@sidhxntt/token-squasher 1.0.0

package/index.js

@@ -0,0 +1,210 @@
+#!/usr/bin/env node
+
+import Anthropic from "@anthropic-ai/sdk";
+import readline from "readline";
+
+// ─── ANSI Colors ──────────────────────────────────────────────────────────────
+const c = {
+  reset: "\x1b[0m", bold: "\x1b[1m", dim: "\x1b[2m",
+  red: "\x1b[38;5;203m", yellow: "\x1b[38;5;220m", green: "\x1b[38;5;114m",
+  cyan: "\x1b[38;5;117m", magenta: "\x1b[38;5;213m", gray: "\x1b[38;5;245m",
+};
+
+const bold    = (s) => `${c.bold}${s}${c.reset}`;
+const dim     = (s) => `${c.dim}${s}${c.reset}`;
+const red     = (s) => `${c.red}${s}${c.reset}`;
+const yellow  = (s) => `${c.yellow}${s}${c.reset}`;
+const green   = (s) => `${c.green}${s}${c.reset}`;
+const cyan    = (s) => `${c.cyan}${s}${c.reset}`;
+const magenta = (s) => `${c.magenta}${s}${c.reset}`;
+const gray    = (s) => `${c.gray}${s}${c.reset}`;
+
+// ─── Token estimate ───────────────────────────────────────────────────────────
+const estimateTokens = (t) => Math.ceil(t.length / 4);
+
+// ─── Spinner ──────────────────────────────────────────────────────────────────
+function makeSpinner(msg) {
+  const frames = ["⠋","⠙","⠹","⠸","⠼","⠴","⠦","⠧","⠇","⠏"];
+  let i = 0;
+  const timer = setInterval(() => {
+    process.stdout.write(`\r  ${cyan(frames[i++ % frames.length])} ${dim(msg)}`);
+  }, 80);
+  return { stop: () => { clearInterval(timer); process.stdout.write("\r\x1b[2K"); } };
+}
+
+// ─── Core squash ──────────────────────────────────────────────────────────────
+async function squashPrompt(promptText) {
+  const client = new Anthropic();
+  const resp = await client.messages.create({
+    model: "claude-sonnet-4-20250514",
+    max_tokens: 4096,
+    system: `You are a prompt compression expert. Convert verbose system prompts into ultra-dense, instruction-only format.
+
+RULES:
+- Strip ALL filler: pleasantries, preambles, rationale, meta-commentary
+- Imperative verbs only: "Return JSON" not "You should return JSON"
+- Collapse repetition: one rule, stated once
+- Symbols: "→" "w/" "&"
+- Numbered lists > prose paragraphs
+- Remove examples unless the ONLY way to convey a rule
+- Preserve ALL logic, constraints, edge cases, behavioral rules — never lose one, never add one
+
+Return ONLY:
+<reasoning>[max 8 bullets of what you compressed and how]</reasoning>
+<squashed>[compressed prompt, raw text only]</squashed>`,
+    messages: [{ role: "user", content: `Compress this system prompt:\n\n${promptText}` }],
+  });
+
+  const raw = resp.content[0].text;
+  const rM  = raw.match(/<reasoning>([\s\S]*?)<\/reasoning>/);
+  const sM  = raw.match(/<squashed>([\s\S]*?)<\/squashed>/);
+  if (!sM) throw new Error("No <squashed> block returned. Try again.");
+  return { squashed: sM[1].trim(), reasoning: rM ? rM[1].trim() : null };
+}
+
+// ─── Stats ────────────────────────────────────────────────────────────────────
+function printStats(orig, sq) {
+  const o = estimateTokens(orig), s = estimateTokens(sq);
+  const pct = (((o - s) / o) * 100).toFixed(1);
+  const filled = Math.round(pct / 100 * 24);
+  const bar = `${c.green}${"█".repeat(filled)}${c.reset}${gray("░".repeat(24 - filled))}`;
+  console.log();
+  console.log(`  ${bold("───── TOKEN STATS ─────")}`);
+  console.log(`  ${gray("Before")}   ${yellow(o.toLocaleString())} tokens  ${gray(`(${orig.length.toLocaleString()} chars)`)}`);
+  console.log(`  ${gray("After")}    ${green(s.toLocaleString())} tokens  ${gray(`(${sq.length.toLocaleString()} chars)`)}`);
+  console.log(`  ${gray("Crushed")}  ${bold(green(pct + "%"))} ${bar}`);
+  console.log(`  ${bold("───────────────────────")}`);
+}
+
+// ─── Welcome ──────────────────────────────────────────────────────────────────
+function printWelcome() {
+  console.clear();
+  console.log();
+  console.log(`  ${bold(cyan("token-squasher"))} ${gray("v2.0.0")}`);
+  console.log(`  ${gray("Compress verbose system prompts. Keep logic. Crush tokens.")}`);
+  console.log();
+
+  const hasKey = !!process.env.ANTHROPIC_API_KEY;
+  if (hasKey) {
+    console.log(`  ${green("✔")} ${dim("ANTHROPIC_API_KEY detected")}`);
+  } else {
+    console.log(`  ${red("✖")} ${bold("ANTHROPIC_API_KEY not set")}  →  ${cyan("export ANTHROPIC_API_KEY=sk-ant-...")}`);
+  }
+
+  console.log();
+  console.log(`  ${dim("Paste your system prompt and press")} ${bold("Enter twice")} ${dim("to squash.")}`);
+  console.log(`  ${dim("Commands:")} ${cyan(":stats")} ${dim("toggle stats")}  ${cyan(":verbose")} ${dim("toggle reasoning")}  ${cyan(":quit")} ${dim("exit")}`);
+  console.log(`  ${dim("─".repeat(58))}`);
+  console.log();
+}
+
+// ─── REPL ─────────────────────────────────────────────────────────────────────
+async function runRepl() {
+  printWelcome();
+
+  let showStats   = false;
+  let showVerbose = false;
+  let session     = 0;
+  let buffer      = [];
+
+  const rl = readline.createInterface({
+    input: process.stdin,
+    output: process.stdout,
+    terminal: true,
+    prompt: `  ${cyan("›")} `,
+  });
+
+  const promptNext = () => {
+    session++;
+    console.log(`  ${dim("─".repeat(58))}`);
+    console.log(`  ${dim(`Prompt #${session} — paste prompt then`)} ${bold("Enter twice")} ${dim("to squash")}  ${gray(`[stats:${showStats?"on":"off"} verbose:${showVerbose?"on":"off"}]`)}`);
+    console.log();
+    rl.prompt();
+  };
+
+  const flush = async () => {
+    const source = buffer.join("\n").trim();
+    buffer = [];
+    if (!source) return;
+
+    // ── Commands ──
+    if (source === ":quit" || source === ":q") {
+      console.log(`\n  ${gray("bye. tokens crushed. 👋\n")}`);
+      process.exit(0);
+    }
+    if (source === ":clear") { printWelcome(); promptNext(); return; }
+    if (source === ":stats")   { showStats   = !showStats;   console.log(`  ${dim(`stats → ${showStats   ? green("on") : "off"}`)}\n`); promptNext(); return; }
+    if (source === ":verbose") { showVerbose = !showVerbose; console.log(`  ${dim(`verbose → ${showVerbose ? green("on") : "off"}`)}\n`); promptNext(); return; }
+
+    // ── Squash ──
+    console.log();
+    const spin = makeSpinner("Squashing tokens…");
+    let result;
+    try {
+      result = await squashPrompt(source);
+    } catch (err) {
+      spin.stop();
+      console.log(`  ${red("✖")} ${err.message}\n`);
+      promptNext();
+      return;
+    }
+    spin.stop();
+
+    if (showVerbose && result.reasoning) {
+      console.log(`  ${bold(magenta("REASONING"))}`);
+      result.reasoning.split("\n").forEach((l) => console.log(`  ${gray(l)}`));
+    }
+
+    if (showStats) printStats(source, result.squashed);
+
+    console.log();
+    console.log(`  ${bold(green("─── squashed ") + "─".repeat(45))}`);
+    console.log();
+    console.log(result.squashed);
+    console.log();
+    console.log(`  ${dim("─".repeat(58))}`);
+
+    promptNext();
+  };
+
+  promptNext();
+
+  rl.on("line", (line) => {
+    // Empty line after content = submit
+    if (line.trim() === "" && buffer.length > 0) {
+      rl.pause();
+      flush().then(() => rl.resume()).catch((e) => {
+        console.error(red(`  ✖ ${e.message}`));
+        rl.resume();
+      });
+    } else {
+      buffer.push(line);
+      rl.prompt();
+    }
+  });
+
+  rl.on("close", () => {
+    if (buffer.length > 0) {
+      flush().then(() => { console.log(`\n  ${gray("bye.\n")}`); process.exit(0); });
+    } else {
+      console.log(`\n  ${gray("bye.\n")}`);
+      process.exit(0);
+    }
+  });
+
+  rl.on("SIGINT", () => {
+    if (buffer.length > 0) {
+      buffer = [];
+      console.log(`\n  ${yellow("⚠")}  ${gray("Input cleared.")}\n`);
+      promptNext();
+    } else {
+      console.log(`\n  ${gray("bye.\n")}`);
+      process.exit(0);
+    }
+  });
+}
+
+runRepl().catch((e) => {
+  console.error(red(`  ✖ ${e.message}`));
+  process.exit(1);
+});

package/package.json

@@ -0,0 +1,30 @@
+{
+  "name": "@sidhxntt/token-squasher",
+  "version": "1.0.0",
+  "description": "Compress verbose system prompts into dense, token-efficient instructions without losing logic",
+  "type": "module",
+  "bin": {
+    "@sidhxntt/token-squasher": "./index.js"
+  },
+  "main": "index.js",
+  "scripts": {
+    "start": "node index.js"
+  },
+  "dependencies": {
+    "@anthropic-ai/sdk": "^0.52.0"
+  },
+  "engines": {
+    "node": ">=18.0.0"
+  },
+  "keywords": [
+    "prompt",
+    "token",
+    "compression",
+    "llm",
+    "anthropic",
+    "claude",
+    "ai"
+  ],
+  "author": "",
+  "license": "MIT"
+}