@hasna/terminal 2.3.0 → 2.3.2

package/dist/ai.js

@@ -0,0 +1,322 @@
+import { cacheGet, cacheSet } from "./cache.js";
+import { getProvider } from "./providers/index.js";
+import { existsSync, readFileSync } from "fs";
+import { join } from "path";
+import { discoverProjectHints, discoverSafetyHints, formatHints } from "./context-hints.js";
+// ── model routing ─────────────────────────────────────────────────────────────
+// Config-driven model selection. Defaults per provider, user can override in ~/.terminal/config.json
+const COMPLEX_SIGNALS = [
+    /\b(undo|revert|rollback|previous|last)\b/i,
+    /\b(all files?|recursively|bulk|batch)\b/i,
+    /\b(pipeline|chain|then|and then|after)\b/i,
+    /\b(if|when|unless|only if)\b/i,
+    /\b(go into|go to|navigate|cd into|enter)\b.*\b(and|then)\b/i,
+    /\b(inside|within|under)\b/i,
+    /[|&;]{2}/,
+];
+/** Default models per provider — user can override in ~/.terminal/config.json under "models" */
+const MODEL_DEFAULTS = {
+    cerebras: { fast: "qwen-3-235b-a22b-instruct-2507", smart: "qwen-3-235b-a22b-instruct-2507" },
+    groq: { fast: "openai/gpt-oss-120b", smart: "moonshotai/kimi-k2-instruct" },
+    xai: { fast: "grok-code-fast-1", smart: "grok-4-fast-non-reasoning" },
+    anthropic: { fast: "claude-haiku-4-5-20251001", smart: "claude-sonnet-4-6" },
+};
+/** Load user model overrides from ~/.terminal/config.json */
+function loadModelOverrides() {
+    try {
+        const configPath = join(process.env.HOME ?? "~", ".terminal", "config.json");
+        if (existsSync(configPath)) {
+            const config = JSON.parse(readFileSync(configPath, "utf8"));
+            return config.models ?? {};
+        }
+    }
+    catch { }
+    return {};
+}
+/** Model routing per provider — config-driven with defaults */
+function pickModel(nl) {
+    const isComplex = COMPLEX_SIGNALS.some((r) => r.test(nl)) || nl.split(" ").length > 10;
+    const provider = getProvider();
+    const defaults = MODEL_DEFAULTS[provider.name] ?? MODEL_DEFAULTS.cerebras;
+    const overrides = loadModelOverrides()[provider.name] ?? {};
+    return {
+        fast: overrides.fast ?? defaults.fast,
+        smart: overrides.smart ?? defaults.smart,
+        pick: isComplex ? "smart" : "fast",
+    };
+}
+// ── irreversibility ───────────────────────────────────────────────────────────
+const IRREVERSIBLE_PATTERNS = [
+    /\brm\s/, /\brmdir\b/, /\btruncate\b/, /\bdrop\s+table\b/i,
+    /\bdelete\s+from\b/i, /\bmv\b.*\/dev\/null/, /\becho\b.*>\s*[^>]/, /\bcat\b.*>\s*[^>]/,
+    /\bdd\b/, /\bmkfs\b/, /\bformat\b/, /\bshred\b/,
+    // Process/service killing
+    /\bkill\b/, /\bkillall\b/, /\bpkill\b/,
+    // Git push/force operations
+    /\bgit\s+push\b/, /\bgit\s+reset\s+--hard\b/, /\bgit\s+force\b/,
+    // Code modification / package installation (security risk)
+    /\bnpx\s+\S+/, /\bnpm\s+install\b/, /\bbun\s+add\b/, /\bpip\s+install\b/,
+    /\bcodemod\b/, /\bsed\s+-i\b/, /\bawk\s.*>\s*\S+\.\w+/, /\bperl\s+-[pi]\b/,
+    // File creation/modification (READ-ONLY terminal)
+    /\btouch\b/, /\bmkdir\b/, /\becho\s.*>/, /\btee\b/, /\bcp\b/, /\bmv\b/,
+    // Starting servers/processes (dangerous from NL)
+    /\b(bun|npm|pnpm|yarn)\s+run\s+dev\b/, /\b(bun|npm)\s+start\b/,
+];
+// Commands that are ALWAYS safe (read-only git, etc.)
+const SAFE_OVERRIDES = [
+    /^\s*git\s+(log|show|diff|branch|status|blame|tag|remote|stash\s+list)\b/,
+    /^\s*git\s+log\b/,
+    // find -exec with read-only tools is safe
+    /\bfind\b.*-exec\s+(wc|cat|head|tail|grep|stat|file|du|ls)\b/,
+    // find without -exec is always safe
+    /^\s*find\b(?!.*-exec\s+(rm|mv|chmod|chown|sed))/,
+    // xargs with read-only tools is safe
+    /\bxargs\s+(wc|cat|head|tail|grep|stat|file|du|ls|git\s+log|git\s+show|git\s+blame)\b/,
+    /\bxargs\s+-I\s*\S+\s+(wc|cat|head|tail|grep|stat|git)\b/,
+];
+export function isIrreversible(command) {
+    // Safe overrides take priority
+    if (SAFE_OVERRIDES.some((r) => r.test(command)))
+        return false;
+    return IRREVERSIBLE_PATTERNS.some((r) => r.test(command));
+}
+// ── permissions ───────────────────────────────────────────────────────────────
+const DESTRUCTIVE_PATTERNS = [/\brm\b/, /\brmdir\b/, /\btruncate\b/, /\bdrop\s+table\b/i, /\bdelete\s+from\b/i];
+const NETWORK_PATTERNS = [/\bcurl\b/, /\bwget\b/, /\bssh\b/, /\bscp\b/, /\bping\b/, /\bnc\b/, /\bnetcat\b/];
+const SUDO_PATTERNS = [/\bsudo\b/];
+const INSTALL_PATTERNS = [/\bbrew\s+install\b/, /\bnpm\s+install\s+-g\b/, /\bpip\s+install\b/, /\bapt\s+install\b/, /\byum\s+install\b/];
+const WRITE_OUTSIDE_PATTERNS = [/\s(\/etc|\/usr|\/var|\/opt|\/root|~\/[^.])/, />\s*\//];
+export function checkPermissions(command, perms) {
+    if (!perms.destructive && DESTRUCTIVE_PATTERNS.some((r) => r.test(command)))
+        return "destructive commands are disabled";
+    if (!perms.network && NETWORK_PATTERNS.some((r) => r.test(command)))
+        return "network commands are disabled";
+    if (!perms.sudo && SUDO_PATTERNS.some((r) => r.test(command)))
+        return "sudo is disabled";
+    if (!perms.install && INSTALL_PATTERNS.some((r) => r.test(command)))
+        return "package installation is disabled";
+    if (!perms.write_outside_cwd && WRITE_OUTSIDE_PATTERNS.some((r) => r.test(command)))
+        return "writing outside cwd is disabled";
+    return null;
+}
+// ── correction memory ───────────────────────────────────────────────────────
+/** Load past corrections relevant to a prompt — injected as negative examples */
+function loadCorrectionHints(prompt) {
+    try {
+        // Dynamic import to avoid circular deps
+        const { findSimilarCorrections } = require("./sessions-db.js");
+        const corrections = findSimilarCorrections(prompt, 3);
+        if (corrections.length === 0)
+            return "";
+        const lines = corrections.map((c) => `AVOID: "${c.failed_command}" (failed: ${c.error_type}). USE: "${c.corrected_command}" instead.`);
+        return `\n\nLEARNED CORRECTIONS (from past failures):\n${lines.join("\n")}`;
+    }
+    catch {
+        return "";
+    }
+}
+// ── project context (powered by context-hints) ──────────────────────────────
+function detectProjectContext() {
+    const hints = discoverProjectHints(process.cwd());
+    return hints.length > 0 ? `\n\n${formatHints(hints)}` : "";
+}
+// ── system prompt ─────────────────────────────────────────────────────────────
+function buildSystemPrompt(perms, sessionEntries, currentPrompt) {
+    const restrictions = [];
+    if (!perms.destructive)
+        restrictions.push("- NEVER generate commands that delete, remove, or overwrite files/data");
+    if (!perms.network)
+        restrictions.push("- NEVER generate commands that make network requests (curl, wget, ssh, etc.)");
+    if (!perms.sudo)
+        restrictions.push("- NEVER generate commands requiring sudo");
+    if (!perms.write_outside_cwd)
+        restrictions.push("- NEVER write to paths outside the current working directory");
+    if (!perms.install)
+        restrictions.push("- NEVER install packages (brew, npm -g, pip, apt, etc.)");
+    const restrictionBlock = restrictions.length > 0
+        ? `\n\nRESTRICTIONS:\n${restrictions.join("\n")}\nIf restricted, output: BLOCKED: <reason>`
+        : "";
+    let contextBlock = "";
+    if (sessionEntries.length > 0) {
+        const lines = [];
+        for (const e of sessionEntries.slice(-5)) {
+            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${lines.join("\n")}`;
+    }
+    const projectContext = detectProjectContext();
+    // Inject safety hints for the command being generated (AI sees what's risky)
+    const safetyBlock = sessionEntries.length > 0
+        ? (() => {
+            const lastCmd = sessionEntries[sessionEntries.length - 1]?.cmd;
+            if (lastCmd) {
+                const safetyHints = discoverSafetyHints(lastCmd);
+                return safetyHints.length > 0 ? `\n\nLAST COMMAND SAFETY:\n${safetyHints.join("\n")}` : "";
+            }
+            return "";
+        })()
+        : "";
+    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.
+
+RULES:
+- SIMPLICITY FIRST: Use the simplest command that works. Prefer grep | sort | head over 10-pipe chains. Complex pipelines are OK when needed, but NEVER pass file:line output to wc or xargs without cleaning it first.
+- ALWAYS use grep -rn (with -r) when searching directories. NEVER use grep without -r on src/ or any directory.
+- When user refers to items from previous output, use the EXACT names shown (e.g., "feature/auth" not "auth", "open-skills" not "open_skills")
+- When user says "the largest/smallest/first/second", look at the previous output to identify the correct item
+- When user says "them all" or "combine them", refer to items from the most recent command output
+- For "show who changed each line" use git blame, for "show remote urls" use git remote -v
+- For text search in code, use grep -rn, NOT nm or objdump (those are for compiled binaries)
+- On macOS: for memory use vm_stat or top -l 1, for disk use df -h, for processes use ps aux
+- macOS uses BSD tools, NOT GNU. Use: du -d 1 (not --max-depth), ls (not ls --color), sort -r (not sort --reverse), ps aux (not ps --sort)
+- NEVER use grep -P (PCRE). macOS grep has NO -P flag. Use grep -E for extended regex, or sed/awk for complex extraction.
+- NEVER invent commands that don't exist. Stick to standard Unix/macOS commands.
+- NEVER install packages (npx, npm install, pip install, brew install). This is a READ-ONLY terminal.
+- NEVER modify source code (sed -i, codemod, awk with redirect). Only observe, never change.
+- Search src/ directory, NOT dist/ or node_modules/ for code queries.
+- Use exact file paths from the project context below. Do NOT guess paths.
+- For "what would break if I deleted X": use grep -rn "from.*X\\|import.*X\\|require.*X" src/ to find all importers.
+- For "find where X is defined": use grep -rn "export.*function X\\|export.*class X\\|export.*const X" src/
+- For "show me the code of function X": if you know the file, use grep -A 30 "function X" src/file.ts. If not, use grep -rn -A 30 "function X" src/ --include="*.ts"
+- ALWAYS use grep -rn (recursive) when searching directories. NEVER use grep without -r on a directory — it will fail.
+- For conceptual questions about what code does: use cat on the relevant file, the AI summary will explain it.
+- For DESTRUCTIVE requests (delete, remove, install, push): output BLOCKED: <reason>. NEVER try to execute destructive commands.
+
+AST-POWERED QUERIES: For code STRUCTURE questions, use the built-in AST tool instead of grep:
+- "find all exported functions" → terminal symbols src/ (lists all functions, classes, interfaces with line numbers)
+- "show all interfaces" → terminal symbols src/ | grep interface
+- "what does file X export" → terminal symbols src/file.ts
+- "show me the class hierarchy" → terminal symbols src/
+The "terminal symbols" command uses AST parsing (not regex) — it understands TypeScript, Python, Go, Rust code structure.
+For TEXT search (TODO, string matches, imports) → use grep as normal.
+
+COMPOUND QUESTIONS: For questions asking multiple things, prefer ONE command that captures all info. Extract multiple answers from a single output.
+- "how many tests and do they pass" → bun test (extract count AND pass/fail from output)
+- "what files changed and how many lines" → git log --stat -3 (shows files AND line counts)
+- "what version of node and bun" → node -v && bun -v (only use && for trivial non-failing commands)
+NEVER split into separate test runs or expensive commands chained with &&.
+
+BLOCKED ALTERNATIVES: If your preferred command would require installing packages (npx, npm install), ALWAYS try a READ-ONLY alternative:
+- Code quality analysis → grep -rn "TODO\\|FIXME\\|HACK\\|XXX" src/
+- Linting → check if "lint" or "typecheck" exists in package.json scripts, run that
+- Security scan → grep -rn "eval\\|exec\\|spawn\\|password\\|secret" src/
+- Dependency audit → cat package.json | grep -A 50 dependencies
+- Test coverage → bun test --coverage (or npm run test:coverage if available)
+NEVER give up. NEVER output BLOCKED for analysis questions. Always try a grep/find/cat/wc/awk read-only alternative.
+- Cyclomatic complexity → grep -rn "if\\|else\\|for\\|while\\|switch\\|case\\|catch\\|&&\\|||" src/ --include="*.ts" | wc -l
+- Unused exports → grep -rn "export function\|export const\|export class" src/ --include="*.ts" | sed 's/.*export [a-z]* //' | sed 's/[(<:].*//' | sort -u
+- Dead code → for each exported name, grep -rn "name" src/ --include="*.ts" | wc -l (if only 1 match = unused)
+- Dependency graph → grep -rn "from " src/ --include="*.ts" | sed 's/:.*from "/→/' | sed 's/".*//' | sort -u
+- Most parameters → grep -rn "function " src/ --include="*.ts" | awk -F'[()]' '{print gsub(/,/,",",$2)+1, $0}' | sort -nr | head -10
+ALWAYS try a heuristic shell approach before giving up. NEVER say BLOCKED for analysis questions.
+
+SEMANTIC MAPPING: When the user references a concept, search the file tree for RELATED terms:
+- Look at directory names: src/agent/ likely contains "agentic" code
+- Look at file names: lazy-executor.ts likely handles "lazy mode"
+- When uncertain: grep -rn "keyword" src/ --include="*.ts" -l (list matching files)
+
+ACTION vs CONCEPTUAL: If the prompt starts with "run", "execute", "check", "test", "build", "show output of" — ALWAYS generate an executable command. NEVER read README for action requests. Only read docs for "explain why", "what does X mean", "how was X designed".
+
+EXISTENCE CHECKS: If the prompt starts with "is there", "does this have", "do we have", "does X exist" — NEVER run/start/launch anything. Use ls, find, or test -d to CHECK existence. These are READ-ONLY questions.
+
+MONOREPO: If the project context says "MONOREPO", search packages/ or apps/ NOT src/. Use: grep -rn "pattern" packages/ --include="*.ts". For specific packages, use packages/PKGNAME/src/.
+cwd: ${process.cwd()}
+shell: zsh / macOS${projectContext}${safetyBlock}${restrictionBlock}${contextBlock}${currentPrompt ? loadCorrectionHints(currentPrompt) : ""}`;
+}
+// ── streaming translate ───────────────────────────────────────────────────────
+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, sessionEntries, nl);
+    let text;
+    if (onToken) {
+        text = await provider.stream(nl, { model, maxTokens: 256, system }, {
+            onToken: (partial) => onToken(partial),
+        });
+    }
+    else {
+        text = await provider.complete(nl, { model, maxTokens: 256, system });
+    }
+    if (text.startsWith("BLOCKED:"))
+        throw new Error(text);
+    // Strip AI reasoning — extract ONLY the shell command (first line)
+    let cleaned = text.trim();
+    // Remove ALL markdown code blocks and their content markers
+    cleaned = cleaned.replace(/```(?:bash|sh|shell)?\n?/g, "").replace(/```/g, "");
+    // Split into lines and find the FIRST one that looks like a SHELL COMMAND
+    const lines = cleaned.split("\n");
+    let command = "";
+    for (const line of lines) {
+        const t = line.trim();
+        if (!t)
+            continue;
+        // Skip lines that are clearly English prose, not commands
+        if (/^(Based on|I |This |The |Let me|Here|Note:|Since|Looking|To |However|BLOCKED:|If |You |We |For |It |A |An |That )/.test(t))
+            continue;
+        if (/^[A-Z][a-z].*[.;:!?,]/.test(t))
+            continue; // English sentence with punctuation anywhere
+        if (t.split(" ").length > 15 && !/[|&;><$]/.test(t))
+            continue; // Long line without shell operators = prose
+        // Must start with a plausible command character (lowercase, /, ., $, or common tool)
+        if (/^[a-z./$~(]/.test(t) || /^[A-Z]+[_=]/.test(t)) {
+            command = t;
+            break;
+        }
+    }
+    cleaned = command || lines[0]?.trim() || cleaned;
+    cacheSet(nl, cleaned);
+    return cleaned;
+}
+// ── prefetch ──────────────────────────────────────────────────────────────────
+export function prefetchNext(lastNl, perms, sessionEntries) {
+    if (sessionEntries.length === 0 && cacheGet(lastNl))
+        return;
+    translateToCommand(lastNl, perms, sessionEntries).catch(() => { });
+}
+// ── explain ───────────────────────────────────────────────────────────────────
+export async function explainCommand(command) {
+    const provider = getProvider();
+    const routing = pickModel("explain"); // simple = fast model
+    return provider.complete(command, {
+        model: routing.fast,
+        maxTokens: 128,
+        system: "Explain what this shell command does in one plain English sentence. No markdown, no code blocks.",
+    });
+}
+// ── auto-fix ──────────────────────────────────────────────────────────────────
+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, sessionEntries, originalNl),
+    });
+    if (text.startsWith("BLOCKED:"))
+        throw new Error(text);
+    return text;
+}
+// ── summarize output (for MCP/agent use) ──────────────────────────────────────
+export async function summarizeOutput(command, output, maxTokens = 200) {
+    const provider = getProvider();
+    const routing = pickModel("summarize");
+    return provider.complete(`Command: ${command}\nOutput:\n${output}\n\nSummarize this output concisely for an AI agent. Focus on: status, key results, errors. Be terse.`, {
+        model: routing.fast,
+        maxTokens,
+        system: "You summarize command output for AI agents. Be extremely concise. Return structured info. No prose.",
+    });
+}

package/dist/cache.js

@@ -0,0 +1,41 @@
+// In-memory LRU cache + disk persistence for command translations
+import { existsSync, readFileSync, writeFileSync } from "fs";
+import { homedir } from "os";
+import { join } from "path";
+const CACHE_FILE = join(homedir(), ".terminal", "cache.json");
+const MAX_ENTRIES = 500;
+let mem = {};
+export function loadCache() {
+    if (!existsSync(CACHE_FILE))
+        return;
+    try {
+        mem = JSON.parse(readFileSync(CACHE_FILE, "utf8"));
+    }
+    catch { }
+}
+function persistCache() {
+    try {
+        writeFileSync(CACHE_FILE, JSON.stringify(mem));
+    }
+    catch { }
+}
+/** Normalize a natural language query for cache lookup */
+export function normalizeNl(nl) {
+    return nl
+        .toLowerCase()
+        .trim()
+        .replace(/[^a-z0-9\s]/g, "") // strip punctuation
+        .replace(/\s+/g, " ");
+}
+export function cacheGet(nl) {
+    return mem[normalizeNl(nl)] ?? null;
+}
+export function cacheSet(nl, command) {
+    const key = normalizeNl(nl);
+    // evict oldest if full
+    const keys = Object.keys(mem);
+    if (keys.length >= MAX_ENTRIES)
+        delete mem[keys[0]];
+    mem[key] = command;
+    persistCache();
+}

package/dist/cli.js

@@ -361,17 +361,52 @@ else if (args[0] === "repo") {
 else if (args[0] === "symbols" && args[1]) {
     const { extractSymbolsFromFile } = await import("./search/semantic.js");
     const { resolve } = await import("path");
-    const filePath = resolve(args[1]);
-    const symbols = extractSymbolsFromFile(filePath);
-    if (symbols.length === 0) {
-        console.log("No symbols found.");
+    const { statSync, readdirSync } = await import("fs");
+    const target = resolve(args[1]);
+    const filter = args[2]; // optional: grep-like filter on symbol name
+    // Support directories — recurse and extract symbols from all source files
+    const files = [];
+    try {
+        if (statSync(target).isDirectory()) {
+            const walk = (dir) => {
+                for (const entry of readdirSync(dir, { withFileTypes: true })) {
+                    if (entry.name.startsWith(".") || entry.name === "node_modules" || entry.name === "dist")
+                        continue;
+                    const full = resolve(dir, entry.name);
+                    if (entry.isDirectory())
+                        walk(full);
+                    else if (/\.(ts|tsx|py|go|rs)$/.test(entry.name) && !/\.(test|spec)\.\w+$/.test(entry.name))
+                        files.push(full);
+                }
+            };
+            walk(target);
+        }
+        else {
+            files.push(target);
+        }
     }
-    else {
-        for (const s of symbols) {
+    catch {
+        files.push(target);
+    }
+    let totalSymbols = 0;
+    for (const file of files) {
+        const symbols = extractSymbolsFromFile(file);
+        const filtered = filter ? symbols.filter(s => s.name.toLowerCase().includes(filter.toLowerCase()) || s.kind.toLowerCase().includes(filter.toLowerCase())) : symbols;
+        if (filtered.length === 0)
+            continue;
+        const relPath = file.replace(process.cwd() + "/", "");
+        if (files.length > 1)
+            console.log(`\n${relPath}:`);
+        for (const s of filtered) {
             const exp = s.exported ? "⬡" : "·";
             console.log(`  ${exp} ${s.kind.padEnd(10)} L${String(s.line).padStart(4)}  ${s.name}`);
         }
+        totalSymbols += filtered.length;
     }
+    if (totalSymbols === 0)
+        console.log("No symbols found.");
+    else if (files.length > 1)
+        console.log(`\n${totalSymbols} symbols across ${files.length} files`);
 }
 // ── History command ──────────────────────────────────────────────────────────
 else if (args[0] === "history") {
@@ -621,7 +656,7 @@ else if (args.length > 0) {
     catch (e) {
         // Empty result (grep exit 1 = no matches) — not a real error
         const errStdout = e.stdout?.toString() ?? "";
-        const errStderr = e.stderr?.toString() ?? "";
+        let errStderr = e.stderr?.toString() ?? "";
         if (e.status === 1 && !errStdout.trim() && !errStderr.trim()) {
             // Empty result — retry with broader scope before giving up
             if (!actualCmd.includes("#(broadened)")) {
@@ -643,24 +678,37 @@ else if (args.length > 0) {
             console.log(`No results found for: ${prompt}`);
             process.exit(0);
         }
-        // Auto-retry: if command failed (exit 2+), ask AI for a simpler alternative
-        if (e.status >= 2 && !actualCmd.includes("(retry)")) {
-            try {
-                const retryCmd = await translateToCommand(`${prompt} (The previous command failed with: ${errStderr.slice(0, 200)}. Try a SIMPLER approach. Use basic commands only.)`, perms, []);
-                if (retryCmd && !isIrreversible(retryCmd) && !checkPermissions(retryCmd, perms)) {
-                    console.error(`[open-terminal] retrying: $ ${retryCmd}`);
-                    const retryResult = execSync(retryCmd + " #(retry)", { encoding: "utf8", maxBuffer: 10 * 1024 * 1024, cwd: process.cwd() });
+        // 3-retry learning loop: each attempt learns from the previous failure
+        if (e.status >= 2) {
+            const retryStrategies = [
+                // Attempt 2: inject error context
+                `${prompt} (Command "${actualCmd}" failed with: ${errStderr.slice(0, 300)}. Fix this specific error. Keep the approach but correct the issue.)`,
+                // Attempt 3: inject corrections + force simplicity
+                `${prompt} (TWO commands already failed for this query. Use the ABSOLUTE SIMPLEST approach: basic grep -rn, find, wc -l, cat. No awk, no xargs, no subshells. Must work on macOS BSD.)`,
+            ];
+            for (let attempt = 0; attempt < retryStrategies.length; attempt++) {
+                try {
+                    const retryCmd = await translateToCommand(retryStrategies[attempt], perms, []);
+                    if (!retryCmd || retryCmd === actualCmd || isIrreversible(retryCmd) || checkPermissions(retryCmd, perms))
+                        continue;
+                    console.error(`[open-terminal] retry ${attempt + 2}/3: $ ${retryCmd}`);
+                    const retryResult = execSync(retryCmd + ` #(retry${attempt + 2})`, { encoding: "utf8", maxBuffer: 10 * 1024 * 1024, cwd: process.cwd() });
                     const retryClean = stripNoise(stripAnsi(retryResult)).cleaned;
                     if (retryClean.length > 5) {
-                        // Record the correction so we learn from it
+                        // Record correction — AI learns for next time
                         recordCorrection(prompt, actualCmd, errStderr.slice(0, 500), retryCmd, true);
                         const processed = await processOutput(retryCmd, retryClean, prompt);
                         console.log(processed.aiProcessed ? processed.summary : retryClean);
                         process.exit(0);
                     }
                 }
+                catch (retryErr) {
+                    // This attempt also failed — record it and try next strategy
+                    const retryStderr = retryErr.stderr?.toString() ?? "";
+                    errStderr = retryStderr; // update for next attempt's context
+                    continue;
+                }
             }
-            catch { /* retry also failed, fall through */ }
         }
         // Combine stdout+stderr and try AI answer framing (for audit/lint/test commands)
         const combined = errStderr && errStdout.includes(errStderr.trim()) ? errStdout : errStdout + errStderr;

package/dist/command-rewriter.js

@@ -0,0 +1,64 @@
+// Command rewriter — auto-optimize commands to produce less output
+// Only rewrites when semantic result is identical
+const rules = [
+    // find | grep -v node_modules → find -not -path
+    {
+        pattern: /find\s+(\S+)\s+(.*?)\|\s*grep\s+-v\s+node_modules/,
+        rewrite: (m, cmd) => cmd.replace(m[0], `find ${m[1]} ${m[2]}-not -path '*/node_modules/*'`),
+        reason: "avoid pipe, filter in-kernel",
+    },
+    // cat file | grep X → grep X file
+    {
+        pattern: /cat\s+(\S+)\s*\|\s*grep\s+(.*)/,
+        rewrite: (m) => `grep ${m[2]} ${m[1]}`,
+        reason: "useless cat",
+    },
+    // find without node_modules exclusion → add it
+    {
+        pattern: /^find\s+\.\s+(.*)(?!.*node_modules)/,
+        rewrite: (m, cmd) => {
+            if (cmd.includes("node_modules") || cmd.includes("-not -path"))
+                return cmd;
+            return cmd.replace(/^find\s+\.\s+/, "find . -not -path '*/node_modules/*' -not -path '*/.git/*' ");
+        },
+        reason: "auto-exclude node_modules and .git",
+    },
+    // git log without limit → add --oneline -20
+    {
+        pattern: /^git\s+log\s*$/,
+        rewrite: () => "git log --oneline -20",
+        reason: "prevent unbounded log output",
+    },
+    // git diff without stat → add --stat for overview
+    {
+        pattern: /^git\s+diff\s*$/,
+        rewrite: () => "git diff --stat",
+        reason: "stat overview is usually sufficient",
+    },
+    // npm ls without depth → add --depth=0
+    {
+        pattern: /^npm\s+ls\s*$/,
+        rewrite: () => "npm ls --depth=0",
+        reason: "full tree is massive, top-level usually enough",
+    },
+    // ps aux without filter → sort by memory and head (macOS compatible)
+    {
+        pattern: /^ps\s+aux\s*$/,
+        rewrite: () => "ps aux | sort -k4 -rn | head -20",
+        reason: "full process list is noise, show top consumers",
+    },
+];
+/** Rewrite a command to produce less output */
+export function rewriteCommand(cmd) {
+    const trimmed = cmd.trim();
+    for (const rule of rules) {
+        const match = trimmed.match(rule.pattern);
+        if (match) {
+            const rewritten = rule.rewrite(match, trimmed);
+            if (rewritten !== trimmed) {
+                return { original: trimmed, rewritten, changed: true, reason: rule.reason };
+            }
+        }
+    }
+    return { original: trimmed, rewritten: trimmed, changed: false };
+}

package/dist/command-validator.js

@@ -0,0 +1,86 @@
+// Command validator — catch invalid commands BEFORE executing
+// Prevents shell errors from hallucinated flags, wrong paths, bad syntax
+import { existsSync } from "fs";
+import { join } from "path";
+/** Extract file paths referenced in a command */
+function extractPaths(command) {
+    const paths = [];
+    // Match quoted paths
+    const quoted = command.match(/["']([^"']+\.\w+)["']/g);
+    if (quoted)
+        paths.push(...quoted.map(q => q.replace(/["']/g, "")));
+    // Match unquoted paths with extensions or directory separators
+    const tokens = command.split(/\s+/);
+    for (const t of tokens) {
+        if (t.includes("/") && !t.startsWith("-") && !t.startsWith("|") && !t.startsWith("&")) {
+            // Clean shell operators from end
+            const clean = t.replace(/[;|&>]+$/, "");
+            if (clean && !clean.startsWith("-"))
+                paths.push(clean);
+        }
+    }
+    return [...new Set(paths)];
+}
+/** Check for obviously broken shell syntax */
+function checkSyntax(command) {
+    const issues = [];
+    // Unmatched quotes
+    const singleQuotes = (command.match(/'/g) || []).length;
+    const doubleQuotes = (command.match(/"/g) || []).length;
+    if (singleQuotes % 2 !== 0)
+        issues.push("unmatched single quote");
+    if (doubleQuotes % 2 !== 0)
+        issues.push("unmatched double quote");
+    // Unmatched parentheses
+    const openParens = (command.match(/\(/g) || []).length;
+    const closeParens = (command.match(/\)/g) || []).length;
+    if (openParens !== closeParens)
+        issues.push("unmatched parentheses");
+    // Empty pipe targets
+    if (/\|\s*$/.test(command))
+        issues.push("pipe with no target");
+    if (/^\s*\|/.test(command))
+        issues.push("pipe with no source");
+    return issues;
+}
+/** Validate a command before execution */
+export function validateCommand(command, cwd) {
+    const issues = [];
+    // Check syntax
+    issues.push(...checkSyntax(command));
+    // Check file paths exist
+    const paths = extractPaths(command);
+    for (const p of paths) {
+        const fullPath = p.startsWith("/") ? p : join(cwd, p);
+        if (p.includes("*") || p.includes("?"))
+            continue; // skip globs
+        if (p.startsWith("-"))
+            continue; // skip flags
+        if ([".", "..", "/", "~"].includes(p))
+            continue; // skip special
+        if (!existsSync(fullPath) && !existsSync(p)) {
+            // Only flag source file paths, not output paths
+            if (/\.(ts|tsx|js|jsx|json|md|yaml|yml|py|go|rs)$/.test(p)) {
+                issues.push(`file not found: ${p}`);
+            }
+        }
+    }
+    // Check for common GNU flags on macOS
+    const gnuFlags = command.match(/--max-depth|--color=|--sort=|--field-type|--no-deps/g);
+    if (gnuFlags) {
+        issues.push(`GNU flag on macOS: ${gnuFlags.join(", ")}`);
+    }
+    // Complexity guard — extreme pipe chains are fragile
+    const pipeCount = (command.match(/\|/g) || []).length;
+    if (pipeCount > 7) {
+        issues.push(`too complex: ${pipeCount} pipes — simplify`);
+    }
+    // grep -P (PCRE) doesn't exist on macOS
+    if (/grep\s+.*-[a-zA-Z]*P/.test(command)) {
+        issues.push("grep -P (PCRE) not available on macOS — use grep -E");
+    }
+    return {
+        valid: issues.length === 0,
+        issues,
+    };
+}