@townco/agent 0.1.50 → 0.1.52

package/dist/utils/context-size-calculator.d.ts

@@ -0,0 +1,29 @@
+/**
+ * Context size calculation utilities
+ * Calculates full context size by counting ALL tokens in messages
+ */
+import type { SessionMessage } from "../acp-server/session-storage.js";
+export interface ContextSize {
+    systemPromptTokens: number;
+    userMessagesTokens: number;
+    assistantMessagesTokens: number;
+    toolInputTokens: number;
+    toolResultsTokens: number;
+    totalEstimated: number;
+    llmReportedInputTokens?: number | undefined;
+}
+/**
+ * Calculate the full context size by counting ALL tokens in the provided messages.
+ * This should be called every time a new context entry is created.
+ *
+ * How LLM-reported tokens work:
+ * - The LLM API returns `usage_metadata.input_tokens` which is the ACTUAL token
+ *   count for EVERYTHING sent to the API: system prompt, tool declarations,
+ *   all messages, and all tool results
+ * - We pass this as `llmReportedTokens` for comparison with our estimate
+ * - This helps us validate the accuracy of our tokenizer estimates
+ * - Tool declarations are NOT counted separately in our estimate since they're
+ *   included in the LLM-reported value
+ */
+export declare function calculateContextSize(messages: SessionMessage[], // Resolved messages from context entry
+systemPrompt?: string, llmReportedTokens?: number): ContextSize;

package/dist/utils/context-size-calculator.js

@@ -0,0 +1,78 @@
+/**
+ * Context size calculation utilities
+ * Calculates full context size by counting ALL tokens in messages
+ */
+import { countTokens, countToolResultTokens } from "./token-counter.js";
+/**
+ * Extract and count tokens from a content block based on its type
+ */
+function countContentBlock(block) {
+    if (block.type === "text") {
+        return {
+            textTokens: countTokens(block.text),
+            toolInputTokens: 0,
+            toolResultTokens: 0,
+        };
+    }
+    if (block.type === "tool_call") {
+        return {
+            textTokens: 0,
+            toolInputTokens: block.rawInput
+                ? countToolResultTokens(block.rawInput)
+                : 0,
+            toolResultTokens: block.rawOutput
+                ? countToolResultTokens(block.rawOutput)
+                : 0,
+        };
+    }
+    return { textTokens: 0, toolInputTokens: 0, toolResultTokens: 0 };
+}
+/**
+ * Calculate the full context size by counting ALL tokens in the provided messages.
+ * This should be called every time a new context entry is created.
+ *
+ * How LLM-reported tokens work:
+ * - The LLM API returns `usage_metadata.input_tokens` which is the ACTUAL token
+ *   count for EVERYTHING sent to the API: system prompt, tool declarations,
+ *   all messages, and all tool results
+ * - We pass this as `llmReportedTokens` for comparison with our estimate
+ * - This helps us validate the accuracy of our tokenizer estimates
+ * - Tool declarations are NOT counted separately in our estimate since they're
+ *   included in the LLM-reported value
+ */
+export function calculateContextSize(messages, // Resolved messages from context entry
+systemPrompt, llmReportedTokens) {
+    const systemPromptTokens = systemPrompt ? countTokens(systemPrompt) : 0;
+    let userMessagesTokens = 0;
+    let assistantMessagesTokens = 0;
+    let toolInputTokens = 0;
+    let toolResultsTokens = 0;
+    // Go through ALL messages in this context snapshot
+    for (const message of messages) {
+        for (const block of message.content) {
+            const counts = countContentBlock(block);
+            // Accumulate based on message role
+            if (message.role === "user") {
+                userMessagesTokens += counts.textTokens;
+            }
+            else if (message.role === "assistant") {
+                assistantMessagesTokens += counts.textTokens;
+                toolInputTokens += counts.toolInputTokens;
+                toolResultsTokens += counts.toolResultTokens;
+            }
+        }
+    }
+    return {
+        systemPromptTokens,
+        userMessagesTokens,
+        assistantMessagesTokens,
+        toolInputTokens,
+        toolResultsTokens,
+        totalEstimated: systemPromptTokens +
+            userMessagesTokens +
+            assistantMessagesTokens +
+            toolInputTokens +
+            toolResultsTokens,
+        llmReportedInputTokens: llmReportedTokens,
+    };
+}

package/dist/utils/index.d.ts

@@ -1,2 +1,4 @@
 export { makeSubagentsTool } from "../runner/langchain/tools/subagent.js";
+export * from "./context-size-calculator.js";
+export * from "./token-counter.js";
 export * from "./tool.js";

package/dist/utils/index.js

@@ -1,2 +1,4 @@
 export { makeSubagentsTool } from "../runner/langchain/tools/subagent.js";
+export * from "./context-size-calculator.js";
+export * from "./token-counter.js";
 export * from "./tool.js";

package/dist/utils/token-counter.d.ts

@@ -0,0 +1,19 @@
+/**
+ * Token counting utilities for tracking context size
+ * Uses Anthropic's tokenizer for rough approximation
+ */
+/**
+ * Count tokens in a string using Anthropic's tokenizer
+ * Note: This is a rough approximation for Claude 3+ models
+ * For exact counts, use the API's usage_metadata where available
+ */
+export declare function countTokens(text: string): number;
+/**
+ * Count tokens in a tool result (rawOutput)
+ * Handles various data types that might be in tool outputs
+ */
+export declare function countToolResultTokens(rawOutput: Record<string, unknown> | undefined): number;
+/**
+ * Count tokens in multiple tool results
+ */
+export declare function countMultipleToolResults(results: Array<Record<string, unknown> | undefined>): number;

package/dist/utils/token-counter.js

@@ -0,0 +1,44 @@
+/**
+ * Token counting utilities for tracking context size
+ * Uses Anthropic's tokenizer for rough approximation
+ */
+import { countTokens as anthropicCountTokens } from "@anthropic-ai/tokenizer";
+/**
+ * Count tokens in a string using Anthropic's tokenizer
+ * Note: This is a rough approximation for Claude 3+ models
+ * For exact counts, use the API's usage_metadata where available
+ */
+export function countTokens(text) {
+    try {
+        return anthropicCountTokens(text);
+    }
+    catch (error) {
+        // Fallback to rough estimation if tokenizer fails
+        // Approximate: ~4 characters per token
+        return Math.ceil(text.length / 4);
+    }
+}
+/**
+ * Count tokens in a tool result (rawOutput)
+ * Handles various data types that might be in tool outputs
+ */
+export function countToolResultTokens(rawOutput) {
+    if (!rawOutput) {
+        return 0;
+    }
+    try {
+        // Convert to string representation (as it would be sent to the LLM)
+        const text = JSON.stringify(rawOutput);
+        return countTokens(text);
+    }
+    catch (error) {
+        // If JSON.stringify fails, return 0
+        return 0;
+    }
+}
+/**
+ * Count tokens in multiple tool results
+ */
+export function countMultipleToolResults(results) {
+    return results.reduce((total, result) => total + countToolResultTokens(result), 0);
+}

package/index.ts

@@ -45,7 +45,7 @@ const exampleAgent: AgentDefinition = {
     {
       type: "context_size",
       setting: {
-        threshold: 95,
+        threshold: 80,
       },
       callback: "compaction_tool",
     },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@townco/agent",
-  "version": "0.1.50",
+  "version": "0.1.52",
   "type": "module",
   "module": "index.ts",
   "files": [
@@ -55,6 +55,7 @@
     "@agentclientprotocol/sdk": "^0.5.1",
     "@anthropic-ai/sandbox-runtime": "^0.0.2",
     "@anthropic-ai/sdk": "^0.70.0",
+    "@anthropic-ai/tokenizer": "^0.0.4",
     "@electric-sql/pglite": "^0.2.15",
     "@langchain/anthropic": "1.0.1",
     "@langchain/core": "^1.0.3",
@@ -72,11 +73,11 @@
     "@opentelemetry/sdk-trace-base": "^1.28.0",
     "@opentelemetry/sdk-trace-node": "^1.28.0",
     "@opentelemetry/semantic-conventions": "^1.28.0",
-    "@townco/core": "0.0.23",
-    "@townco/gui-template": "0.1.42",
-    "@townco/tui-template": "0.1.42",
-    "@townco/tsconfig": "0.1.42",
-    "@townco/ui": "0.1.45",
+    "@townco/core": "0.0.25",
+    "@townco/gui-template": "0.1.44",
+    "@townco/tsconfig": "0.1.44",
+    "@townco/tui-template": "0.1.44",
+    "@townco/ui": "0.1.47",
     "exa-js": "^2.0.0",
     "hono": "^4.10.4",
     "langchain": "^1.0.3",

package/templates/index.ts

@@ -20,11 +20,23 @@ export interface TemplateVars {
   systemPrompt: string | null;
   hasWebSearch: boolean;
   hooks?:
-    | Array<{
-        type: "context_size";
-        setting?: { threshold: number } | undefined;
-        callback: string;
-      }>
+    | Array<
+        | {
+            type: "context_size";
+            setting?: { threshold: number } | undefined;
+            callback: string;
+          }
+        | {
+            type: "tool_response";
+            setting?:
+              | {
+                  maxContextThreshold?: number | undefined;
+                  responseTruncationThreshold?: number | undefined;
+                }
+              | undefined;
+            callback: string;
+          }
+      >
     | undefined;
 }
 
@@ -41,7 +53,7 @@ export function getTemplateVars(
     hasWebSearch: tools.some(
       (tool) => typeof tool === "string" && tool === "web_search",
     ),
-    hooks: definition.hooks,
+    hooks: definition.hooks as TemplateVars["hooks"],
   };
 }
 

package/dist/definition/tools/todo.d.ts

@@ -1,49 +0,0 @@
-import { z } from "zod";
-export declare const todoItemSchema: z.ZodObject<
-	{
-		content: z.ZodString;
-		status: z.ZodEnum<{
-			pending: "pending";
-			in_progress: "in_progress";
-			completed: "completed";
-		}>;
-		activeForm: z.ZodString;
-	},
-	z.core.$strip
->;
-export declare const todoWrite: import("langchain").DynamicStructuredTool<
-	z.ZodObject<
-		{
-			todos: z.ZodArray<
-				z.ZodObject<
-					{
-						content: z.ZodString;
-						status: z.ZodEnum<{
-							pending: "pending";
-							in_progress: "in_progress";
-							completed: "completed";
-						}>;
-						activeForm: z.ZodString;
-					},
-					z.core.$strip
-				>
-			>;
-		},
-		z.core.$strip
-	>,
-	{
-		todos: {
-			content: string;
-			status: "pending" | "in_progress" | "completed";
-			activeForm: string;
-		}[];
-	},
-	{
-		todos: {
-			content: string;
-			status: "pending" | "in_progress" | "completed";
-			activeForm: string;
-		}[];
-	},
-	string
->;

package/dist/definition/tools/todo.js

@@ -1,80 +0,0 @@
-import { tool } from "langchain";
-import { z } from "zod";
-export const todoItemSchema = z.object({
-	content: z.string().min(1),
-	status: z.enum(["pending", "in_progress", "completed"]),
-	activeForm: z.string().min(1),
-});
-export const todoWrite = tool(
-	({ todos }) => {
-		// Simple implementation that confirms the todos were written
-		return `Successfully updated todo list with ${todos.length} items`;
-	},
-	{
-		name: "todo_write",
-		description: `Use this tool to create and manage a structured task list for your current coding session. This helps you track progress, organize complex tasks, and demonstrate thoroughness to the user.
-It also helps the user understand the progress of the task and overall progress of their requests.
-
-## When to Use This Tool
-Use this tool proactively in these scenarios:
-
-1. Complex multi-step tasks - When a task requires 3 or more distinct steps or actions
-2. Non-trivial and complex tasks - Tasks that require careful planning or multiple operations
-3. User explicitly requests todo list - When the user directly asks you to use the todo list
-4. User provides multiple tasks - When users provide a list of things to be done (numbered or comma-separated)
-5. After receiving new instructions - Immediately capture user requirements as todos
-6. When you start working on a task - Mark it as in_progress BEFORE beginning work. Ideally you should only have one todo as in_progress at a time
-7. After completing a task - Mark it as completed and add any new follow-up tasks discovered during implementation
-
-## When NOT to Use This Tool
-
-Skip using this tool when:
-1. There is only a single, straightforward task
-2. The task is trivial and tracking it provides no organizational benefit
-3. The task can be completed in less than 3 trivial steps
-4. The task is purely conversational or informational
-
-NOTE that you should not use this tool if there is only one trivial task to do. In this case you are better off just doing the task directly.
-
-## Task States and Management
-
-1. **Task States**: Use these states to track progress:
-   - pending: Task not yet started
-   - in_progress: Currently working on (limit to ONE task at a time)
-   - completed: Task finished successfully
-
-   **IMPORTANT**: Task descriptions must have two forms:
-   - content: The imperative form describing what needs to be done (e.g., "Run tests", "Build the project")
-   - activeForm: The present continuous form shown during execution (e.g., "Running tests", "Building the project")
-
-2. **Task Management**:
-   - Update task status in real-time as you work
-   - Mark tasks complete IMMEDIATELY after finishing (don't batch completions)
-   - Exactly ONE task must be in_progress at any time (not less, not more)
-   - Complete current tasks before starting new ones
-   - Remove tasks that are no longer relevant from the list entirely
-
-3. **Task Completion Requirements**:
-   - ONLY mark a task as completed when you have FULLY accomplished it
-   - If you encounter errors, blockers, or cannot finish, keep the task as in_progress
-   - When blocked, create a new task describing what needs to be resolved
-   - Never mark a task as completed if:
-     - Tests are failing
-     - Implementation is partial
-     - You encountered unresolved errors
-     - You couldn't find necessary files or dependencies
-
-4. **Task Breakdown**:
-   - Create specific, actionable items
-   - Break complex tasks into smaller, manageable steps
-   - Use clear, descriptive task names
-   - Always provide both forms:
-     - content: "Fix authentication bug"
-     - activeForm: "Fixing authentication bug"
-
-When in doubt, use this tool. Being proactive with task management demonstrates attentiveness and ensures you complete all requirements successfully.`,
-		schema: z.object({
-			todos: z.array(todoItemSchema),
-		}),
-	},
-);

package/dist/definition/tools/web_search.d.ts

@@ -1,4 +0,0 @@
-import { ExaSearchResults } from "@langchain/exa";
-export declare function makeWebSearchTool(): ExaSearchResults<{
-	text: true;
-}>;

package/dist/definition/tools/web_search.js

@@ -1,26 +0,0 @@
-import { ExaSearchResults } from "@langchain/exa";
-import Exa from "exa-js";
-
-let _webSearchInstance = null;
-export function makeWebSearchTool() {
-	if (_webSearchInstance) {
-		return _webSearchInstance;
-	}
-	const apiKey = process.env.EXA_API_KEY;
-	if (!apiKey) {
-		throw new Error(
-			"EXA_API_KEY environment variable is required to use the web_search tool. " +
-				"Please set it to your Exa API key from https://exa.ai",
-		);
-	}
-	const client = new Exa(apiKey);
-	_webSearchInstance = new ExaSearchResults({
-		client,
-		searchArgs: {
-			numResults: 5,
-			type: "auto",
-			text: true,
-		},
-	});
-	return _webSearchInstance;
-}

package/dist/dev-agent/index.d.ts

@@ -1,2 +0,0 @@
-#!/usr/bin/env bun
-export {};

package/dist/dev-agent/index.js

@@ -1,18 +0,0 @@
-#!/usr/bin/env bun
-import { readFileSync } from "node:fs";
-import { join } from "node:path";
-import { makeHttpTransport, makeStdioTransport } from "../acp-server/index";
-// Load agent definition from JSON file
-const configPath = join(import.meta.dir, "agent.json");
-const agent = JSON.parse(readFileSync(configPath, "utf-8"));
-const transport = process.argv[2] || "stdio";
-if (transport === "http") {
-    makeHttpTransport(agent);
-}
-else if (transport === "stdio") {
-    makeStdioTransport(agent);
-}
-else {
-    console.error(`Invalid transport: ${transport}`);
-    process.exit(1);
-}

package/dist/example.d.ts

@@ -1,2 +0,0 @@
-#!/usr/bin/env bun
-export {};

package/dist/example.js

@@ -1,19 +0,0 @@
-#!/usr/bin/env bun
-import { makeHttpTransport, makeStdioTransport } from "./acp-server/index.js";
-
-const exampleAgent = {
-	model: "claude-sonnet-4-5-20250929",
-	systemPrompt: "You are a helpful assistant.",
-	tools: ["todo_write", "get_weather", "web_search"],
-};
-// Parse transport type from command line argument
-const transport = process.argv[2] || "stdio";
-if (transport === "http") {
-	makeHttpTransport(exampleAgent);
-} else if (transport === "stdio") {
-	makeStdioTransport(exampleAgent);
-} else {
-	console.error(`Invalid transport: ${transport}`);
-	console.error("Usage: bun run example.ts [stdio|http]");
-	process.exit(1);
-}