@tomsun28/pizza 0.0.9 → 0.1.1

package/dist/core/compaction/compaction.js

@@ -0,0 +1,295 @@
+/**
+ * Context compaction for long sessions.
+ *
+ * Pure functions for compaction logic. The session manager handles I/O,
+ * and after compaction the session is reloaded.
+ */
+export const DEFAULT_COMPACTION_SETTINGS = {
+    enabled: true,
+    reserveTokens: 16384,
+    keepRecentTokens: 20000,
+};
+// ============================================================================
+// Token calculation
+// ============================================================================
+/**
+ * Calculate total context tokens from usage.
+ * Uses the native totalTokens field when available, falls back to computing from components.
+ */
+export function calculateContextTokens(usage) {
+    return usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite;
+}
+/**
+ * Get usage from an assistant message if available.
+ * Skips aborted and error messages as they don't have valid usage data.
+ */
+function getAssistantUsage(msg) {
+    if (msg.role === "assistant" && "usage" in msg) {
+        const assistantMsg = msg;
+        if (assistantMsg.stopReason !== "aborted" && assistantMsg.stopReason !== "error" && assistantMsg.usage) {
+            return assistantMsg.usage;
+        }
+    }
+    return undefined;
+}
+/**
+ * Find the last non-aborted assistant message usage from session entries.
+ */
+export function getLastAssistantUsage(entries) {
+    for (let i = entries.length - 1; i >= 0; i--) {
+        const entry = entries[i];
+        if (entry.type === "message") {
+            const usage = getAssistantUsage(entry.message);
+            if (usage)
+                return usage;
+        }
+    }
+    return undefined;
+}
+function getLastAssistantUsageInfo(messages) {
+    for (let i = messages.length - 1; i >= 0; i--) {
+        const usage = getAssistantUsage(messages[i]);
+        if (usage)
+            return { usage, index: i };
+    }
+    return undefined;
+}
+/**
+ * Estimate context tokens from messages, using the last assistant usage when available.
+ * If there are messages after the last usage, estimate their tokens with estimateTokens.
+ */
+export function estimateContextTokens(messages) {
+    const usageInfo = getLastAssistantUsageInfo(messages);
+    if (!usageInfo) {
+        let estimated = 0;
+        for (const message of messages) {
+            estimated += estimateTokens(message);
+        }
+        return {
+            tokens: estimated,
+            usageTokens: 0,
+            trailingTokens: estimated,
+            lastUsageIndex: null,
+        };
+    }
+    const usageTokens = calculateContextTokens(usageInfo.usage);
+    let trailingTokens = 0;
+    for (let i = usageInfo.index + 1; i < messages.length; i++) {
+        trailingTokens += estimateTokens(messages[i]);
+    }
+    return {
+        tokens: usageTokens + trailingTokens,
+        usageTokens,
+        trailingTokens,
+        lastUsageIndex: usageInfo.index,
+    };
+}
+/**
+ * Check if compaction should trigger based on context usage.
+ */
+export function shouldCompact(contextTokens, contextWindow, settings) {
+    if (!settings.enabled)
+        return false;
+    return contextTokens > contextWindow - settings.reserveTokens;
+}
+// ============================================================================
+// Cut point detection
+// ============================================================================
+/**
+ * Estimate token count for a message using chars/4 heuristic.
+ * This is conservative (overestimates tokens).
+ */
+export function estimateTokens(message) {
+    let chars = 0;
+    switch (message.role) {
+        case "user": {
+            const content = message.content;
+            if (typeof content === "string") {
+                chars = content.length;
+            }
+            else if (Array.isArray(content)) {
+                for (const block of content) {
+                    if (block.type === "text" && block.text) {
+                        chars += block.text.length;
+                    }
+                }
+            }
+            return Math.ceil(chars / 4);
+        }
+        case "assistant": {
+            const assistant = message;
+            for (const block of assistant.content) {
+                if (block.type === "text") {
+                    chars += block.text.length;
+                }
+                else if (block.type === "thinking") {
+                    chars += block.thinking.length;
+                }
+                else if (block.type === "toolCall") {
+                    chars += block.name.length + JSON.stringify(block.arguments).length;
+                }
+            }
+            return Math.ceil(chars / 4);
+        }
+        case "custom":
+        case "toolResult": {
+            if (typeof message.content === "string") {
+                chars = message.content.length;
+            }
+            else {
+                for (const block of message.content) {
+                    if (block.type === "text" && block.text) {
+                        chars += block.text.length;
+                    }
+                    if (block.type === "image") {
+                        chars += 4800; // Estimate images as 4000 chars, or 1200 tokens
+                    }
+                }
+            }
+            return Math.ceil(chars / 4);
+        }
+        case "bashExecution": {
+            chars = message.command.length + message.output.length;
+            return Math.ceil(chars / 4);
+        }
+        case "branchSummary":
+        case "compactionSummary": {
+            chars = message.summary.length;
+            return Math.ceil(chars / 4);
+        }
+    }
+    return 0;
+}
+/**
+ * Find valid cut points: indices of user, assistant, custom, or bashExecution messages.
+ * Never cut at tool results (they must follow their tool call).
+ * When we cut at an assistant message with tool calls, its tool results follow it
+ * and will be kept.
+ * BashExecutionMessage is treated like a user message (user-initiated context).
+ */
+function findValidCutPoints(entries, startIndex, endIndex) {
+    const cutPoints = [];
+    for (let i = startIndex; i < endIndex; i++) {
+        const entry = entries[i];
+        switch (entry.type) {
+            case "message": {
+                const role = entry.message.role;
+                switch (role) {
+                    case "bashExecution":
+                    case "custom":
+                    case "branchSummary":
+                    case "compactionSummary":
+                    case "user":
+                    case "assistant":
+                        cutPoints.push(i);
+                        break;
+                    case "toolResult":
+                        break;
+                }
+                break;
+            }
+            case "thinking_level_change":
+            case "model_change":
+            case "compaction":
+            case "branch_summary":
+            case "custom":
+            case "custom_message":
+            case "label":
+            case "session_info":
+                break;
+        }
+        // branch_summary and custom_message are user-role messages, valid cut points
+        if (entry.type === "branch_summary" || entry.type === "custom_message") {
+            cutPoints.push(i);
+        }
+    }
+    return cutPoints;
+}
+/**
+ * Find the user message (or bashExecution) that starts the turn containing the given entry index.
+ * Returns -1 if no turn start found before the index.
+ * BashExecutionMessage is treated like a user message for turn boundaries.
+ */
+export function findTurnStartIndex(entries, entryIndex, startIndex) {
+    for (let i = entryIndex; i >= startIndex; i--) {
+        const entry = entries[i];
+        // branch_summary and custom_message are user-role messages, can start a turn
+        if (entry.type === "branch_summary" || entry.type === "custom_message") {
+            return i;
+        }
+        if (entry.type === "message") {
+            const role = entry.message.role;
+            if (role === "user" || role === "bashExecution") {
+                return i;
+            }
+        }
+    }
+    return -1;
+}
+/**
+ * Find the cut point in session entries that keeps approximately `keepRecentTokens`.
+ *
+ * Algorithm: Walk backwards from newest, accumulating estimated message sizes.
+ * Stop when we've accumulated >= keepRecentTokens. Cut at that point.
+ *
+ * Can cut at user OR assistant messages (never tool results). When cutting at an
+ * assistant message with tool calls, its tool results come after and will be kept.
+ *
+ * Returns CutPointResult with:
+ * - firstKeptEntryIndex: the entry index to start keeping from
+ * - turnStartIndex: if cutting mid-turn, the user message that started that turn
+ * - isSplitTurn: whether we're cutting in the middle of a turn
+ *
+ * Only considers entries between `startIndex` and `endIndex` (exclusive).
+ */
+export function findCutPoint(entries, startIndex, endIndex, keepRecentTokens) {
+    const cutPoints = findValidCutPoints(entries, startIndex, endIndex);
+    if (cutPoints.length === 0) {
+        return { firstKeptEntryIndex: startIndex, turnStartIndex: -1, isSplitTurn: false };
+    }
+    // Walk backwards from newest, accumulating estimated message sizes
+    let accumulatedTokens = 0;
+    let cutIndex = cutPoints[0]; // Default: keep from first message (not header)
+    for (let i = endIndex - 1; i >= startIndex; i--) {
+        const entry = entries[i];
+        if (entry.type !== "message")
+            continue;
+        // Estimate this message's size
+        const messageTokens = estimateTokens(entry.message);
+        accumulatedTokens += messageTokens;
+        // Check if we've exceeded the budget
+        if (accumulatedTokens >= keepRecentTokens) {
+            // Find the closest valid cut point at or after this entry
+            for (let c = 0; c < cutPoints.length; c++) {
+                if (cutPoints[c] >= i) {
+                    cutIndex = cutPoints[c];
+                    break;
+                }
+            }
+            break;
+        }
+    }
+    // Scan backwards from cutIndex to include any non-message entries (bash, settings, etc.)
+    while (cutIndex > startIndex) {
+        const prevEntry = entries[cutIndex - 1];
+        // Stop at session header or compaction boundaries
+        if (prevEntry.type === "compaction") {
+            break;
+        }
+        if (prevEntry.type === "message") {
+            // Stop if we hit any message
+            break;
+        }
+        // Include this non-message entry (bash, settings change, etc.)
+        cutIndex--;
+    }
+    // Determine if this is a split turn
+    const cutEntry = entries[cutIndex];
+    const isUserMessage = cutEntry.type === "message" && cutEntry.message.role === "user";
+    const turnStartIndex = isUserMessage ? -1 : findTurnStartIndex(entries, cutIndex, startIndex);
+    return {
+        firstKeptEntryIndex: cutIndex,
+        turnStartIndex,
+        isSplitTurn: !isUserMessage && turnStartIndex !== -1,
+    };
+}

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

@@ -0,0 +1,7 @@
+/**
+ * Compaction and summarization utilities.
+ */
+export * from "./branch-summarization.js";
+export * from "./compaction.js";
+export * from "./compaction-engine.js";
+export * from "./utils.js";

package/dist/core/compaction/index.js

@@ -0,0 +1,7 @@
+/**
+ * Compaction and summarization utilities.
+ */
+export * from "./branch-summarization.js";
+export * from "./compaction.js";
+export * from "./compaction-engine.js";
+export * from "./utils.js";

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

@@ -0,0 +1,37 @@
+/**
+ * Shared utilities for compaction and branch summarization.
+ */
+import type { AgentMessage } from "../agent/types.js";
+import type { Message } from "@earendil-works/pi-ai/compat";
+export interface FileOperations {
+    read: Set<string>;
+    written: Set<string>;
+    edited: Set<string>;
+}
+export declare function createFileOps(): FileOperations;
+/**
+ * Extract file operations from tool calls in an assistant message.
+ */
+export declare function extractFileOpsFromMessage(message: AgentMessage, fileOps: FileOperations): void;
+/**
+ * Compute final file lists from file operations.
+ * Returns readFiles (files only read, not modified) and modifiedFiles.
+ */
+export declare function computeFileLists(fileOps: FileOperations): {
+    readFiles: string[];
+    modifiedFiles: string[];
+};
+/**
+ * Format file operations as XML tags for summary.
+ */
+export declare function formatFileOperations(readFiles: string[], modifiedFiles: string[]): string;
+/**
+ * Serialize LLM messages to text for summarization.
+ * This prevents the model from treating it as a conversation to continue.
+ * Call convertToLlm() first to handle custom message types.
+ *
+ * Tool results are truncated to keep the summarization request within
+ * reasonable token budgets. Full content is not needed for summarization.
+ */
+export declare function serializeConversation(messages: Message[]): string;
+export declare const SUMMARIZATION_SYSTEM_PROMPT = "You are a context summarization assistant. Your task is to read a conversation between a user and an AI coding assistant, then produce a structured summary following the exact format specified.\n\nDo NOT continue the conversation. Do NOT respond to any questions in the conversation. ONLY output the structured summary.";

package/dist/core/compaction/utils.js

@@ -0,0 +1,152 @@
+/**
+ * Shared utilities for compaction and branch summarization.
+ */
+export function createFileOps() {
+    return {
+        read: new Set(),
+        written: new Set(),
+        edited: new Set(),
+    };
+}
+/**
+ * Extract file operations from tool calls in an assistant message.
+ */
+export function extractFileOpsFromMessage(message, fileOps) {
+    if (message.role !== "assistant")
+        return;
+    if (!("content" in message) || !Array.isArray(message.content))
+        return;
+    for (const block of message.content) {
+        if (typeof block !== "object" || block === null)
+            continue;
+        if (!("type" in block) || block.type !== "toolCall")
+            continue;
+        if (!("arguments" in block) || !("name" in block))
+            continue;
+        const args = block.arguments;
+        if (!args)
+            continue;
+        const path = typeof args.path === "string" ? args.path : undefined;
+        if (!path)
+            continue;
+        switch (block.name) {
+            case "read":
+                fileOps.read.add(path);
+                break;
+            case "write":
+                fileOps.written.add(path);
+                break;
+            case "edit":
+                fileOps.edited.add(path);
+                break;
+        }
+    }
+}
+/**
+ * Compute final file lists from file operations.
+ * Returns readFiles (files only read, not modified) and modifiedFiles.
+ */
+export function computeFileLists(fileOps) {
+    const modified = new Set([...fileOps.edited, ...fileOps.written]);
+    const readOnly = [...fileOps.read].filter((f) => !modified.has(f)).sort();
+    const modifiedFiles = [...modified].sort();
+    return { readFiles: readOnly, modifiedFiles };
+}
+/**
+ * Format file operations as XML tags for summary.
+ */
+export function formatFileOperations(readFiles, modifiedFiles) {
+    const sections = [];
+    if (readFiles.length > 0) {
+        sections.push(`<read-files>\n${readFiles.join("\n")}\n</read-files>`);
+    }
+    if (modifiedFiles.length > 0) {
+        sections.push(`<modified-files>\n${modifiedFiles.join("\n")}\n</modified-files>`);
+    }
+    if (sections.length === 0)
+        return "";
+    return `\n\n${sections.join("\n\n")}`;
+}
+// ============================================================================
+// Message Serialization
+// ============================================================================
+/** Maximum characters for a tool result in serialized summaries. */
+const TOOL_RESULT_MAX_CHARS = 2000;
+/**
+ * Truncate text to a maximum character length for summarization.
+ * Keeps the beginning and appends a truncation marker.
+ */
+function truncateForSummary(text, maxChars) {
+    if (text.length <= maxChars)
+        return text;
+    const truncatedChars = text.length - maxChars;
+    return `${text.slice(0, maxChars)}\n\n[... ${truncatedChars} more characters truncated]`;
+}
+/**
+ * Serialize LLM messages to text for summarization.
+ * This prevents the model from treating it as a conversation to continue.
+ * Call convertToLlm() first to handle custom message types.
+ *
+ * Tool results are truncated to keep the summarization request within
+ * reasonable token budgets. Full content is not needed for summarization.
+ */
+export function serializeConversation(messages) {
+    const parts = [];
+    for (const msg of messages) {
+        if (msg.role === "user") {
+            const content = typeof msg.content === "string"
+                ? msg.content
+                : msg.content
+                    .filter((c) => c.type === "text")
+                    .map((c) => c.text)
+                    .join("");
+            if (content)
+                parts.push(`[User]: ${content}`);
+        }
+        else if (msg.role === "assistant") {
+            const textParts = [];
+            const thinkingParts = [];
+            const toolCalls = [];
+            for (const block of msg.content) {
+                if (block.type === "text") {
+                    textParts.push(block.text);
+                }
+                else if (block.type === "thinking") {
+                    thinkingParts.push(block.thinking);
+                }
+                else if (block.type === "toolCall") {
+                    const args = block.arguments;
+                    const argsStr = Object.entries(args)
+                        .map(([k, v]) => `${k}=${JSON.stringify(v)}`)
+                        .join(", ");
+                    toolCalls.push(`${block.name}(${argsStr})`);
+                }
+            }
+            if (thinkingParts.length > 0) {
+                parts.push(`[Assistant thinking]: ${thinkingParts.join("\n")}`);
+            }
+            if (textParts.length > 0) {
+                parts.push(`[Assistant]: ${textParts.join("\n")}`);
+            }
+            if (toolCalls.length > 0) {
+                parts.push(`[Assistant tool calls]: ${toolCalls.join("; ")}`);
+            }
+        }
+        else if (msg.role === "toolResult") {
+            const content = msg.content
+                .filter((c) => c.type === "text")
+                .map((c) => c.text)
+                .join("");
+            if (content) {
+                parts.push(`[Tool result]: ${truncateForSummary(content, TOOL_RESULT_MAX_CHARS)}`);
+            }
+        }
+    }
+    return parts.join("\n\n");
+}
+// ============================================================================
+// 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 coding 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.`;

package/dist/core/defaults.d.ts

@@ -0,0 +1,2 @@
+import type { ThinkingLevel } from "./agent/types.js";
+export declare const DEFAULT_THINKING_LEVEL: ThinkingLevel;

package/dist/core/defaults.js

@@ -0,0 +1 @@
+export const DEFAULT_THINKING_LEVEL = "medium";

package/dist/core/diagnostics.d.ts

@@ -0,0 +1,14 @@
+export interface ResourceCollision {
+    resourceType: "extension" | "skill" | "prompt" | "theme";
+    name: string;
+    winnerPath: string;
+    loserPath: string;
+    winnerSource?: string;
+    loserSource?: string;
+}
+export interface ResourceDiagnostic {
+    type: "warning" | "error" | "collision";
+    message: string;
+    path?: string;
+    collision?: ResourceCollision;
+}

package/dist/core/diagnostics.js

@@ -0,0 +1 @@
+export {};

package/dist/core/event-bus.d.ts

@@ -0,0 +1,8 @@
+export interface EventBus {
+    emit(channel: string, data: unknown): void;
+    on(channel: string, handler: (data: unknown) => void): () => void;
+}
+export interface EventBusController extends EventBus {
+    clear(): void;
+}
+export declare function createEventBus(): EventBusController;

package/dist/core/event-bus.js

@@ -0,0 +1,24 @@
+import { EventEmitter } from "node:events";
+export function createEventBus() {
+    const emitter = new EventEmitter();
+    return {
+        emit: (channel, data) => {
+            emitter.emit(channel, data);
+        },
+        on: (channel, handler) => {
+            const safeHandler = async (data) => {
+                try {
+                    await handler(data);
+                }
+                catch (err) {
+                    console.error(`Event handler error (${channel}):`, err);
+                }
+            };
+            emitter.on(channel, safeHandler);
+            return () => emitter.off(channel, safeHandler);
+        },
+        clear: () => {
+            emitter.removeAllListeners();
+        },
+    };
+}