@voidwire/llm-summarize 3.0.0 → 3.1.0

package/cli.ts

@@ -29,7 +29,12 @@
  *   2 - Client error (missing args, invalid config)
  */
 
-import { summarize, loadConfig, type SummarizeOptions } from "./index";
+import {
+  summarize,
+  loadConfig,
+  type SummarizeOptions,
+  type SummarizeMode,
+} from "./index";
 
 /**
  * Read text from stdin
@@ -60,11 +65,16 @@ Usage: llm-summarize [options] <text>
        llm-summarize --stdin
 
 Options:
+  --mode <mode>         Summarization mode: quick or insights (default: insights)
   --model <name>        Override model from config
   --max-tokens <n>      Max output tokens (default: from config or 1024)
   --stdin               Read text from stdin
   -h, --help            Show this help
 
+Modes:
+  quick     - Fast one-liner summary (for user prompts)
+  insights  - Full SessionInsights extraction (for responses)
+
 Config file: ~/.config/llm/config.toml
   [llm]
   provider = "ollama"
@@ -127,6 +137,7 @@ async function parseArgs(argv: string[]): Promise<ParsedArgs | null> {
 
   let modelOverride: string | undefined;
   let maxTokensOverride: number | undefined;
+  let modeOverride: SummarizeMode | undefined;
   let useStdin = false;
   let text = "";
 
@@ -137,6 +148,14 @@ async function parseArgs(argv: string[]): Promise<ParsedArgs | null> {
       modelOverride = args[++i];
     } else if (arg === "--max-tokens" && i + 1 < args.length) {
       maxTokensOverride = parseInt(args[++i], 10);
+    } else if (arg === "--mode" && i + 1 < args.length) {
+      const mode = args[++i];
+      if (mode === "quick" || mode === "insights") {
+        modeOverride = mode;
+      } else {
+        console.error(`Invalid mode: ${mode}. Use 'quick' or 'insights'.`);
+        process.exit(2);
+      }
     } else if (arg === "--stdin") {
       useStdin = true;
     } else if (!arg.startsWith("-")) {
@@ -154,6 +173,7 @@ async function parseArgs(argv: string[]): Promise<ParsedArgs | null> {
     options: {
       model: modelOverride,
       maxTokens: maxTokensOverride,
+      mode: modeOverride,
     },
   };
 }

package/index.ts

@@ -45,15 +45,28 @@ export interface LLMConfig {
 export interface SummarizeOptions {
   model?: string;
   maxTokens?: number;
+  mode?: "quick" | "insights";
 }
 
 export type ProviderType = "anthropic" | "openai" | "ollama";
+export type SummarizeMode = "quick" | "insights";
 
 // ============================================================================
-// System Prompt
+// System Prompts
 // ============================================================================
 
-const SYSTEM_PROMPT = `You are an experienced engineering manager reviewing session transcripts to extract actionable team insights.
+/**
+ * Quick mode: Fast one-liner summary for user prompts
+ */
+const QUICK_PROMPT = `Summarize what the user is asking or doing in one sentence.
+Use the user's name from the context in your summary (e.g., "Rudy asked about...").
+
+Output JSON only: {"summary": "One sentence summary"}`;
+
+/**
+ * Insights mode: Full SessionInsights extraction for responses
+ */
+const INSIGHTS_PROMPT = `You are an experienced engineering manager reviewing session transcripts to extract actionable team insights.
 
 Analyze the development session conversation and extract structured observations.
 
@@ -69,6 +82,7 @@ Analyze the development session conversation and extract structured observations
 </output_schema>
 
 <rules>
+- Use the user's name from the context in the summary field (e.g., "Rudy implemented...")
 - Include a field ONLY when the conversation provides clear evidence
 - Extract specifics: "Chose SQLite over Postgres for single-user simplicity" not "Made a database decision"
 - Omit empty arrays entirely
@@ -76,6 +90,13 @@ Analyze the development session conversation and extract structured observations
 
 Output valid JSON only. No markdown code blocks, no explanation.`;
 
+/**
+ * Get prompt for the specified mode
+ */
+function getPromptForMode(mode: SummarizeMode): string {
+  return mode === "quick" ? QUICK_PROMPT : INSIGHTS_PROMPT;
+}
+
 // ============================================================================
 // Response Parsing
 // ============================================================================
@@ -258,6 +279,7 @@ async function callAnthropic(
   model: string,
   maxTokens: number,
   apiKey: string,
+  systemPrompt: string,
   apiBase?: string,
 ): Promise<SummarizeResult> {
   const endpoint = apiBase || "https://api.anthropic.com/v1/messages";
@@ -274,7 +296,7 @@ async function callAnthropic(
         model,
         max_tokens: maxTokens,
         temperature: 0.3,
-        system: SYSTEM_PROMPT,
+        system: systemPrompt,
         messages: [
           {
             role: "user",
@@ -321,6 +343,7 @@ async function callOpenAI(
   model: string,
   maxTokens: number,
   apiKey: string,
+  systemPrompt: string,
   apiBase?: string,
 ): Promise<SummarizeResult> {
   const endpoint = apiBase || "https://api.openai.com/v1/chat/completions";
@@ -339,7 +362,7 @@ async function callOpenAI(
         messages: [
           {
             role: "system",
-            content: SYSTEM_PROMPT,
+            content: systemPrompt,
           },
           {
             role: "user",
@@ -386,6 +409,7 @@ async function callOllama(
   model: string,
   maxTokens: number,
   apiBase: string,
+  systemPrompt: string,
 ): Promise<SummarizeResult> {
   const endpoint = `${apiBase}/api/chat`;
 
@@ -400,7 +424,7 @@ async function callOllama(
         messages: [
           {
             role: "system",
-            content: SYSTEM_PROMPT,
+            content: systemPrompt,
           },
           {
             role: "user",
@@ -453,8 +477,12 @@ async function callOllama(
  *
  * @param text - Text to summarize
  * @param config - LLM configuration (from loadConfig())
- * @param options - Optional overrides for model and maxTokens
- * @returns SummarizeResult with summary or error
+ * @param options - Optional overrides for model, maxTokens, and mode
+ * @returns SummarizeResult with insights or error
+ *
+ * Modes:
+ * - "quick": Fast one-liner summary (for user prompts)
+ * - "insights": Full SessionInsights extraction (for responses, default)
  */
 export async function summarize(
   text: string,
@@ -465,6 +493,8 @@ export async function summarize(
   const model = options?.model || config.model;
   const maxTokens = options?.maxTokens || config.maxTokens;
   const apiKey = config.apiKey;
+  const mode: SummarizeMode = options?.mode || "insights";
+  const systemPrompt = getPromptForMode(mode);
 
   // Validate config
   if (!provider) {
@@ -493,6 +523,7 @@ export async function summarize(
       model,
       maxTokens,
       apiKey!,
+      systemPrompt,
       config.apiBase || undefined,
     );
   } else if (provider === "openai") {
@@ -501,6 +532,7 @@ export async function summarize(
       model,
       maxTokens,
       apiKey!,
+      systemPrompt,
       config.apiBase || undefined,
     );
   } else if (provider === "ollama") {
@@ -509,7 +541,7 @@ export async function summarize(
         error: `No api_base configured for ollama. Set api_base in ~/.config/llm/config.toml`,
       };
     }
-    return callOllama(text, model, maxTokens, config.apiBase);
+    return callOllama(text, model, maxTokens, config.apiBase, systemPrompt);
   } else {
     return {
       error: `Unknown provider: ${provider}. Supported: anthropic, openai, ollama`,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@voidwire/llm-summarize",
-  "version": "3.0.0",
+  "version": "3.1.0",
   "description": "Structured session insight extraction for knowledge systems",
   "type": "module",
   "main": "./index.ts",
@@ -18,6 +18,9 @@
     "README.md",
     "LICENSE"
   ],
+  "scripts": {
+    "test": "bun test"
+  },
   "keywords": [
     "llm",
     "summarize",
@@ -39,8 +42,5 @@
   },
   "engines": {
     "bun": ">=1.0.0"
-  },
-  "scripts": {
-    "test": "bun test"
   }
-}
+}