@averyyy/pi-coding-agent 0.84.2-piclient.3 → 0.84.4-piclient.1

package/dist/core/agent-session.js

@@ -24,7 +24,7 @@ import { normalizeToolResultImages } from "../utils/tool-result-images.js";
 import { formatNoApiKeyFoundMessage, formatNoModelSelectedMessage } from "./auth-guidance.js";
 import { executeBashWithOperations } from "./bash-executor.js";
 import { calculateContextTokens, collectEntriesForBranchSummary, compact, estimateContextTokens, estimateTokens, generateBranchSummary, prepareCompaction, shouldCompact, } from "./compaction/index.js";
-import { DEFAULT_THINKING_LEVEL } from "./defaults.js";
+import { DEFAULT_THINKING_LEVEL, THINKING_LEVEL_OPTIONS } from "./defaults.js";
 import { exportSessionToHtml } from "./export-html/index.js";
 import { createToolHtmlRenderer } from "./export-html/tool-renderer.js";
 import { ExtensionRunner, wrapRegisteredTools, } from "./extensions/index.js";
@@ -72,8 +72,6 @@ function estimateMessagesTokens(messages) {
 // ============================================================================
 // Constants
 // ============================================================================
-/** Standard thinking levels */
-const THINKING_LEVELS = ["off", "minimal", "low", "medium", "high"];
 const INTRA_TURN_COMPACTION_MARKER_TYPE = "pi:intra-turn-compaction";
 const INTRA_TURN_COMPACTION_MARKER_TEXT = "Continue from the compaction summary above. The previous tool loop was compacted before the next provider request; use the summary's Operational State, file lists, failures, last command, and next steps.";
 const VALIDATION_HINT_MESSAGE_TYPE = "pi:validation-hint";
@@ -172,6 +170,8 @@ 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;
@@ -474,6 +474,11 @@ export class AgentSession {
             followUp: [...this._followUpMessages],
         });
     }
+    async _emitSessionCompactFailed(event) {
+        if (this._extensionRunner.hasHandlers("session_compact_failed")) {
+            await this._extensionRunner.emit({ type: "session_compact_failed", ...event });
+        }
+    }
     _getIdleWaitPromise() {
         if (!this._idleWaitPromise) {
             this._idleWaitPromise = new Promise((resolve) => {
@@ -564,6 +569,14 @@ 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();
@@ -992,6 +1005,7 @@ export class AgentSession {
         finally {
             this._systemPromptOverride = undefined;
             this._flushPendingBashMessages();
+            this._flushPendingCustomMessages();
             await this._emitAgentSettled();
         }
     }
@@ -1150,8 +1164,9 @@ 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());
@@ -1368,8 +1383,9 @@ 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
      *
@@ -1401,11 +1417,34 @@ 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);
         }
     }
     /**
@@ -1503,10 +1542,11 @@ export class AgentSession {
     }
     /**
      * Set model directly.
-     * Validates that auth is configured, saves to session and settings.
+     * Validates that auth is configured and saves to the session transcript.
+     * Persists to global defaults only when options.persist is true.
      * @throws Error if no auth is configured for the model
      */
-    async setModel(model) {
+    async setModel(model, options = {}) {
         if (!(await this._modelRuntime.checkAuth(model.provider))) {
             throw new Error(`No API key for ${model.provider}/${model.id}`);
         }
@@ -1514,27 +1554,46 @@ export class AgentSession {
             await this.abort();
         }
         const previousModel = this.model;
-        const thinkingLevel = this._getThinkingLevelForModelSwitch();
+        const thinkingLevel = this._getThinkingLevelForModelSwitch(model);
         this.agent.state.model = model;
         this.sessionManager.appendModelChange(model.provider, model.id);
-        this.settingsManager.setDefaultModelAndProvider(model.provider, model.id);
-        // Re-clamp thinking level for new model's capabilities
+        if (options.persist) {
+            this.settingsManager.setDefaultModelAndProvider(model.provider, model.id);
+            this._addPersistedDefaultToNonEmptyScope(model);
+        }
+        // Apply thinking level for the new model.
+        // Per-model thinking level overrides take priority over the global default.
+        // Model persistence does not implicitly rewrite the global thinking default.
         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.
      * @param direction - "forward" (default) or "backward"
      * @returns The new model info, or undefined if only one model available
      */
-    async cycleModel(direction = "forward") {
+    async cycleModel(direction = "forward", options = {}) {
         if (this._scopedModels.length > 0) {
-            return this._cycleScopedModel(direction);
+            return this._cycleScopedModel(direction, options);
         }
-        return this._cycleAvailableModel(direction);
+        return this._cycleAvailableModel(direction, options);
     }
-    async _cycleScopedModel(direction) {
+    async _cycleScopedModel(direction, options) {
         const availableIds = new Set(this._modelRuntime.getAvailableSnapshot().map((model) => `${model.provider}\0${model.id}`));
         const scopedModels = this._scopedModels.filter((scoped) => availableIds.has(`${scoped.model.provider}\0${scoped.model.id}`));
         if (scopedModels.length <= 1)
@@ -1546,20 +1605,24 @@ export class AgentSession {
         const len = scopedModels.length;
         const nextIndex = direction === "forward" ? (currentIndex + 1) % len : (currentIndex - 1 + len) % len;
         const next = scopedModels[nextIndex];
-        const thinkingLevel = this._getThinkingLevelForModelSwitch(next.thinkingLevel);
+        const thinkingLevel = this._getThinkingLevelForModelSwitch(next.model, next.thinkingLevel);
         // Apply model
         this.agent.state.model = next.model;
         this.sessionManager.appendModelChange(next.model.provider, next.model.id);
-        this.settingsManager.setDefaultModelAndProvider(next.model.provider, next.model.id);
-        // Apply thinking level.
-        // - Explicit scoped model thinking level overrides current session level
-        // - Undefined scoped model thinking level inherits the current session preference
+        if (options.persist) {
+            this.settingsManager.setDefaultModelAndProvider(next.model.provider, next.model.id);
+            this._addPersistedDefaultToNonEmptyScope(next.model);
+        }
+        // Apply thinking level for the new model.
+        // - Explicit scoped model thinking level overrides defaults
+        // - Per-model thinking level overrides take priority over the global default
         // setThinkingLevel clamps to model capabilities.
+        // Model persistence does not implicitly rewrite the global thinking default.
         this.setThinkingLevel(thinkingLevel);
         await this._emitModelSelect(next.model, currentModel, "cycle");
         return { model: next.model, thinkingLevel: this.thinkingLevel, isScoped: true };
     }
-    async _cycleAvailableModel(direction) {
+    async _cycleAvailableModel(direction, options) {
         const availableModels = this._modelRuntime.getAvailableSnapshot();
         if (availableModels.length <= 1)
             return undefined;
@@ -1570,11 +1633,15 @@ export class AgentSession {
         const len = availableModels.length;
         const nextIndex = direction === "forward" ? (currentIndex + 1) % len : (currentIndex - 1 + len) % len;
         const nextModel = availableModels[nextIndex];
-        const thinkingLevel = this._getThinkingLevelForModelSwitch();
+        const thinkingLevel = this._getThinkingLevelForModelSwitch(nextModel);
         this.agent.state.model = nextModel;
         this.sessionManager.appendModelChange(nextModel.provider, nextModel.id);
-        this.settingsManager.setDefaultModelAndProvider(nextModel.provider, nextModel.id);
-        // Re-clamp thinking level for new model's capabilities
+        if (options.persist) {
+            this.settingsManager.setDefaultModelAndProvider(nextModel.provider, nextModel.id);
+            this._addPersistedDefaultToNonEmptyScope(nextModel);
+        }
+        // Apply thinking level for the new model.
+        // Model persistence does not implicitly rewrite the global thinking default.
         this.setThinkingLevel(thinkingLevel);
         await this._emitModelSelect(nextModel, currentModel, "cycle");
         return { model: nextModel, thinkingLevel: this.thinkingLevel, isScoped: false };
@@ -1585,20 +1652,21 @@ export class AgentSession {
     /**
      * Set thinking level.
      * Clamps to model capabilities based on available thinking levels.
-     * Saves to session and settings only if the level actually changes.
+     * Saves the clamped level to the session transcript only if the level actually changes.
+     * Persists the requested level to global defaults only when options.persist is true.
      */
-    setThinkingLevel(level) {
+    setThinkingLevel(level, options = {}) {
         const availableLevels = this.getAvailableThinkingLevels();
         const effectiveLevel = availableLevels.includes(level) ? level : this._clampThinkingLevel(level, availableLevels);
         // Only persist if actually changing
         const previousLevel = this.agent.state.thinkingLevel;
         const isChanging = effectiveLevel !== previousLevel;
         this.agent.state.thinkingLevel = effectiveLevel;
+        if (options.persist) {
+            this.settingsManager.setDefaultThinkingLevel(level);
+        }
         if (isChanging) {
             this.sessionManager.appendThinkingLevelChange(effectiveLevel);
-            if (this.supportsThinking() || effectiveLevel !== "off") {
-                this.settingsManager.setDefaultThinkingLevel(effectiveLevel);
-            }
             this._emit({ type: "thinking_level_changed", level: effectiveLevel });
             void this._extensionRunner.emit({
                 type: "thinking_level_select",
@@ -1611,14 +1679,14 @@ export class AgentSession {
      * Cycle to next thinking level.
      * @returns New level, or undefined if model doesn't support thinking
      */
-    cycleThinkingLevel() {
+    cycleThinkingLevel(options = {}) {
         if (!this.supportsThinking())
             return undefined;
         const levels = this.getAvailableThinkingLevels();
         const currentIndex = levels.indexOf(this.thinkingLevel);
         const nextIndex = (currentIndex + 1) % levels.length;
         const nextLevel = levels[nextIndex];
-        this.setThinkingLevel(nextLevel);
+        this.setThinkingLevel(nextLevel, options);
         return nextLevel;
     }
     /**
@@ -1627,7 +1695,7 @@ export class AgentSession {
      */
     getAvailableThinkingLevels() {
         if (!this.model)
-            return THINKING_LEVELS;
+            return [...THINKING_LEVEL_OPTIONS];
         return getSupportedThinkingLevels(this.model);
     }
     /**
@@ -1636,14 +1704,17 @@ export class AgentSession {
     supportsThinking() {
         return !!this.model?.reasoning;
     }
-    _getThinkingLevelForModelSwitch(explicitLevel) {
+    _getThinkingLevelForModelSwitch(targetModel, explicitLevel) {
         if (explicitLevel !== undefined) {
             return explicitLevel;
         }
-        if (!this.supportsThinking()) {
-            return this.settingsManager.getDefaultThinkingLevel() ?? DEFAULT_THINKING_LEVEL;
+        if (targetModel) {
+            const perModel = this.settingsManager.getModelThinkingLevel(targetModel.provider, targetModel.id);
+            if (perModel !== undefined) {
+                return perModel;
+            }
         }
-        return this.thinkingLevel;
+        return this.settingsManager.getDefaultThinkingLevel() ?? this.thinkingLevel ?? DEFAULT_THINKING_LEVEL;
     }
     _clampThinkingLevel(level, _availableLevels) {
         return this.model ? clampThinkingLevel(this.model, level) : "off";
@@ -1907,20 +1978,24 @@ export class AgentSession {
         // independent of the configured context size or any context-clamped provider request limit.
         // A successful response over the configured window should compact but must not retry: the
         // assistant answer already completed and agent.continue() cannot continue from an assistant.
+        const contextOverflow = sameModel && isContextOverflow(assistantMessage, contextWindow);
         const recoverableLength = sameModel && isRecoverableLength(assistantMessage, this.model?.maxTokens ?? 0);
-        if (sameModel && (isContextOverflow(assistantMessage, contextWindow) || recoverableLength)) {
+        if (contextOverflow || recoverableLength) {
             const willRetry = assistantMessage.stopReason !== "stop";
             if (!willRetry) {
                 return await this._runAutoCompaction("overflow", false);
             }
             if (this._overflowRecoveryAttempted) {
+                const errorMessage = contextOverflow
+                    ? "Context overflow recovery failed after one compact-and-retry attempt. Try reducing context or switching to a larger-context model."
+                    : "Truncated response recovery failed after one compact-and-retry attempt.";
                 this._emit({
                     type: "compaction_end",
                     reason: "overflow",
                     result: undefined,
                     aborted: false,
                     willRetry: false,
-                    errorMessage: "Context overflow recovery failed after one compact-and-retry attempt. Try reducing context or switching to a larger-context model.",
+                    errorMessage,
                 });
                 return false;
             }
@@ -1950,16 +2025,16 @@ export class AgentSession {
         if (assistantMessage.stopReason === "error" || directContextTokens === 0) {
             const messages = this.agent.state.messages;
             const estimate = estimateContextTokens(messages);
-            if (estimate.lastUsageIndex === null)
-                return false; // No usage data at all
-            // Verify the usage source is post-compaction. Kept pre-compaction messages
-            // have stale usage reflecting the old (larger) context and would falsely
-            // trigger compaction right after one just finished.
-            const usageMsg = messages[estimate.lastUsageIndex];
-            if (compactionTimestamp > 0 &&
-                usageMsg.role === "assistant" &&
-                usageMsg.timestamp <= compactionTimestamp) {
-                return false;
+            // When a valid usage entry exists, verify it is post-compaction. Kept
+            // pre-compaction messages have stale usage reflecting the old (larger)
+            // context and would falsely trigger compaction right after one finished.
+            if (estimate.lastUsageIndex !== null) {
+                const usageMsg = messages[estimate.lastUsageIndex];
+                if (compactionTimestamp > 0 &&
+                    usageMsg.role === "assistant" &&
+                    usageMsg.timestamp <= compactionTimestamp) {
+                    return false;
+                }
             }
             contextTokens = estimate.tokens;
         }
@@ -1977,6 +2052,7 @@ export class AgentSession {
     async _runAutoCompaction(reason, willRetry, preparationOverride, preparationOptions) {
         const settings = this.settingsManager.getCompactionSettings();
         let started = false;
+        let fromExtension = false;
         try {
             if (!this.model) {
                 return false;
@@ -2016,7 +2092,6 @@ export class AgentSession {
             this._autoCompactionAbortController = new AbortController();
             started = true;
             let extensionCompaction;
-            let fromExtension = false;
             if (this._extensionRunner.hasHandlers("session_before_compact")) {
                 const extensionResult = (await this._extensionRunner.emit({
                     type: "session_before_compact",
@@ -2035,6 +2110,12 @@ export class AgentSession {
                         aborted: true,
                         willRetry: false,
                     });
+                    await this._emitSessionCompactFailed({
+                        reason,
+                        aborted: true,
+                        willRetry: false,
+                        fromExtension: false,
+                    });
                     return false;
                 }
                 if (extensionResult?.compaction) {
@@ -2072,6 +2153,12 @@ export class AgentSession {
                     aborted: true,
                     willRetry: false,
                 });
+                await this._emitSessionCompactFailed({
+                    reason,
+                    aborted: true,
+                    willRetry: false,
+                    fromExtension,
+                });
                 return false;
             }
             this.sessionManager.appendCompaction(summary, firstKeptEntryId, tokensBefore, details, fromExtension, usage);
@@ -2128,6 +2215,15 @@ export class AgentSession {
                         ? `Context overflow recovery failed: ${errorMessage}`
                         : `Auto-compaction failed: ${errorMessage}`,
                 });
+                await this._emitSessionCompactFailed({
+                    reason,
+                    errorMessage: reason === "overflow"
+                        ? `Context overflow recovery failed: ${errorMessage}`
+                        : `Auto-compaction failed: ${errorMessage}`,
+                    aborted: false,
+                    willRetry: false,
+                    fromExtension,
+                });
             }
             return false;
         }
@@ -2537,11 +2633,9 @@ export class AgentSession {
     // =========================================================================
     // Auto-Retry
     // =========================================================================
-    _isNonRetryableProviderLimitError(errorMessage) {
-        return /GoUsageLimitError|FreeUsageLimitError|Monthly usage limit reached|available balance|insufficient.?balance|balance.?insufficient|insufficient_quota|out of budget|quota exceeded|billing|payment required|402/i.test(errorMessage);
-    }
     /**
-     * Check if an error is retryable (overloaded, rate limit, server errors).
+     * Check if an error is retryable. Provider error text is retryable by default,
+     * except usage/quota/balance limits and session or context failures.
      * Context overflow errors are NOT retryable (handled by compaction instead).
      */
     _isRetryableError(message) {
@@ -2553,8 +2647,6 @@ export class AgentSession {
         const piServerPhase = getPiServerFailurePhase(message);
         if (piServerPhase && piServerPhase !== "provider_stream")
             return false;
-        if (message.errorMessage && this._isNonRetryableProviderLimitError(message.errorMessage))
-            return false;
         return isRetryableAssistantError(message);
     }
     /**