that was too large to keep. The SUFFIX (recent work) is retained.
+
+Summarize the prefix to provide context for the retained suffix:
+
+## Original Request
+[What did the user ask for in this turn?]
+
+## Early Progress
+- [Key decisions and work done in the prefix]
+
+## Context for Suffix
+- [Information needed to understand the retained recent work]
+
 Be concise. Focus on what's needed to understand the kept suffix.`;
 /**
  * Generate summaries for compaction using prepared data.
@@ -634,8 +648,9 @@ async function generateTurnPrefixSummary(messages, model, reserveTokens, apiKey,
         },
     ];
     const response = await completeSummarization(model, { systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages }, createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel), streamFn, retry, callbacks);
-    if (response.stopReason === "error") {
-        throw new Error(`Turn prefix summarization failed: ${response.errorMessage || "Unknown error"}`);
+    const failure = getSummarizationFailure(response, "Turn prefix summarization");
+    if (failure) {
+        throw new Error(failure);
     }
     return {
         text: contentText(response.content),

package/dist/core/compaction/utils.js

@@ -136,6 +136,6 @@ export function serializeConversation(messages) {
 // ============================================================================
 // Summarization System Prompt
 // ============================================================================
-export const SUMMARIZATION_SYSTEM_PROMPT = `You are a context summarization assistant. Your task is to read a conversation between a user and an AI assistant, then produce a structured summary following the exact format specified.
-
+export const SUMMARIZATION_SYSTEM_PROMPT = `You are a context summarization assistant. Your task is to read a conversation between a user and an AI assistant, then produce a structured summary following the exact format specified.
+
 Do NOT continue the conversation. Do NOT respond to any questions in the conversation. ONLY output the structured summary.`;