@hasna/terminal 4.1.0 → 4.3.0

package/dist/mcp/server.js

@@ -67,10 +67,22 @@ function resolvePath(p, cwd) {
 export function createServer() {
     const server = new McpServer({
         name: "terminal",
-        version: "3.4.0",
+        version: "4.2.0",
     });
     // Create a session for this MCP server instance
     const sessionId = createSession(process.cwd(), "mcp");
+    // ── Mementos: cross-session project memory ────────────────────────────────
+    let mementosProjectId = null;
+    try {
+        const mementos = require("@hasna/mementos");
+        const projectName = process.cwd().split("/").pop() ?? "unknown";
+        const project = mementos.registerProject(projectName, process.cwd());
+        mementosProjectId = project?.id ?? null;
+        mementos.registerAgent("terminal-mcp");
+        if (mementosProjectId)
+            mementos.setFocus(mementosProjectId);
+    }
+    catch { } // mementos optional — works without it
     /** Log a tool call to sessions.db for economy tracking */
     function logCall(tool, data) {
         try {
@@ -166,11 +178,12 @@ export function createServer() {
         command: z.string().describe("Shell command to execute"),
         cwd: z.string().optional().describe("Working directory"),
         timeout: z.number().optional().describe("Timeout in ms (default: 30000)"),
-    }, async ({ command, cwd, timeout }) => {
+        verbosity: z.enum(["minimal", "normal", "detailed"]).optional().describe("Summary detail level (default: normal)"),
+    }, async ({ command, cwd, timeout, verbosity }) => {
         const start = Date.now();
         const result = await exec(command, cwd, timeout ?? 30000, true);
         const output = (result.stdout + result.stderr).trim();
-        const processed = await processOutput(command, output);
+        const processed = await processOutput(command, output, undefined, verbosity);
         const detailKey = output.split("\n").length > 15 ? storeOutput(command, output) : undefined;
         logCall("execute_smart", { command, outputTokens: estimateTokens(output), tokensSaved: processed.tokensSaved, durationMs: Date.now() - start, exitCode: result.exitCode, aiProcessed: processed.aiProcessed });
         return {
@@ -498,18 +511,21 @@ export function createServer() {
         offset: z.number().optional().describe("Start line (0-indexed)"),
         limit: z.number().optional().describe("Max lines to return"),
         summarize: z.boolean().optional().describe("Return AI summary instead of full content (saves ~90% tokens)"),
-    }, async ({ path: rawPath, offset, limit, summarize }) => {
+        focus: z.string().optional().describe("Focus hint for summary (e.g., 'public API', 'error handling', 'auth logic')"),
+    }, async ({ path: rawPath, offset, limit, summarize, focus }) => {
         const start = Date.now();
         const path = resolvePath(rawPath);
         const result = cachedRead(path, { offset, limit });
         if (summarize && result.content.length > 500) {
-            // AI-native file summary — ask directly what the file does
             const provider = getOutputProvider();
             const outputModel = provider.name === "groq" ? "llama-3.1-8b-instant" : undefined;
             const content = result.content.length > 8000 ? result.content.slice(0, 8000) : result.content;
+            const focusInstruction = focus
+                ? `Focus specifically on: ${focus}. Describe only aspects related to "${focus}".`
+                : `Describe what this source file does in 2-4 lines. Include: main class/module name, key methods/functions, what it exports, and its purpose.`;
             const summary = await provider.complete(`File: ${path}\n\n${content}`, {
                 model: outputModel,
-                system: `Describe what this source file does in 2-4 lines. Include: main class/module name, key methods/functions, what it exports, and its purpose. Be specific — name the actual functions and what they do. Never just say "N lines of code."`,
+                system: `${focusInstruction} Be specific — name the actual functions and what they do. Never just say "N lines of code."`,
                 maxTokens: 300,
                 temperature: 0.2,
             });
@@ -1208,6 +1224,146 @@ Be specific, not generic. Only flag real problems.`,
         });
         return { content: [{ type: "text", text: recommendation }] };
     });
+    // ── batch: multiple operations in one round trip ───────────────────────────
+    server.tool("batch", "Run multiple operations in ONE call. Saves N-1 round trips. Each op can be: execute (run command), read (file read/summarize), search (grep pattern), or symbols (file outline).", {
+        ops: z.array(z.object({
+            type: z.enum(["execute", "read", "write", "search", "symbols"]).describe("Operation type"),
+            command: z.string().optional().describe("Shell command (for execute)"),
+            path: z.string().optional().describe("File path (for read/write/symbols)"),
+            content: z.string().optional().describe("File content (for write)"),
+            pattern: z.string().optional().describe("Search pattern (for search)"),
+            summarize: z.boolean().optional().describe("AI summarize (for read)"),
+            format: z.enum(["raw", "summary"]).optional().describe("Output format (for execute)"),
+        })).describe("Array of operations to run"),
+        cwd: z.string().optional().describe("Working directory for all ops"),
+    }, async ({ ops, cwd }) => {
+        const start = Date.now();
+        const workDir = cwd ?? process.cwd();
+        const results = [];
+        for (let i = 0; i < ops.slice(0, 10).length; i++) {
+            const op = ops[i];
+            try {
+                if (op.type === "execute" && op.command) {
+                    const result = await exec(op.command, workDir, 30000);
+                    const output = (result.stdout + result.stderr).trim();
+                    if (op.format === "summary" && output.split("\n").length > 15) {
+                        const processed = await processOutput(op.command, output);
+                        results.push({ op: i, type: "execute", summary: processed.summary, exitCode: result.exitCode, tokensSaved: processed.tokensSaved });
+                    }
+                    else {
+                        results.push({ op: i, type: "execute", output: stripAnsi(output).slice(0, 2000), exitCode: result.exitCode });
+                    }
+                }
+                else if (op.type === "read" && op.path) {
+                    const filePath = resolvePath(op.path, workDir);
+                    const result = cachedRead(filePath, {});
+                    if (op.summarize && result.content.length > 500) {
+                        const provider = getOutputProvider();
+                        const outputModel = provider.name === "groq" ? "llama-3.1-8b-instant" : undefined;
+                        const content = result.content.length > 8000 ? result.content.slice(0, 8000) : result.content;
+                        const summary = await provider.complete(`File: ${filePath}\n\n${content}`, {
+                            model: outputModel,
+                            system: `Describe what this source file does in 2-4 lines. Include: main class/module name, key methods/functions, what it exports, and its purpose. Be specific.`,
+                            maxTokens: 300, temperature: 0.2,
+                        });
+                        results.push({ op: i, type: "read", path: op.path, summary, lines: result.content.split("\n").length });
+                    }
+                    else {
+                        results.push({ op: i, type: "read", path: op.path, content: result.content, lines: result.content.split("\n").length });
+                    }
+                }
+                else if (op.type === "write" && op.path && op.content !== undefined) {
+                    const filePath = resolvePath(op.path, workDir);
+                    const { writeFileSync, mkdirSync, existsSync } = await import("fs");
+                    const { dirname } = await import("path");
+                    const dir = dirname(filePath);
+                    if (!existsSync(dir))
+                        mkdirSync(dir, { recursive: true });
+                    writeFileSync(filePath, op.content);
+                    results.push({ op: i, type: "write", path: op.path, ok: true, bytes: op.content.length });
+                }
+                else if (op.type === "search" && op.pattern) {
+                    // Search accepts both files and directories — resolve to parent dir if file
+                    let searchPath = op.path ? resolvePath(op.path, workDir) : workDir;
+                    try {
+                        const { statSync } = await import("fs");
+                        if (statSync(searchPath).isFile())
+                            searchPath = searchPath.replace(/\/[^/]+$/, "");
+                    }
+                    catch { }
+                    const result = await searchContent(op.pattern, searchPath, {});
+                    results.push({ op: i, type: "search", pattern: op.pattern, totalMatches: result.totalMatches, files: result.files.slice(0, 10) });
+                }
+                else if (op.type === "symbols" && op.path) {
+                    const filePath = resolvePath(op.path, workDir);
+                    const result = cachedRead(filePath, {});
+                    if (result.content && !result.content.startsWith("Error:")) {
+                        const provider = getOutputProvider();
+                        const outputModel = provider.name === "groq" ? "llama-3.1-8b-instant" : undefined;
+                        const content = result.content.length > 8000 ? result.content.slice(0, 8000) : result.content;
+                        const summary = await provider.complete(`File: ${filePath}\n\n${content}`, {
+                            model: outputModel,
+                            system: `Extract all symbols. Return ONLY a JSON array. Each: {"name":"x","kind":"function|class|method|interface|type","line":N,"signature":"brief"}. For class methods use "Class.method". Exclude imports.`,
+                            maxTokens: 2000, temperature: 0,
+                        });
+                        let symbols = [];
+                        try {
+                            const m = summary.match(/\[[\s\S]*\]/);
+                            if (m)
+                                symbols = JSON.parse(m[0]);
+                        }
+                        catch { }
+                        results.push({ op: i, type: "symbols", path: op.path, symbols });
+                    }
+                    else {
+                        results.push({ op: i, type: "symbols", path: op.path, error: "Cannot read file" });
+                    }
+                }
+            }
+            catch (err) {
+                results.push({ op: i, type: op.type, error: err.message?.slice(0, 200) });
+            }
+        }
+        logCall("batch", { command: `${ops.length} ops`, durationMs: Date.now() - start, aiProcessed: true });
+        return { content: [{ type: "text", text: JSON.stringify({ results, total: results.length, durationMs: Date.now() - start }) }] };
+    });
+    // ── Cross-session memory (mementos SDK) ────────────────────────────────────
+    server.tool("remember", "Save a learning about this project for future sessions. Persists across restarts. Use for: project patterns, conventions, toolchain quirks, architectural decisions.", {
+        key: z.string().describe("Short key (e.g., 'test-command', 'deploy-process', 'auth-pattern')"),
+        value: z.string().describe("What to remember"),
+        importance: z.number().optional().describe("1-10, default 7"),
+    }, async ({ key, value, importance }) => {
+        try {
+            const mementos = require("@hasna/mementos");
+            mementos.createMemory({ key, value, scope: "shared", category: "knowledge", importance: importance ?? 7 });
+            logCall("remember", { command: `remember: ${key}` });
+            return { content: [{ type: "text", text: JSON.stringify({ saved: key }) }] };
+        }
+        catch (e) {
+            return { content: [{ type: "text", text: JSON.stringify({ error: e.message?.slice(0, 200) }) }] };
+        }
+    });
+    server.tool("recall", "Recall project memories from previous sessions. Returns all saved learnings, patterns, and decisions for this project.", {
+        search: z.string().optional().describe("Search query to filter memories"),
+        limit: z.number().optional().describe("Max memories to return (default: 20)"),
+    }, async ({ search, limit }) => {
+        try {
+            const mementos = require("@hasna/mementos");
+            let memories;
+            if (search) {
+                memories = mementos.searchMemories(search, { limit: limit ?? 20 });
+            }
+            else {
+                memories = mementos.listMemories({ scope: "shared", limit: limit ?? 20 });
+            }
+            const items = (memories ?? []).map((m) => ({ key: m.key, value: m.value, importance: m.importance }));
+            logCall("recall", { command: `recall${search ? `: ${search}` : ""}` });
+            return { content: [{ type: "text", text: JSON.stringify({ memories: items, total: items.length }) }] };
+        }
+        catch (e) {
+            return { content: [{ type: "text", text: JSON.stringify({ error: e.message?.slice(0, 200), memories: [] }) }] };
+        }
+    });
     return server;
 }
 // ── main: start MCP server via stdio ─────────────────────────────────────────

package/dist/output-processor.js

@@ -102,7 +102,7 @@ RULES:
  * Process command output through AI summarization.
  * Cheap AI call (~100 tokens) saves 1000+ tokens downstream.
  */
-export async function processOutput(command, output, originalPrompt) {
+export async function processOutput(command, output, originalPrompt, verbosity) {
     const lines = output.split("\n");
     // Fingerprint check — skip AI entirely for known patterns (0ms, $0)
     const fp = fingerprint(command, output);
@@ -157,10 +157,14 @@ export async function processOutput(command, output, originalPrompt) {
         // Falls back to main provider if Groq unavailable
         const provider = getOutputProvider();
         const outputModel = provider.name === "groq" ? "llama-3.1-8b-instant" : undefined;
+        const verbosityHint = verbosity === "minimal" ? "\nBe ULTRA concise — 1-2 lines max. Status + key number only."
+            : verbosity === "detailed" ? "\nBe thorough — include all relevant details, up to 15 lines."
+                : ""; // normal = default 8 lines from SUMMARIZE_PROMPT
+        const maxTok = verbosity === "minimal" ? 100 : verbosity === "detailed" ? 500 : 300;
         const summary = await provider.complete(`${originalPrompt ? `User asked: ${originalPrompt}\n` : ""}Command: ${command}\nOutput (${lines.length} lines):\n${toSummarize}${hintsBlock}${profileHints}`, {
             model: outputModel,
-            system: SUMMARIZE_PROMPT,
-            maxTokens: 300,
+            system: SUMMARIZE_PROMPT + verbosityHint,
+            maxTokens: maxTok,
             temperature: 0.2,
         });
         const originalTokens = estimateTokens(output);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hasna/terminal",
-  "version": "4.1.0",
+  "version": "4.3.0",
   "description": "Smart terminal wrapper for AI agents and humans — structured output, token compression, MCP server, natural language",
   "type": "module",
   "files": [
@@ -22,6 +22,7 @@
   },
   "dependencies": {
     "@anthropic-ai/sdk": "^0.39.0",
+    "@hasna/mementos": "^0.10.0",
     "@modelcontextprotocol/sdk": "^1.27.1",
     "@typescript/vfs": "^1.6.4",
     "ink": "^5.0.1",

package/src/mcp/server.ts

@@ -71,12 +71,23 @@ function resolvePath(p: string, cwd?: string): string {
 export function createServer(): McpServer {
   const server = new McpServer({
     name: "terminal",
-    version: "3.4.0",
+    version: "4.2.0",
   });
 
   // Create a session for this MCP server instance
   const sessionId = createSession(process.cwd(), "mcp");
 
+  // ── Mementos: cross-session project memory ────────────────────────────────
+  let mementosProjectId: string | null = null;
+  try {
+    const mementos = require("@hasna/mementos");
+    const projectName = process.cwd().split("/").pop() ?? "unknown";
+    const project = mementos.registerProject(projectName, process.cwd());
+    mementosProjectId = project?.id ?? null;
+    mementos.registerAgent("terminal-mcp");
+    if (mementosProjectId) mementos.setFocus(mementosProjectId);
+  } catch {} // mementos optional — works without it
+
   /** Log a tool call to sessions.db for economy tracking */
   function logCall(tool: string, data: {
     command?: string;
@@ -194,12 +205,13 @@ export function createServer(): McpServer {
       command: z.string().describe("Shell command to execute"),
       cwd: z.string().optional().describe("Working directory"),
       timeout: z.number().optional().describe("Timeout in ms (default: 30000)"),
+      verbosity: z.enum(["minimal", "normal", "detailed"]).optional().describe("Summary detail level (default: normal)"),
     },
-    async ({ command, cwd, timeout }) => {
+    async ({ command, cwd, timeout, verbosity }) => {
       const start = Date.now();
       const result = await exec(command, cwd, timeout ?? 30000, true);
       const output = (result.stdout + result.stderr).trim();
-      const processed = await processOutput(command, output);
+      const processed = await processOutput(command, output, undefined, verbosity);
 
       const detailKey = output.split("\n").length > 15 ? storeOutput(command, output) : undefined;
       logCall("execute_smart", { command, outputTokens: estimateTokens(output), tokensSaved: processed.tokensSaved, durationMs: Date.now() - start, exitCode: result.exitCode, aiProcessed: processed.aiProcessed });
@@ -698,22 +710,25 @@ export function createServer(): McpServer {
       offset: z.number().optional().describe("Start line (0-indexed)"),
       limit: z.number().optional().describe("Max lines to return"),
       summarize: z.boolean().optional().describe("Return AI summary instead of full content (saves ~90% tokens)"),
+      focus: z.string().optional().describe("Focus hint for summary (e.g., 'public API', 'error handling', 'auth logic')"),
     },
-    async ({ path: rawPath, offset, limit, summarize }) => {
+    async ({ path: rawPath, offset, limit, summarize, focus }) => {
       const start = Date.now();
       const path = resolvePath(rawPath);
       const result = cachedRead(path, { offset, limit });
 
       if (summarize && result.content.length > 500) {
-        // AI-native file summary — ask directly what the file does
         const provider = getOutputProvider();
         const outputModel = provider.name === "groq" ? "llama-3.1-8b-instant" : undefined;
         const content = result.content.length > 8000 ? result.content.slice(0, 8000) : result.content;
+        const focusInstruction = focus
+          ? `Focus specifically on: ${focus}. Describe only aspects related to "${focus}".`
+          : `Describe what this source file does in 2-4 lines. Include: main class/module name, key methods/functions, what it exports, and its purpose.`;
         const summary = await provider.complete(
           `File: ${path}\n\n${content}`,
           {
             model: outputModel,
-            system: `Describe what this source file does in 2-4 lines. Include: main class/module name, key methods/functions, what it exports, and its purpose. Be specific — name the actual functions and what they do. Never just say "N lines of code."`,
+            system: `${focusInstruction} Be specific — name the actual functions and what they do. Never just say "N lines of code."`,
             maxTokens: 300,
             temperature: 0.2,
           }
@@ -1570,6 +1585,149 @@ Be specific, not generic. Only flag real problems.`,
     }
   );
 
+  // ── batch: multiple operations in one round trip ───────────────────────────
+
+  server.tool(
+    "batch",
+    "Run multiple operations in ONE call. Saves N-1 round trips. Each op can be: execute (run command), read (file read/summarize), search (grep pattern), or symbols (file outline).",
+    {
+      ops: z.array(z.object({
+        type: z.enum(["execute", "read", "write", "search", "symbols"]).describe("Operation type"),
+        command: z.string().optional().describe("Shell command (for execute)"),
+        path: z.string().optional().describe("File path (for read/write/symbols)"),
+        content: z.string().optional().describe("File content (for write)"),
+        pattern: z.string().optional().describe("Search pattern (for search)"),
+        summarize: z.boolean().optional().describe("AI summarize (for read)"),
+        format: z.enum(["raw", "summary"]).optional().describe("Output format (for execute)"),
+      })).describe("Array of operations to run"),
+      cwd: z.string().optional().describe("Working directory for all ops"),
+    },
+    async ({ ops, cwd }) => {
+      const start = Date.now();
+      const workDir = cwd ?? process.cwd();
+      const results: Record<string, any>[] = [];
+
+      for (let i = 0; i < ops.slice(0, 10).length; i++) {
+        const op = ops[i];
+        try {
+          if (op.type === "execute" && op.command) {
+            const result = await exec(op.command, workDir, 30000);
+            const output = (result.stdout + result.stderr).trim();
+            if (op.format === "summary" && output.split("\n").length > 15) {
+              const processed = await processOutput(op.command, output);
+              results.push({ op: i, type: "execute", summary: processed.summary, exitCode: result.exitCode, tokensSaved: processed.tokensSaved });
+            } else {
+              results.push({ op: i, type: "execute", output: stripAnsi(output).slice(0, 2000), exitCode: result.exitCode });
+            }
+          } else if (op.type === "read" && op.path) {
+            const filePath = resolvePath(op.path, workDir);
+            const result = cachedRead(filePath, {});
+            if (op.summarize && result.content.length > 500) {
+              const provider = getOutputProvider();
+              const outputModel = provider.name === "groq" ? "llama-3.1-8b-instant" : undefined;
+              const content = result.content.length > 8000 ? result.content.slice(0, 8000) : result.content;
+              const summary = await provider.complete(`File: ${filePath}\n\n${content}`, {
+                model: outputModel,
+                system: `Describe what this source file does in 2-4 lines. Include: main class/module name, key methods/functions, what it exports, and its purpose. Be specific.`,
+                maxTokens: 300, temperature: 0.2,
+              });
+              results.push({ op: i, type: "read", path: op.path, summary, lines: result.content.split("\n").length });
+            } else {
+              results.push({ op: i, type: "read", path: op.path, content: result.content, lines: result.content.split("\n").length });
+            }
+          } else if (op.type === "write" && op.path && op.content !== undefined) {
+            const filePath = resolvePath(op.path, workDir);
+            const { writeFileSync, mkdirSync, existsSync } = await import("fs");
+            const { dirname } = await import("path");
+            const dir = dirname(filePath);
+            if (!existsSync(dir)) mkdirSync(dir, { recursive: true });
+            writeFileSync(filePath, op.content);
+            results.push({ op: i, type: "write", path: op.path, ok: true, bytes: op.content.length });
+          } else if (op.type === "search" && op.pattern) {
+            // Search accepts both files and directories — resolve to parent dir if file
+            let searchPath = op.path ? resolvePath(op.path, workDir) : workDir;
+            try {
+              const { statSync } = await import("fs");
+              if (statSync(searchPath).isFile()) searchPath = searchPath.replace(/\/[^/]+$/, "");
+            } catch {}
+            const result = await searchContent(op.pattern, searchPath, {});
+            results.push({ op: i, type: "search", pattern: op.pattern, totalMatches: result.totalMatches, files: result.files.slice(0, 10) });
+          } else if (op.type === "symbols" && op.path) {
+            const filePath = resolvePath(op.path, workDir);
+            const result = cachedRead(filePath, {});
+            if (result.content && !result.content.startsWith("Error:")) {
+              const provider = getOutputProvider();
+              const outputModel = provider.name === "groq" ? "llama-3.1-8b-instant" : undefined;
+              const content = result.content.length > 8000 ? result.content.slice(0, 8000) : result.content;
+              const summary = await provider.complete(`File: ${filePath}\n\n${content}`, {
+                model: outputModel,
+                system: `Extract all symbols. Return ONLY a JSON array. Each: {"name":"x","kind":"function|class|method|interface|type","line":N,"signature":"brief"}. For class methods use "Class.method". Exclude imports.`,
+                maxTokens: 2000, temperature: 0,
+              });
+              let symbols: any[] = [];
+              try { const m = summary.match(/\[[\s\S]*\]/); if (m) symbols = JSON.parse(m[0]); } catch {}
+              results.push({ op: i, type: "symbols", path: op.path, symbols });
+            } else {
+              results.push({ op: i, type: "symbols", path: op.path, error: "Cannot read file" });
+            }
+          }
+        } catch (err: any) {
+          results.push({ op: i, type: op.type, error: err.message?.slice(0, 200) });
+        }
+      }
+
+      logCall("batch", { command: `${ops.length} ops`, durationMs: Date.now() - start, aiProcessed: true });
+      return { content: [{ type: "text" as const, text: JSON.stringify({ results, total: results.length, durationMs: Date.now() - start }) }] };
+    }
+  );
+
+  // ── Cross-session memory (mementos SDK) ────────────────────────────────────
+
+  server.tool(
+    "remember",
+    "Save a learning about this project for future sessions. Persists across restarts. Use for: project patterns, conventions, toolchain quirks, architectural decisions.",
+    {
+      key: z.string().describe("Short key (e.g., 'test-command', 'deploy-process', 'auth-pattern')"),
+      value: z.string().describe("What to remember"),
+      importance: z.number().optional().describe("1-10, default 7"),
+    },
+    async ({ key, value, importance }) => {
+      try {
+        const mementos = require("@hasna/mementos");
+        mementos.createMemory({ key, value, scope: "shared", category: "knowledge", importance: importance ?? 7 });
+        logCall("remember", { command: `remember: ${key}` });
+        return { content: [{ type: "text" as const, text: JSON.stringify({ saved: key }) }] };
+      } catch (e: any) {
+        return { content: [{ type: "text" as const, text: JSON.stringify({ error: e.message?.slice(0, 200) }) }] };
+      }
+    }
+  );
+
+  server.tool(
+    "recall",
+    "Recall project memories from previous sessions. Returns all saved learnings, patterns, and decisions for this project.",
+    {
+      search: z.string().optional().describe("Search query to filter memories"),
+      limit: z.number().optional().describe("Max memories to return (default: 20)"),
+    },
+    async ({ search, limit }) => {
+      try {
+        const mementos = require("@hasna/mementos");
+        let memories;
+        if (search) {
+          memories = mementos.searchMemories(search, { limit: limit ?? 20 });
+        } else {
+          memories = mementos.listMemories({ scope: "shared", limit: limit ?? 20 });
+        }
+        const items = (memories ?? []).map((m: any) => ({ key: m.key, value: m.value, importance: m.importance }));
+        logCall("recall", { command: `recall${search ? `: ${search}` : ""}` });
+        return { content: [{ type: "text" as const, text: JSON.stringify({ memories: items, total: items.length }) }] };
+      } catch (e: any) {
+        return { content: [{ type: "text" as const, text: JSON.stringify({ error: e.message?.slice(0, 200), memories: [] }) }] };
+      }
+    }
+  );
+
   return server;
 }
 

package/src/output-processor.ts

@@ -140,6 +140,7 @@ export async function processOutput(
   command: string,
   output: string,
   originalPrompt?: string,
+  verbosity?: "minimal" | "normal" | "detailed",
 ): Promise<ProcessedOutput> {
   const lines = output.split("\n");
 
@@ -201,12 +202,16 @@ export async function processOutput(
     // Falls back to main provider if Groq unavailable
     const provider = getOutputProvider();
     const outputModel = provider.name === "groq" ? "llama-3.1-8b-instant" : undefined;
+    const verbosityHint = verbosity === "minimal" ? "\nBe ULTRA concise — 1-2 lines max. Status + key number only."
+      : verbosity === "detailed" ? "\nBe thorough — include all relevant details, up to 15 lines."
+      : ""; // normal = default 8 lines from SUMMARIZE_PROMPT
+    const maxTok = verbosity === "minimal" ? 100 : verbosity === "detailed" ? 500 : 300;
     const summary = await provider.complete(
       `${originalPrompt ? `User asked: ${originalPrompt}\n` : ""}Command: ${command}\nOutput (${lines.length} lines):\n${toSummarize}${hintsBlock}${profileHints}`,
       {
         model: outputModel,
-        system: SUMMARIZE_PROMPT,
-        maxTokens: 300,
+        system: SUMMARIZE_PROMPT + verbosityHint,
+        maxTokens: maxTok,
         temperature: 0.2,
       }
     );