@codebolt/agent 5.0.7 → 5.0.8

package/dist/processor-pieces/postToolCallProcessors/conversationCompactorModifier.js

@@ -2,39 +2,846 @@
 /**
  * Conversation Compactor Modifier
  * Post-Tool Call processor for compacting conversation history to reduce context size
+ *
+ * Inspired by gemini-cli's ChatCompressionService approach:
+ * - Truncates large tool outputs
+ * - Summarizes old conversation history using LLM
+ * - Preserves recent context for continuity
  */
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.ConversationCompactorModifier = void 0;
+exports.ConversationCompactorModifier = exports.CompressionStatus = void 0;
 const basePostToolCallProcessor_1 = require("../base/basePostToolCallProcessor");
+const codeboltjs_1 = __importDefault(require("@codebolt/codeboltjs"));
+// ============================================================================
+// Constants (from gemini-cli)
+// ============================================================================
+/** Default threshold for compression - compress when exceeding 50% of model limit */
+const DEFAULT_COMPRESSION_TOKEN_THRESHOLD = 0.5;
+/** Fraction of recent history to preserve - keep last 30% */
+const DEFAULT_COMPRESSION_PRESERVE_THRESHOLD = 0.3;
+/** Maximum token budget for all tool outputs combined */
+const DEFAULT_TOOL_RESPONSE_TOKEN_BUDGET = 50000;
+/** Lines to keep when truncating tool outputs */
+const DEFAULT_TRUNCATE_LINES = 30;
+/** Default model token limit */
+const DEFAULT_MODEL_TOKEN_LIMIT = 128000;
+// ============================================================================
+// Model Token Limits Lookup
+// ============================================================================
+/**
+ * Known model token limits
+ * Based on common LLM provider specifications
+ */
+const MODEL_TOKEN_LIMITS = {
+    // OpenAI models
+    'gpt-4': 8192,
+    'gpt-4-32k': 32768,
+    'gpt-4-turbo': 128000,
+    'gpt-4-turbo-preview': 128000,
+    'gpt-4o': 128000,
+    'gpt-4o-mini': 128000,
+    'gpt-3.5-turbo': 16385,
+    'gpt-3.5-turbo-16k': 16385,
+    // Anthropic models
+    'claude-3-opus': 200000,
+    'claude-3-sonnet': 200000,
+    'claude-3-haiku': 200000,
+    'claude-3-5-sonnet': 200000,
+    'claude-2': 100000,
+    'claude-2.1': 200000,
+    // Google models
+    'gemini-pro': 32768,
+    'gemini-1.5-pro': 1048576,
+    'gemini-1.5-flash': 1048576,
+    // Mistral models
+    'mistral-large': 32768,
+    'mistral-medium': 32768,
+    'mistral-small': 32768,
+};
+/**
+ * Gets the token limit for a model
+ * Falls back to default if model not found
+ */
+function getModelTokenLimit(modelName) {
+    if (!modelName)
+        return DEFAULT_MODEL_TOKEN_LIMIT;
+    // Try exact match first
+    if (MODEL_TOKEN_LIMITS[modelName]) {
+        return MODEL_TOKEN_LIMITS[modelName];
+    }
+    // Try partial match (model names often have version suffixes)
+    const lowerModel = modelName.toLowerCase();
+    for (const [key, limit] of Object.entries(MODEL_TOKEN_LIMITS)) {
+        if (lowerModel.includes(key.toLowerCase()) || key.toLowerCase().includes(lowerModel)) {
+            return limit;
+        }
+    }
+    return DEFAULT_MODEL_TOKEN_LIMIT;
+}
+// ============================================================================
+// Types and Interfaces
+// ============================================================================
+/**
+ * Compression Status Enum - tracks the result of compression attempts
+ */
+var CompressionStatus;
+(function (CompressionStatus) {
+    /** Compression was successful */
+    CompressionStatus["COMPRESSED"] = "COMPRESSED";
+    /** Compression not needed - under threshold */
+    CompressionStatus["NOOP"] = "NOOP";
+    /** Compression failed - summary larger than original */
+    CompressionStatus["FAILED_INFLATED_TOKEN_COUNT"] = "FAILED_INFLATED_TOKEN_COUNT";
+    /** Compression failed - token counting error */
+    CompressionStatus["FAILED_TOKEN_COUNT_ERROR"] = "FAILED_TOKEN_COUNT_ERROR";
+    /** Compression failed - LLM summarization error */
+    CompressionStatus["FAILED_SUMMARIZATION_ERROR"] = "FAILED_SUMMARIZATION_ERROR";
+    /** Tool output truncation was applied (without full compression) */
+    CompressionStatus["TOOL_OUTPUT_TRUNCATED"] = "TOOL_OUTPUT_TRUNCATED";
+})(CompressionStatus || (exports.CompressionStatus = CompressionStatus = {}));
+// ============================================================================
+// Main Class
+// ============================================================================
 class ConversationCompactorModifier extends basePostToolCallProcessor_1.BasePostToolCallProcessor {
     constructor(options = {}) {
+        var _a, _b, _c, _d, _e, _f, _g, _h, _j, _k, _l;
         super();
+        this.failedCompressionAttempts = 0;
+        this.hasFailedCompression = false;
         this.options = {
-            maxMessages: 50,
-            preserveSystemMessages: true,
-            preserveLastUserMessages: 5,
-            preserveLastToolResults: 10,
-            compactStrategy: 'smart',
-            ...options
+            compressionTokenThreshold: (_a = options.compressionTokenThreshold) !== null && _a !== void 0 ? _a : DEFAULT_COMPRESSION_TOKEN_THRESHOLD,
+            preserveThreshold: (_b = options.preserveThreshold) !== null && _b !== void 0 ? _b : DEFAULT_COMPRESSION_PRESERVE_THRESHOLD,
+            toolResponseTokenBudget: (_c = options.toolResponseTokenBudget) !== null && _c !== void 0 ? _c : DEFAULT_TOOL_RESPONSE_TOKEN_BUDGET,
+            truncateLines: (_d = options.truncateLines) !== null && _d !== void 0 ? _d : DEFAULT_TRUNCATE_LINES,
+            compactStrategy: (_e = options.compactStrategy) !== null && _e !== void 0 ? _e : 'smart',
+            preserveSystemMessages: (_f = options.preserveSystemMessages) !== null && _f !== void 0 ? _f : true,
+            minPreservedUserMessages: (_g = options.minPreservedUserMessages) !== null && _g !== void 0 ? _g : 3,
+            enableVerificationPass: (_h = options.enableVerificationPass) !== null && _h !== void 0 ? _h : false,
+            modelTokenLimit: (_j = options.modelTokenLimit) !== null && _j !== void 0 ? _j : DEFAULT_MODEL_TOKEN_LIMIT,
+            llmRole: (_k = options.llmRole) !== null && _k !== void 0 ? _k : 'summarizer',
+            enableLogging: (_l = options.enableLogging) !== null && _l !== void 0 ? _l : false
         };
     }
-    async modify(input) {
-        // For now, pass through the nextPrompt unchanged since we need to understand
-        // the ProcessedMessage structure better for compaction
-        // TODO: Implement conversation compaction logic once we understand the message structure
-        // Add metadata to track that this processor was applied
-        const modifiedNextPrompt = {
-            ...input.nextPrompt,
+    // ========================================================================
+    // Token Counting Utilities
+    // ========================================================================
+    /**
+     * Estimates token count for a string
+     * Uses the ~4 characters per token approximation
+     */
+    estimateTokens(text) {
+        if (!text)
+            return 0;
+        return Math.ceil(text.length / 4);
+    }
+    /**
+     * Counts tokens for an array of messages
+     */
+    countMessageTokens(messages) {
+        if (!messages || !Array.isArray(messages))
+            return 0;
+        return messages.reduce((total, msg) => {
+            if (!msg)
+                return total;
+            const content = !msg.content ? '' :
+                (typeof msg.content === 'string' ? msg.content : JSON.stringify(msg.content));
+            // Add overhead for role and structure (~4 tokens per message)
+            return total + this.estimateTokens(content) + 4;
+        }, 0);
+    }
+    /**
+     * Gets content length from a message (handles string and array content)
+     */
+    getMessageContentLength(message) {
+        if (!message || message.content === undefined || message.content === null) {
+            return 0;
+        }
+        if (typeof message.content === 'string') {
+            return message.content.length;
+        }
+        return JSON.stringify(message.content).length;
+    }
+    /**
+     * Extracts text content from a message
+     */
+    getMessageContent(message) {
+        if (!message || message.content === undefined || message.content === null) {
+            return '';
+        }
+        if (typeof message.content === 'string') {
+            return message.content;
+        }
+        // Extract text from content blocks
+        if (!Array.isArray(message.content)) {
+            return JSON.stringify(message.content);
+        }
+        return message.content
+            .filter(block => block && block.text)
+            .map(block => block.text)
+            .join('\n');
+    }
+    // ========================================================================
+    // Tool Output Truncation (Phase 1)
+    // ========================================================================
+    /**
+     * Checks if a message is a function/tool response
+     */
+    isToolResponseMessage(message) {
+        if (!message)
+            return false;
+        if (message.role === 'tool')
+            return true;
+        if (message.tool_call_id)
+            return true;
+        // Check for function response in content
+        if (typeof message.content === 'object' && message.content !== null) {
+            if (Array.isArray(message.content)) {
+                return message.content.some(block => block && (block.type === 'tool_result' ||
+                    block.type === 'function_response'));
+            }
+        }
+        return false;
+    }
+    /**
+     * Truncates content to a specified number of lines
+     * Keeps first half and last half of allowed lines for context
+     */
+    truncateToLines(content, maxLines) {
+        const lines = content.split('\n');
+        if (lines.length <= maxLines) {
+            return content;
+        }
+        // Keep first half and last half of allowed lines for context
+        const halfLines = Math.floor(maxLines / 2);
+        const firstPart = lines.slice(0, halfLines);
+        const lastPart = lines.slice(-halfLines);
+        const omittedCount = lines.length - maxLines;
+        return [
+            ...firstPart,
+            `\n... [${omittedCount} lines truncated] ...\n`,
+            ...lastPart
+        ].join('\n');
+    }
+    /**
+     * Truncates large tool output messages to reduce token count
+     * Iterates backwards (prioritizing recent messages) following gemini-cli approach
+     */
+    truncateLargeToolOutputs(messages) {
+        const result = [...messages];
+        let truncated = false;
+        let tokensSaved = 0;
+        // Calculate current tool response tokens and identify tool message indices
+        let toolResponseTokens = 0;
+        const toolMessageInfo = [];
+        for (let i = 0; i < result.length; i++) {
+            const msg = result[i];
+            if (msg && (msg.role === 'tool' || this.isToolResponseMessage(msg))) {
+                const content = this.getMessageContent(msg);
+                const tokens = this.estimateTokens(content);
+                toolMessageInfo.push({ index: i, tokens });
+                toolResponseTokens += tokens;
+            }
+        }
+        // If under budget, no truncation needed
+        if (toolResponseTokens <= this.options.toolResponseTokenBudget) {
+            return { messages: result, truncated: false, tokensSaved: 0 };
+        }
+        // Iterate from oldest to newest tool messages for truncation
+        // (We truncate older tool outputs first to preserve recent context)
+        for (const info of toolMessageInfo) {
+            if (toolResponseTokens <= this.options.toolResponseTokenBudget) {
+                break;
+            }
+            const msg = result[info.index];
+            if (!msg)
+                continue;
+            const originalContent = this.getMessageContent(msg);
+            const originalTokens = this.estimateTokens(originalContent);
+            // Only truncate if content is substantial
+            if (originalTokens > 100) {
+                const truncatedContent = this.truncateToLines(originalContent, this.options.truncateLines);
+                const newTokens = this.estimateTokens(truncatedContent);
+                if (newTokens < originalTokens) {
+                    // Update the message with truncated content
+                    const updatedMsg = {
+                        role: msg.role,
+                        content: truncatedContent
+                    };
+                    if (msg.tool_call_id)
+                        updatedMsg.tool_call_id = msg.tool_call_id;
+                    if (msg.tool_calls)
+                        updatedMsg.tool_calls = msg.tool_calls;
+                    if (msg.name)
+                        updatedMsg.name = msg.name;
+                    result[info.index] = updatedMsg;
+                    const saved = originalTokens - newTokens;
+                    tokensSaved += saved;
+                    toolResponseTokens -= saved;
+                    truncated = true;
+                    if (this.options.enableLogging) {
+                        console.log(`[ConversationCompactor] Truncated tool output at index ${info.index}, saved ${saved} tokens`);
+                    }
+                }
+            }
+        }
+        return { messages: result, truncated, tokensSaved };
+    }
+    // ========================================================================
+    // History Split Point Calculation (Phase 2)
+    // ========================================================================
+    /**
+     * Adjusts split index to the start of the next turn (user message)
+     * Ensures we don't split in the middle of a conversation turn
+     */
+    adjustToTurnBoundary(messages, index) {
+        while (index < messages.length) {
+            const msg = messages[index];
+            if (!msg) {
+                index++;
+                continue;
+            }
+            // Stop at a user message that isn't a tool response
+            if (msg.role === 'user' && !this.isToolResponseMessage(msg)) {
+                break;
+            }
+            // Also stop at assistant message without pending tool calls
+            if (msg.role === 'assistant' && (!msg.tool_calls || msg.tool_calls.length === 0)) {
+                break;
+            }
+            index++;
+        }
+        return index;
+    }
+    /**
+     * Maps filtered array index back to original messages array
+     */
+    mapToOriginalIndex(original, filtered, filteredIndex) {
+        if (filteredIndex >= filtered.length)
+            return original.length;
+        const targetMessage = filtered[filteredIndex];
+        return original.findIndex(msg => msg === targetMessage);
+    }
+    /**
+     * Finds the index to split history for compression
+     * Returns the index after which messages should be preserved
+     */
+    findCompressionSplitIndex(messages) {
+        // Filter out system messages if preserving them
+        const processableMessages = this.options.preserveSystemMessages
+            ? messages.filter(m => m.role !== 'system')
+            : messages;
+        if (processableMessages.length <= this.options.minPreservedUserMessages * 2) {
+            // Too few messages to compress
+            return 0;
+        }
+        // Calculate content lengths for fraction-based splitting
+        const contentLengths = processableMessages.map(msg => this.getMessageContentLength(msg));
+        const totalLength = contentLengths.reduce((sum, len) => sum + len, 0);
+        // Find index after (1 - preserveThreshold) of content
+        // e.g., if preserveThreshold is 0.3, find index after 70% of content
+        const targetLength = totalLength * (1 - this.options.preserveThreshold);
+        let accumulatedLength = 0;
+        let splitIndex = 0;
+        for (let i = 0; i < contentLengths.length; i++) {
+            const len = contentLengths[i];
+            if (len !== undefined) {
+                accumulatedLength += len;
+            }
+            if (accumulatedLength >= targetLength) {
+                splitIndex = i;
+                break;
+            }
+        }
+        // Ensure we don't split too early
+        if (splitIndex === 0) {
+            splitIndex = Math.floor(processableMessages.length * (1 - this.options.preserveThreshold));
+        }
+        // Adjust to turn boundary - find the next user message
+        splitIndex = this.adjustToTurnBoundary(processableMessages, splitIndex);
+        // Map back to original messages array if we filtered system messages
+        if (this.options.preserveSystemMessages) {
+            return this.mapToOriginalIndex(messages, processableMessages, splitIndex);
+        }
+        return splitIndex;
+    }
+    // ========================================================================
+    // LLM Summarization (Phase 3 - 'summarize' strategy only)
+    // ========================================================================
+    /**
+     * Generates the compression prompt with state_snapshot XML format
+     * Based on gemini-cli's approach
+     */
+    getCompressionPrompt(historyToCompress) {
+        const historyText = historyToCompress.map((msg, i) => {
+            const content = this.getMessageContent(msg);
+            const truncated = content.length > 500 ? content.substring(0, 500) + '...' : content;
+            return `[${i + 1}] ${msg.role}: ${truncated}`;
+        }).join('\n\n');
+        return `You are a conversation compression assistant. Analyze the following conversation history and create a concise state snapshot that preserves all critical information.
+
+CONVERSATION HISTORY TO COMPRESS:
+${historyText}
+
+Generate a state snapshot in the following XML format. Be extremely concise but preserve all critical technical details:
+
+<state_snapshot>
+    <overall_goal>
+        [Describe the user's high-level objective in 1-2 sentences]
+    </overall_goal>
+
+    <active_constraints>
+        [List any user-specified rules, preferences, or constraints that must be maintained]
+    </active_constraints>
+
+    <key_knowledge>
+        [List crucial technical facts discovered: file structures, dependencies, API details, error patterns, etc.]
+    </key_knowledge>
+
+    <artifact_trail>
+        [Track evolution of critical files/symbols that were modified or are important]
+    </artifact_trail>
+
+    <file_system_state>
+        [Describe current relevant file system state: created files, modified files, directory structure if relevant]
+    </file_system_state>
+
+    <recent_actions>
+        [Fact-based summary of tool calls and their outcomes - what was done, not how]
+    </recent_actions>
+
+    <task_state>
+        [Current state of the task: what's complete, what's pending, and the IMMEDIATE next step]
+    </task_state>
+</state_snapshot>
+
+IMPORTANT:
+- Be extremely concise - aim for maximum information density
+- Preserve technical accuracy - exact file paths, function names, error messages
+- Focus on facts that affect future decisions
+- Do not include conversational pleasantries or redundant information`;
+    }
+    /**
+     * Generates a state snapshot summary using LLM
+     */
+    async generateStateSummary(historyToCompress) {
+        var _a;
+        try {
+            const prompt = this.getCompressionPrompt(historyToCompress);
+            const response = await codeboltjs_1.default.llm.inference({
+                messages: [
+                    { role: 'system', content: 'You are a precise conversation compression assistant.' },
+                    { role: 'user', content: prompt }
+                ],
+            });
+            const summary = ((_a = response.completion) === null || _a === void 0 ? void 0 : _a.content) || '';
+            if (this.options.enableLogging) {
+                console.log(`[ConversationCompactor] Generated summary (${summary.length} chars)`);
+            }
+            return summary;
+        }
+        catch (error) {
+            console.error('[ConversationCompactor] Error generating summary:', error);
+            throw error;
+        }
+    }
+    /**
+     * Optional verification pass - asks LLM to verify and correct the summary
+     */
+    async verifySummary(summary, originalHistory) {
+        var _a;
+        if (!this.options.enableVerificationPass) {
+            return summary;
+        }
+        try {
+            const verificationPrompt = `Review this conversation state snapshot for accuracy and completeness.
+
+STATE SNAPSHOT:
+${summary}
+
+ORIGINAL MESSAGES COUNT: ${originalHistory.length}
+
+If the snapshot is accurate and complete, respond with just the original snapshot.
+If there are inaccuracies or missing critical information, provide a corrected version.
+Keep the same XML format.`;
+            const response = await codeboltjs_1.default.llm.inference({
+                messages: [
+                    { role: 'system', content: 'You are verifying a conversation compression for accuracy.' },
+                    { role: 'user', content: verificationPrompt }
+                ]
+            });
+            return ((_a = response.completion) === null || _a === void 0 ? void 0 : _a.content) || summary;
+        }
+        catch (error) {
+            console.warn('[ConversationCompactor] Verification pass failed, using original summary');
+            return summary;
+        }
+    }
+    // ========================================================================
+    // Compression Strategies
+    // ========================================================================
+    /**
+     * Simple compression strategy - just removes old messages
+     */
+    async compressSimple(messages) {
+        const splitIndex = this.findCompressionSplitIndex(messages);
+        if (splitIndex === 0) {
+            return {
+                compressedMessages: messages,
+                metadata: {
+                    strategy: 'simple',
+                    messagesCompressed: 0,
+                    messagesPreserved: messages.length
+                }
+            };
+        }
+        // Preserve system messages
+        const systemMessages = this.options.preserveSystemMessages
+            ? messages.filter(m => m.role === 'system')
+            : [];
+        // Keep messages after split point
+        const preservedMessages = messages.slice(splitIndex);
+        // Combine: system messages first, then placeholder, then preserved history
+        const compressedMessages = [
+            ...systemMessages.filter(sm => !preservedMessages.includes(sm)),
+            {
+                role: 'user',
+                content: `[Previous conversation context has been compressed. ${splitIndex} messages were removed to manage context length.]`
+            },
+            ...preservedMessages
+        ];
+        return {
+            compressedMessages,
+            metadata: {
+                strategy: 'simple',
+                messagesCompressed: splitIndex,
+                messagesPreserved: preservedMessages.length,
+                toolOutputsTruncated: false
+            }
+        };
+    }
+    /**
+     * Smart compression strategy - truncates tool outputs + intelligent message removal
+     */
+    async compressSmart(messages) {
+        // Phase 1: Truncate large tool outputs
+        const { messages: truncatedMessages, truncated } = this.truncateLargeToolOutputs(messages);
+        // Check if truncation alone was sufficient
+        const tokensAfterTruncation = this.countMessageTokens(truncatedMessages);
+        const threshold = this.options.modelTokenLimit * this.options.compressionTokenThreshold;
+        if (tokensAfterTruncation < threshold) {
+            return {
+                compressedMessages: truncatedMessages,
+                metadata: {
+                    strategy: 'smart',
+                    toolOutputsTruncated: truncated,
+                    messagesCompressed: 0,
+                    messagesPreserved: truncatedMessages.length
+                }
+            };
+        }
+        // Phase 2: Remove old messages
+        const splitIndex = this.findCompressionSplitIndex(truncatedMessages);
+        if (splitIndex === 0) {
+            return {
+                compressedMessages: truncatedMessages,
+                metadata: {
+                    strategy: 'smart',
+                    toolOutputsTruncated: truncated,
+                    messagesCompressed: 0,
+                    messagesPreserved: truncatedMessages.length
+                }
+            };
+        }
+        const systemMessages = this.options.preserveSystemMessages
+            ? truncatedMessages.filter(m => m.role === 'system')
+            : [];
+        const preservedMessages = truncatedMessages.slice(splitIndex);
+        const compressedMessages = [
+            ...systemMessages.filter(sm => !preservedMessages.includes(sm)),
+            {
+                role: 'user',
+                content: `[Conversation compressed. ${splitIndex} older messages summarized. Tool outputs may have been truncated.]`
+            },
+            ...preservedMessages
+        ];
+        return {
+            compressedMessages,
             metadata: {
-                ...input.nextPrompt.metadata,
-                compactedBy: 'ConversationCompactorModifier',
-                compactedAt: new Date().toISOString(),
+                strategy: 'smart',
+                toolOutputsTruncated: truncated,
+                messagesCompressed: splitIndex,
+                messagesPreserved: preservedMessages.length
+            }
+        };
+    }
+    /**
+     * Summarize compression strategy - uses LLM to create state snapshot
+     */
+    async compressSummarize(messages) {
+        // Phase 1: Truncate large tool outputs
+        const { messages: truncatedMessages, truncated } = this.truncateLargeToolOutputs(messages);
+        // Phase 2: Find split point
+        const splitIndex = this.findCompressionSplitIndex(truncatedMessages);
+        if (splitIndex === 0) {
+            // Nothing to compress
+            return {
+                compressedMessages: truncatedMessages,
+                metadata: {
+                    strategy: 'summarize',
+                    toolOutputsTruncated: truncated,
+                    messagesCompressed: 0,
+                    messagesPreserved: truncatedMessages.length
+                }
+            };
+        }
+        // Phase 3: Generate summary of messages to compress
+        const systemMessages = this.options.preserveSystemMessages
+            ? truncatedMessages.filter(m => m.role === 'system')
+            : [];
+        const historyToCompress = truncatedMessages
+            .slice(0, splitIndex)
+            .filter(m => m.role !== 'system');
+        const preservedMessages = truncatedMessages.slice(splitIndex);
+        try {
+            // Generate LLM summary
+            let summary = await this.generateStateSummary(historyToCompress);
+            // Optional verification pass
+            summary = await this.verifySummary(summary, historyToCompress);
+            if (!summary || summary.trim().length === 0) {
+                throw new Error('Empty summary generated');
+            }
+            // Build compressed message array
+            const compressedMessages = [
+                ...systemMessages.filter(sm => !preservedMessages.includes(sm)),
+                {
+                    role: 'user',
+                    content: summary
+                },
+                {
+                    role: 'assistant',
+                    content: 'Understood. I have processed the conversation state snapshot and am ready to continue from where we left off.'
+                },
+                ...preservedMessages
+            ];
+            return {
+                compressedMessages,
+                metadata: {
+                    strategy: 'summarize',
+                    toolOutputsTruncated: truncated,
+                    messagesCompressed: historyToCompress.length,
+                    messagesPreserved: preservedMessages.length
+                }
+            };
+        }
+        catch (error) {
+            // Fall back to smart compression if summarization fails
+            console.warn('[ConversationCompactor] Summarization failed, falling back to smart compression:', error);
+            this.failedCompressionAttempts++;
+            return this.compressSmart(messages);
+        }
+    }
+    // ========================================================================
+    // Main Entry Point
+    // ========================================================================
+    async modify(input) {
+        var _a, _b, _c;
+        const { nextPrompt, rawLLMResponseMessage, tokenLimit } = input;
+        try {
+            // Safety check: ensure messages array exists
+            if (!((_a = nextPrompt === null || nextPrompt === void 0 ? void 0 : nextPrompt.message) === null || _a === void 0 ? void 0 : _a.messages) || !Array.isArray(nextPrompt.message.messages)) {
+                if (this.options.enableLogging) {
+                    console.warn('[ConversationCompactor] No messages array found, skipping compression');
+                }
+                return {
+                    nextPrompt,
+                    shouldExit: false
+                };
+            }
+            const messages = nextPrompt.message.messages;
+            // Get model token limit: prefer tokenLimit from LLM response, then lookup by model name, then fallback to config
+            const modelName = rawLLMResponseMessage === null || rawLLMResponseMessage === void 0 ? void 0 : rawLLMResponseMessage.model;
+            const modelTokenLimit = tokenLimit !== null && tokenLimit !== void 0 ? tokenLimit : (modelName ? getModelTokenLimit(modelName) : this.options.modelTokenLimit);
+            // Use actual token count from LLM response if available, otherwise estimate
+            const currentTokens = (_c = (_b = rawLLMResponseMessage === null || rawLLMResponseMessage === void 0 ? void 0 : rawLLMResponseMessage.usage) === null || _b === void 0 ? void 0 : _b.prompt_tokens) !== null && _c !== void 0 ? _c : this.countMessageTokens(messages);
+            const threshold = modelTokenLimit * this.options.compressionTokenThreshold;
+            if (this.options.enableLogging) {
+                console.log(`[ConversationCompactor] Current tokens: ${currentTokens}, Threshold: ${threshold}`);
+            }
+            // Check if compression is needed
+            if (currentTokens < threshold) {
+                return {
+                    nextPrompt: {
+                        ...nextPrompt,
+                        metadata: {
+                            ...nextPrompt.metadata,
+                            compression: {
+                                status: CompressionStatus.NOOP,
+                                originalTokenCount: currentTokens,
+                                newTokenCount: currentTokens,
+                                messagesCompressed: 0,
+                                messagesPreserved: messages.length,
+                                toolOutputsTruncated: false,
+                                failedAttempts: this.failedCompressionAttempts,
+                                timestamp: new Date().toISOString(),
+                                strategy: this.options.compactStrategy
+                            }
+                        }
+                    },
+                    shouldExit: false
+                };
+            }
+            // Skip if we've had too many failed attempts
+            if (this.hasFailedCompression && this.failedCompressionAttempts >= 3) {
+                if (this.options.enableLogging) {
+                    console.warn('[ConversationCompactor] Skipping - too many failed attempts');
+                }
+                return {
+                    nextPrompt: {
+                        ...nextPrompt,
+                        metadata: {
+                            ...nextPrompt.metadata,
+                            compressionSkipped: true,
+                            compressionSkipReason: 'Too many failed attempts'
+                        }
+                    },
+                    shouldExit: false
+                };
+            }
+            // Apply compression based on strategy
+            let result;
+            switch (this.options.compactStrategy) {
+                case 'simple':
+                    result = await this.compressSimple(messages);
+                    break;
+                case 'smart':
+                    result = await this.compressSmart(messages);
+                    break;
+                case 'summarize':
+                    result = await this.compressSummarize(messages);
+                    break;
+                default:
+                    result = await this.compressSmart(messages);
+            }
+            const newTokens = this.countMessageTokens(result.compressedMessages);
+            // Verify compression was effective
+            if (newTokens >= currentTokens) {
+                this.failedCompressionAttempts++;
+                this.hasFailedCompression = true;
+                if (this.options.enableLogging) {
+                    console.warn(`[ConversationCompactor] Compression inflated token count: ${currentTokens} -> ${newTokens}`);
+                }
+                return {
+                    nextPrompt: {
+                        ...nextPrompt,
+                        metadata: {
+                            ...nextPrompt.metadata,
+                            compression: {
+                                status: CompressionStatus.FAILED_INFLATED_TOKEN_COUNT,
+                                originalTokenCount: currentTokens,
+                                newTokenCount: newTokens,
+                                messagesCompressed: 0,
+                                messagesPreserved: messages.length,
+                                toolOutputsTruncated: false,
+                                failedAttempts: this.failedCompressionAttempts,
+                                timestamp: new Date().toISOString(),
+                                strategy: this.options.compactStrategy
+                            }
+                        }
+                    },
+                    shouldExit: false
+                };
+            }
+            // Build final compression metadata
+            const compressionMetadata = {
+                status: CompressionStatus.COMPRESSED,
+                originalTokenCount: currentTokens,
+                newTokenCount: newTokens,
+                messagesCompressed: result.metadata.messagesCompressed || 0,
+                messagesPreserved: result.metadata.messagesPreserved || 0,
+                toolOutputsTruncated: result.metadata.toolOutputsTruncated || false,
+                failedAttempts: this.failedCompressionAttempts,
+                timestamp: new Date().toISOString(),
                 strategy: this.options.compactStrategy
+            };
+            if (this.options.enableLogging) {
+                console.log(`[ConversationCompactor] Compression complete: ${currentTokens} -> ${newTokens} tokens (saved ${currentTokens - newTokens})`);
             }
+            return {
+                nextPrompt: {
+                    message: {
+                        ...nextPrompt.message,
+                        messages: result.compressedMessages
+                    },
+                    metadata: {
+                        ...nextPrompt.metadata,
+                        compression: compressionMetadata,
+                        compactedBy: 'ConversationCompactorModifier',
+                        compactedAt: compressionMetadata.timestamp
+                    }
+                },
+                shouldExit: false
+            };
+        }
+        catch (error) {
+            console.error('[ConversationCompactor] Error during compression:', error);
+            this.failedCompressionAttempts++;
+            this.hasFailedCompression = true;
+            return {
+                nextPrompt: {
+                    ...nextPrompt,
+                    metadata: {
+                        ...nextPrompt.metadata,
+                        compression: {
+                            status: CompressionStatus.FAILED_SUMMARIZATION_ERROR,
+                            originalTokenCount: this.countMessageTokens(nextPrompt.message.messages),
+                            newTokenCount: this.countMessageTokens(nextPrompt.message.messages),
+                            messagesCompressed: 0,
+                            messagesPreserved: nextPrompt.message.messages.length,
+                            toolOutputsTruncated: false,
+                            failedAttempts: this.failedCompressionAttempts,
+                            timestamp: new Date().toISOString(),
+                            strategy: this.options.compactStrategy
+                        },
+                        compressionError: error instanceof Error ? error.message : 'Unknown error'
+                    }
+                },
+                shouldExit: false
+            };
+        }
+    }
+    // ========================================================================
+    // Public Utility Methods
+    // ========================================================================
+    /**
+     * Resets the compression state (useful for new conversations)
+     */
+    resetCompressionState() {
+        this.failedCompressionAttempts = 0;
+        this.hasFailedCompression = false;
+    }
+    /**
+     * Gets the current compression statistics
+     */
+    getCompressionStats() {
+        return {
+            failedAttempts: this.failedCompressionAttempts,
+            hasFailedCompression: this.hasFailedCompression,
+            options: this.options
         };
+    }
+    /**
+     * Forces compression regardless of threshold (for testing or manual trigger)
+     */
+    async forceCompress(messages) {
+        const result = await (this.options.compactStrategy === 'summarize'
+            ? this.compressSummarize(messages)
+            : this.compressSmart(messages));
         return {
-            nextPrompt: modifiedNextPrompt,
-            shouldExit: false
+            messages: result.compressedMessages,
+            metadata: result.metadata
         };
     }
 }