@hasna/terminal 0.2.0 → 0.2.2

package/dist/App.js

@@ -40,7 +40,7 @@ function maybeCd(command) {
 function newTab(id, cwd) {
     return {
         id, cwd,
-        scroll: [], sessionCmds: [], sessionNl: [],
+        scroll: [], sessionEntries: [], sessionNl: [],
         phase: { type: "input", value: "", cursor: 0, histIdx: -1, raw: false },
         streamLines: [],
     };
@@ -83,10 +83,13 @@ export default function App() {
     const commitStream = (nl, cmd, lines, error) => {
         const truncated = lines.length > MAX_LINES;
         const filePaths = !error ? extractFilePaths(lines) : [];
+        // Build short output summary for session context (first 10 lines)
+        const shortOutput = lines.slice(0, 10).join("\n") + (lines.length > 10 ? `\n... (${lines.length} lines total)` : "");
+        const entry = { nl, cmd, output: shortOutput, error: error || undefined };
         updateTab(t => ({
             ...t,
             streamLines: [],
-            sessionCmds: [...t.sessionCmds.slice(-9), cmd],
+            sessionEntries: [...t.sessionEntries.slice(-9), entry],
             scroll: [...t.scroll, {
                     nl, cmd,
                     lines: truncated ? lines.slice(0, MAX_LINES) : lines,
@@ -137,10 +140,10 @@ export default function App() {
             await runPhase(nl, nl, true);
             return;
         }
-        const sessionCmds = tabs[activeTab].sessionCmds;
+        const sessionEntries = tabs[activeTab].sessionEntries;
         setPhase({ type: "thinking", nl, partial: "" });
         try {
-            const command = await translateToCommand(nl, config.permissions, sessionCmds, partial => setPhase({ type: "thinking", nl, partial }));
+            const command = await translateToCommand(nl, config.permissions, sessionEntries, partial => setPhase({ type: "thinking", nl, partial }));
             const blocked = checkPermissions(command, config.permissions);
             if (blocked) {
                 pushScroll({ nl, cmd: command, lines: [`blocked: ${blocked}`], truncated: false, error: true });
@@ -309,7 +312,7 @@ export default function App() {
                 const { nl, command, errorOutput } = phase;
                 setPhase({ type: "thinking", nl, partial: "" });
                 try {
-                    const fixed = await fixCommand(nl, command, errorOutput, config.permissions, tab.sessionCmds);
+                    const fixed = await fixCommand(nl, command, errorOutput, config.permissions, tab.sessionEntries);
                     const danger = isIrreversible(fixed);
                     if (!config.confirm && !danger) {
                         await runPhase(nl, fixed, false);

package/dist/ai.js

@@ -20,10 +20,10 @@ function pickModel(nl) {
             pick: isComplex ? "smart" : "fast",
         };
     }
-    // Cerebras — single fast model (Llama is already fast)
+    // Cerebras — fast model for simple, smart model for complex
     return {
-        fast: "llama-4-scout-17b-16e",
-        smart: "llama-4-scout-17b-16e",
+        fast: "llama3.1-8b",
+        smart: "llama3.1-8b",
         pick: isComplex ? "smart" : "fast",
     };
 }
@@ -56,7 +56,7 @@ export function checkPermissions(command, perms) {
     return null;
 }
 // ── system prompt ─────────────────────────────────────────────────────────────
-function buildSystemPrompt(perms, sessionCmds) {
+function buildSystemPrompt(perms, sessionEntries) {
     const restrictions = [];
     if (!perms.destructive)
         restrictions.push("- NEVER generate commands that delete, remove, or overwrite files/data");
@@ -71,25 +71,40 @@ function buildSystemPrompt(perms, sessionCmds) {
     const restrictionBlock = restrictions.length > 0
         ? `\n\nRESTRICTIONS:\n${restrictions.join("\n")}\nIf restricted, output: BLOCKED: <reason>`
         : "";
-    const contextBlock = sessionCmds.length > 0
-        ? `\n\nSESSION HISTORY:\n${sessionCmds.map((c) => `$ ${c}`).join("\n")}`
-        : "";
+    let contextBlock = "";
+    if (sessionEntries.length > 0) {
+        const lines = [];
+        for (const e of sessionEntries.slice(-5)) { // last 5 interactions
+            lines.push(`> ${e.nl}`);
+            lines.push(`$ ${e.cmd}`);
+            if (e.output)
+                lines.push(e.output);
+            if (e.error)
+                lines.push("(command failed)");
+        }
+        contextBlock = `\n\nSESSION HISTORY (user intent > command $ output):\n${lines.join("\n")}`;
+    }
     return `You are a terminal assistant. Output ONLY the exact shell command — no explanation, no markdown, no backticks.
+The user describes what they want in plain English. You translate to the exact shell command.
+Pay attention to session history — when the user says "inside X folder" they mean a folder visible in the previous output.
+If the user refers to a relative path from a previous command, resolve it correctly.
 cwd: ${process.cwd()}
 shell: zsh / macOS${restrictionBlock}${contextBlock}`;
 }
 // ── streaming translate ───────────────────────────────────────────────────────
-export async function translateToCommand(nl, perms, sessionCmds, onToken) {
-    // cache hit — instant
-    const cached = cacheGet(nl);
-    if (cached) {
-        onToken?.(cached);
-        return cached;
+export async function translateToCommand(nl, perms, sessionEntries, onToken) {
+    // Only use cache when there's no session context (context makes same NL produce different commands)
+    if (sessionEntries.length === 0) {
+        const cached = cacheGet(nl);
+        if (cached) {
+            onToken?.(cached);
+            return cached;
+        }
     }
     const provider = getProvider();
     const routing = pickModel(nl);
     const model = routing.pick === "smart" ? routing.smart : routing.fast;
-    const system = buildSystemPrompt(perms, sessionCmds);
+    const system = buildSystemPrompt(perms, sessionEntries);
     let text;
     if (onToken) {
         text = await provider.stream(nl, { model, maxTokens: 256, system }, {
@@ -105,10 +120,10 @@ export async function translateToCommand(nl, perms, sessionCmds, onToken) {
     return text;
 }
 // ── prefetch ──────────────────────────────────────────────────────────────────
-export function prefetchNext(lastNl, perms, sessionCmds) {
-    if (cacheGet(lastNl))
+export function prefetchNext(lastNl, perms, sessionEntries) {
+    if (sessionEntries.length === 0 && cacheGet(lastNl))
         return;
-    translateToCommand(lastNl, perms, sessionCmds).catch(() => { });
+    translateToCommand(lastNl, perms, sessionEntries).catch(() => { });
 }
 // ── explain ───────────────────────────────────────────────────────────────────
 export async function explainCommand(command) {
@@ -121,13 +136,13 @@ export async function explainCommand(command) {
     });
 }
 // ── auto-fix ──────────────────────────────────────────────────────────────────
-export async function fixCommand(originalNl, failedCommand, errorOutput, perms, sessionCmds) {
+export async function fixCommand(originalNl, failedCommand, errorOutput, perms, sessionEntries) {
     const provider = getProvider();
     const routing = pickModel(originalNl);
     const text = await provider.complete(`I wanted to: ${originalNl}\nI ran: ${failedCommand}\nError:\n${errorOutput}\n\nGive me the corrected command only.`, {
         model: routing.smart, // always use smart model for fixes
         maxTokens: 256,
-        system: buildSystemPrompt(perms, sessionCmds),
+        system: buildSystemPrompt(perms, sessionEntries),
     });
     if (text.startsWith("BLOCKED:"))
         throw new Error(text);

package/dist/providers/cerebras.js

@@ -1,7 +1,7 @@
 // Cerebras provider — uses OpenAI-compatible API
 // Default for open-source users. Fast inference on Llama models.
 const CEREBRAS_BASE_URL = "https://api.cerebras.ai/v1";
-const DEFAULT_MODEL = "llama-4-scout-17b-16e";
+const DEFAULT_MODEL = "llama3.1-8b";
 export class CerebrasProvider {
     name = "cerebras";
     apiKey;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hasna/terminal",
-  "version": "0.2.0",
+  "version": "0.2.2",
   "description": "Smart terminal wrapper for AI agents and humans — structured output, token compression, MCP server, natural language",
   "type": "module",
   "bin": {

package/src/App.tsx

@@ -1,7 +1,7 @@
 import React, { useState, useCallback, useRef } from "react";
 import { Box, Text, useInput, useApp } from "ink";
 import { spawn } from "child_process";
-import { translateToCommand, explainCommand, fixCommand, checkPermissions, isIrreversible } from "./ai.js";
+import { translateToCommand, explainCommand, fixCommand, checkPermissions, isIrreversible, type SessionEntry } from "./ai.js";
 import { loadHistory, appendHistory, loadConfig, saveConfig, type Permissions } from "./history.js";
 import { loadCache } from "./cache.js";
 import Onboarding from "./Onboarding.js";
@@ -39,7 +39,7 @@ interface TabState {
   id: number;
   cwd: string;
   scroll: ScrollEntry[];
-  sessionCmds: string[];
+  sessionEntries: SessionEntry[];
   sessionNl: string[];
   phase: Phase;
   streamLines: string[];
@@ -77,7 +77,7 @@ function maybeCd(command: string): string | null {
 function newTab(id: number, cwd: string): TabState {
   return {
     id, cwd,
-    scroll: [], sessionCmds: [], sessionNl: [],
+    scroll: [], sessionEntries: [], sessionNl: [],
     phase: { type: "input", value: "", cursor: 0, histIdx: -1, raw: false },
     streamLines: [],
   };
@@ -133,10 +133,13 @@ export default function App() {
   const commitStream = (nl: string, cmd: string, lines: string[], error: boolean) => {
     const truncated = lines.length > MAX_LINES;
     const filePaths = !error ? extractFilePaths(lines) : [];
+    // Build short output summary for session context (first 10 lines)
+    const shortOutput = lines.slice(0, 10).join("\n") + (lines.length > 10 ? `\n... (${lines.length} lines total)` : "");
+    const entry: SessionEntry = { nl, cmd, output: shortOutput, error: error || undefined };
     updateTab(t => ({
       ...t,
       streamLines: [],
-      sessionCmds: [...t.sessionCmds.slice(-9), cmd],
+      sessionEntries: [...t.sessionEntries.slice(-9), entry],
       scroll: [...t.scroll, {
         nl, cmd,
         lines: truncated ? lines.slice(0, MAX_LINES) : lines,
@@ -192,10 +195,10 @@ export default function App() {
     updateTab(t => ({ ...t, sessionNl: [...t.sessionNl, nl] }));
     if (raw) { await runPhase(nl, nl, true); return; }
 
-    const sessionCmds = tabs[activeTab].sessionCmds;
+    const sessionEntries = tabs[activeTab].sessionEntries;
     setPhase({ type: "thinking", nl, partial: "" });
     try {
-      const command = await translateToCommand(nl, config.permissions, sessionCmds, partial =>
+      const command = await translateToCommand(nl, config.permissions, sessionEntries, partial =>
         setPhase({ type: "thinking", nl, partial })
       );
       const blocked = checkPermissions(command, config.permissions);
@@ -346,7 +349,7 @@ export default function App() {
         const { nl, command, errorOutput } = phase;
         setPhase({ type: "thinking", nl, partial: "" });
         try {
-          const fixed = await fixCommand(nl, command, errorOutput, config.permissions, tab.sessionCmds);
+          const fixed = await fixCommand(nl, command, errorOutput, config.permissions, tab.sessionEntries);
           const danger = isIrreversible(fixed);
           if (!config.confirm && !danger) { await runPhase(nl, fixed, false); return; }
           setPhase({ type: "confirm", nl, command: fixed, danger });

package/src/ai.ts

@@ -26,10 +26,10 @@ function pickModel(nl: string): { fast: string; smart: string; pick: "fast" | "s
     };
   }
 
-  // Cerebras — single fast model (Llama is already fast)
+  // Cerebras — fast model for simple, smart model for complex
   return {
-    fast: "llama-4-scout-17b-16e",
-    smart: "llama-4-scout-17b-16e",
+    fast: "llama3.1-8b",
+    smart: "llama3.1-8b",
     pick: isComplex ? "smart" : "fast",
   };
 }
@@ -68,9 +68,18 @@ export function checkPermissions(command: string, perms: Permissions): string |
   return null;
 }
 
+// ── session context ──────────────────────────────────────────────────────────
+
+export interface SessionEntry {
+  nl: string;
+  cmd: string;
+  output?: string; // short output (first few lines)
+  error?: boolean;
+}
+
 // ── system prompt ─────────────────────────────────────────────────────────────
 
-function buildSystemPrompt(perms: Permissions, sessionCmds: string[]): string {
+function buildSystemPrompt(perms: Permissions, sessionEntries: SessionEntry[]): string {
   const restrictions: string[] = [];
   if (!perms.destructive)
     restrictions.push("- NEVER generate commands that delete, remove, or overwrite files/data");
@@ -87,11 +96,22 @@ function buildSystemPrompt(perms: Permissions, sessionCmds: string[]): string {
     ? `\n\nRESTRICTIONS:\n${restrictions.join("\n")}\nIf restricted, output: BLOCKED: <reason>`
     : "";
 
-  const contextBlock = sessionCmds.length > 0
-    ? `\n\nSESSION HISTORY:\n${sessionCmds.map((c) => `$ ${c}`).join("\n")}`
-    : "";
+  let contextBlock = "";
+  if (sessionEntries.length > 0) {
+    const lines: string[] = [];
+    for (const e of sessionEntries.slice(-5)) { // last 5 interactions
+      lines.push(`> ${e.nl}`);
+      lines.push(`$ ${e.cmd}`);
+      if (e.output) lines.push(e.output);
+      if (e.error) lines.push("(command failed)");
+    }
+    contextBlock = `\n\nSESSION HISTORY (user intent > command $ output):\n${lines.join("\n")}`;
+  }
 
   return `You are a terminal assistant. Output ONLY the exact shell command — no explanation, no markdown, no backticks.
+The user describes what they want in plain English. You translate to the exact shell command.
+Pay attention to session history — when the user says "inside X folder" they mean a folder visible in the previous output.
+If the user refers to a relative path from a previous command, resolve it correctly.
 cwd: ${process.cwd()}
 shell: zsh / macOS${restrictionBlock}${contextBlock}`;
 }
@@ -101,17 +121,19 @@ shell: zsh / macOS${restrictionBlock}${contextBlock}`;
 export async function translateToCommand(
   nl: string,
   perms: Permissions,
-  sessionCmds: string[],
+  sessionEntries: SessionEntry[],
   onToken?: (partial: string) => void
 ): Promise<string> {
-  // cache hit — instant
-  const cached = cacheGet(nl);
-  if (cached) { onToken?.(cached); return cached; }
+  // Only use cache when there's no session context (context makes same NL produce different commands)
+  if (sessionEntries.length === 0) {
+    const cached = cacheGet(nl);
+    if (cached) { onToken?.(cached); return cached; }
+  }
 
   const provider = getProvider();
   const routing = pickModel(nl);
   const model = routing.pick === "smart" ? routing.smart : routing.fast;
-  const system = buildSystemPrompt(perms, sessionCmds);
+  const system = buildSystemPrompt(perms, sessionEntries);
 
   let text: string;
 
@@ -133,10 +155,10 @@ export async function translateToCommand(
 export function prefetchNext(
   lastNl: string,
   perms: Permissions,
-  sessionCmds: string[]
+  sessionEntries: SessionEntry[]
 ) {
-  if (cacheGet(lastNl)) return;
-  translateToCommand(lastNl, perms, sessionCmds).catch(() => {});
+  if (sessionEntries.length === 0 && cacheGet(lastNl)) return;
+  translateToCommand(lastNl, perms, sessionEntries).catch(() => {});
 }
 
 // ── explain ───────────────────────────────────────────────────────────────────
@@ -158,7 +180,7 @@ export async function fixCommand(
   failedCommand: string,
   errorOutput: string,
   perms: Permissions,
-  sessionCmds: string[]
+  sessionEntries: SessionEntry[]
 ): Promise<string> {
   const provider = getProvider();
   const routing = pickModel(originalNl);
@@ -167,7 +189,7 @@ export async function fixCommand(
     {
       model: routing.smart, // always use smart model for fixes
       maxTokens: 256,
-      system: buildSystemPrompt(perms, sessionCmds),
+      system: buildSystemPrompt(perms, sessionEntries),
     }
   );
   if (text.startsWith("BLOCKED:")) throw new Error(text);

package/src/providers/cerebras.ts

@@ -4,7 +4,7 @@
 import type { LLMProvider, ProviderOptions, StreamCallbacks } from "./base.js";
 
 const CEREBRAS_BASE_URL = "https://api.cerebras.ai/v1";
-const DEFAULT_MODEL = "llama-4-scout-17b-16e";
+const DEFAULT_MODEL = "llama3.1-8b";
 
 export class CerebrasProvider implements LLMProvider {
   readonly name = "cerebras";