promptpilot 0.1.5 → 0.1.7

package/README.md

@@ -75,6 +75,43 @@ ollama pull qwen2.5:3b
 ollama pull phi3:mini
 ```
 
+## Custom local compressor model
+
+PromptPilot ships a `Modelfile` that defines `promptpilot-compressor`, a text-only compression model built on top of `qwen2.5:3b`. It is tuned to output only the compressed prompt with no reasoning, analysis, or commentary.
+
+Build and verify it:
+
+```bash
+ollama pull qwen2.5:3b
+ollama create promptpilot-compressor -f ./Modelfile
+ollama run promptpilot-compressor "explain recursion simply"
+```
+
+Use it via the CLI after installing from npm:
+
+```bash
+# Plain output — pipe directly into Claude
+promptpilot optimize "help me refactor this auth middleware" \
+  --model promptpilot-compressor \
+  --preset code \
+  --plain
+
+# JSON output with debug info
+promptpilot optimize "help me refactor this auth middleware" \
+  --model promptpilot-compressor \
+  --preset code \
+  --json --debug
+
+# With session memory, piped into Claude
+promptpilot optimize "continue the refactor" \
+  --model promptpilot-compressor \
+  --session repo-refactor \
+  --save-context \
+  --plain | claude
+```
+
+`promptpilot-compressor` outputs plain text rather than JSON. PromptPilot detects this automatically and falls back to text-only mode, stripping any reasoning leakage before using the output. Explicit `--model` always takes priority over automatic local model selection.
+
 ## Core behavior
 
 PromptPilot has two distinct routing layers.

package/dist/cli.js

@@ -3,6 +3,7 @@
 // src/cli.ts
 import { readFileSync, realpathSync } from "fs";
 import { fileURLToPath } from "url";
+import { execSync } from "child_process";
 
 // src/errors.ts
 var InvalidPromptError = class extends Error {
@@ -353,6 +354,17 @@ var OllamaClient = class {
     }
     throw new OllamaUnavailableError("Ollama returned JSON that could not be parsed.");
   }
+  async generateJsonWithTextFallback(options, textFallbackHandler) {
+    try {
+      return await this.generateJson(options);
+    } catch {
+      const raw = await this.generate({
+        ...options,
+        format: void 0
+      });
+      return textFallbackHandler(raw);
+    }
+  }
 };
 
 // src/core/systemPrompt.ts
@@ -1046,6 +1058,11 @@ var PromptOptimizer = class {
         contextSummary: relevantContext.summary
       });
     }
+    const originalPromptTokens = this.estimator.estimateText(originalPrompt);
+    const promptCompressionSavings = Math.max(0, originalPromptTokens - estimatedTokensAfter.prompt);
+    const contextCompressionSavings = Math.max(0, estimatedTokensBefore.context - estimatedTokensAfter.context);
+    const wrapperOverhead = Math.max(0, estimatedTokensAfter.total - (estimatedTokensAfter.prompt + estimatedTokensAfter.context));
+    const tokenSavings = promptCompressionSavings + contextCompressionSavings;
     return {
       originalPrompt,
       optimizedPrompt,
@@ -1054,7 +1071,10 @@ var PromptOptimizer = class {
       contextSummary: relevantContext.summary,
       estimatedTokensBefore,
       estimatedTokensAfter,
-      tokenSavings: Math.max(0, estimatedTokensBefore.total - estimatedTokensAfter.total),
+      tokenSavings,
+      promptCompressionSavings,
+      contextCompressionSavings,
+      wrapperOverhead,
       mode,
       provider,
       model,
@@ -1121,29 +1141,52 @@ Mode: Ultra compression. Minimize tokens aggressively.` : getOptimizationSystemP
       let optimizedPrompt = "";
       let responseChanges = [];
       let responseWarnings = [];
-      try {
-        const response = await this.client.generateJson({
-          systemPrompt,
-          prompt: optimizationPrompt,
-          timeoutMs,
-          model: options.model,
-          temperature: this.config.temperature,
-          format: "json"
-        });
-        optimizedPrompt = normalizeWhitespace(response.optimizedPrompt ?? "");
-        responseChanges = response.changes ?? [];
-        responseWarnings = response.warnings ?? [];
-      } catch {
-        const raw = await this.client.generate({
-          systemPrompt,
-          prompt: optimizationPrompt,
-          timeoutMs,
-          model: options.model,
-          temperature: this.config.temperature
-        });
-        optimizedPrompt = sanitizeTextOptimizationOutput(raw);
-        responseChanges = [`Applied text-only Ollama optimization with ${options.model}.`];
-      }
+      const generateWithFallback = this.client.generateJsonWithTextFallback ? async () => {
+        const response2 = await this.client.generateJsonWithTextFallback(
+          {
+            systemPrompt,
+            prompt: optimizationPrompt,
+            timeoutMs,
+            model: options.model,
+            temperature: this.config.temperature,
+            format: "json"
+          },
+          (text) => ({
+            optimizedPrompt: sanitizeTextOptimizationOutput(text),
+            changes: [`Applied text-only Ollama optimization with ${options.model}.`],
+            warnings: []
+          })
+        );
+        return response2;
+      } : async () => {
+        try {
+          return await this.client.generateJson({
+            systemPrompt,
+            prompt: optimizationPrompt,
+            timeoutMs,
+            model: options.model,
+            temperature: this.config.temperature,
+            format: "json"
+          });
+        } catch {
+          const raw = await this.client.generate({
+            systemPrompt,
+            prompt: optimizationPrompt,
+            timeoutMs,
+            model: options.model,
+            temperature: this.config.temperature
+          });
+          return {
+            optimizedPrompt: sanitizeTextOptimizationOutput(raw),
+            changes: [`Applied text-only Ollama optimization with ${options.model}.`],
+            warnings: []
+          };
+        }
+      };
+      const response = await generateWithFallback();
+      optimizedPrompt = normalizeWhitespace(response.optimizedPrompt ?? "");
+      responseChanges = response.changes ?? [];
+      responseWarnings = response.warnings ?? [];
       if (!optimizedPrompt) {
         return {
           optimizedPrompt: preprocessedPrompt,
@@ -1567,14 +1610,6 @@ ${contextBlock}`);
   if (constraints.length > 0) {
     sections.push(`Constraints:
 - ${constraints.join("\n- ")}`);
-  }
-  if (isCodeFirstRequest(input.input)) {
-    sections.push(`Execution loop:
-- Inspect the relevant files and current behavior.
-- Plan the smallest safe next step.
-- Act with minimal, reversible changes.
-- Test or validate the result.
-- Reflect on gaps or risks, then repeat.`);
   }
   const desiredOutput = [
     input.routingDecision.selectedTarget ? `Selected target: ${formatTargetLabel(input.routingDecision.selectedTarget)}` : input.input.targetModel ? `Target model: ${input.input.targetModel}` : "Target model: claude",
@@ -1685,14 +1720,16 @@ function sanitizeTextOptimizationOutput(raw) {
   if (!normalized) {
     return "";
   }
-  if (!containsReasoningLeak(normalized)) {
-    return stripWrappingQuotes(normalized);
+  let cleaned = normalized.replace(/<think>[\s\S]*?<\/think>/gi, "").replace(/<reasoning>[\s\S]*?<\/reasoning>/gi, "").replace(/<analysis>[\s\S]*?<\/analysis>/gi, "").replace(/^(thinking|thinking process|analysis|critique|attempt|final decision|role|task|guidelines)[:=]?[\s\S]*?(?=\n\n|\n[A-Z]|$)/gim, "");
+  if (!containsReasoningLeak(cleaned)) {
+    return stripWrappingQuotes(cleaned);
   }
-  const candidates = raw.split(/\n{2,}/).map((chunk) => stripWrappingQuotes(normalizeWhitespace(chunk))).filter(Boolean).filter((chunk) => !containsReasoningLeak(chunk)).filter((chunk) => !/^(role|task|guidelines|thinking|thinking process|attempt|critique|final decision|analysis)\b/i.test(chunk)).filter((chunk) => !/^[-*]\s/.test(chunk)).filter((chunk) => !/^\d+\.\s/.test(chunk));
-  return candidates.at(-1) ?? stripWrappingQuotes(normalized);
+  const candidates = cleaned.split(/\n{2,}/).map((chunk) => stripWrappingQuotes(normalizeWhitespace(chunk))).filter(Boolean).filter((chunk) => !containsReasoningLeak(chunk)).filter((chunk) => !/^(role|task|guidelines|thinking|thinking process|attempt|critique|final decision|analysis)\b/i.test(chunk)).filter((chunk) => chunk.length > 10);
+  const selected = candidates.reduce((a, b) => a.length > b.length ? a : b, "");
+  return selected || stripWrappingQuotes(normalized);
 }
 function containsReasoningLeak(text) {
-  return /(thinking process|analyze the request|drafting the optimized prompt|critique \d|attempt \d|final decision)/i.test(text);
+  return /(thinking process|analyze the request|drafting the optimized prompt|critique \d|attempt \d|final decision|^thinking:|^analysis:|<think>|<reasoning>|<analysis>)/i.test(text);
 }
 function stripWrappingQuotes(text) {
   return text.replace(/^["'`]+|["'`]+$/g, "").trim();
@@ -1942,6 +1979,44 @@ function capitalize(value) {
   return value[0].toUpperCase() + value.slice(1);
 }
 
+// src/utils/spinner.ts
+var Spinner = class {
+  message = "";
+  frame = 0;
+  interval = null;
+  writer;
+  isTTY;
+  frames = ["\u280B", "\u2819", "\u2839", "\u2838", "\u283C", "\u2834", "\u2826", "\u2827", "\u2807", "\u280F"];
+  constructor(writer, isTTY = false) {
+    this.writer = writer;
+    this.isTTY = isTTY;
+  }
+  start(message) {
+    if (!this.isTTY) {
+      return;
+    }
+    this.message = message;
+    this.frame = 0;
+    this.interval = setInterval(() => {
+      const spinner = this.frames[this.frame % this.frames.length];
+      this.writer.write(`\r${spinner} ${this.message}`);
+      this.frame += 1;
+    }, 80);
+  }
+  stop() {
+    if (this.interval) {
+      clearInterval(this.interval);
+      this.interval = null;
+    }
+    if (this.isTTY) {
+      this.writer.write("\r\x1B[K");
+    }
+  }
+};
+function createSpinner(writer, isTTY = false) {
+  return new Spinner(writer, isTTY);
+}
+
 // src/cli.ts
 async function runCli(argv, io = { stdout: process.stdout, stderr: process.stderr, stdin: process.stdin }, dependencies = { createOptimizer, readStdin, getCliInfo }) {
   const [command, ...rest] = argv;
@@ -2004,7 +2079,9 @@ async function runCli(argv, io = { stdout: process.stdout, stderr: process.stder
     io.stderr.write("A prompt is required.\n");
     return 1;
   }
+  const spinner = createSpinner(io.stderr, io.stderr.isTTY ?? false);
   try {
+    spinner.start("optimizing");
     const result = await optimizer.optimize({
       prompt: parsed.prompt,
       task: parsed.task,
@@ -2033,11 +2110,26 @@ async function runCli(argv, io = { stdout: process.stdout, stderr: process.stder
       timeoutMs: parsed.timeoutMs,
       bypassOptimization: parsed.bypassOptimization
     });
+    spinner.stop();
     if (parsed.json) {
       io.stdout.write(`${toPrettyJson(result)}
 `);
       return 0;
     }
+    if (parsed.clipboard) {
+      const copied = copyToClipboard(result.finalPrompt);
+      if (copied) {
+        io.stderr.write(`\u2713 Copied optimized prompt to clipboard
+`);
+        return 0;
+      } else {
+        io.stderr.write(`\u2717 Failed to copy to clipboard. Install xclip, xsel, or wl-copy.
+`);
+        io.stdout.write(`${result.finalPrompt}
+`);
+        return 1;
+      }
+    }
     if (parsed.plain) {
       io.stdout.write(`${result.finalPrompt}
 `);
@@ -2047,6 +2139,8 @@ async function runCli(argv, io = { stdout: process.stdout, stderr: process.stder
 
 `);
     io.stdout.write(`provider=${result.provider} model=${result.model} tokens=${result.estimatedTokensAfter.total} savings=${result.tokenSavings}
+`);
+    io.stdout.write(`  prompt_savings=${result.promptCompressionSavings} context_savings=${result.contextCompressionSavings} wrapper_overhead=${result.wrapperOverhead}
 `);
     if (result.selectedTarget) {
       io.stdout.write(`selected_target=${formatTarget(result.selectedTarget)}
@@ -2058,6 +2152,7 @@ async function runCli(argv, io = { stdout: process.stdout, stderr: process.stder
     }
     return 0;
   } catch (error) {
+    spinner.stop();
     const message = error instanceof Error ? error.message : "Unknown CLI error.";
     io.stderr.write(`${message}
 `);
@@ -2069,6 +2164,7 @@ function parseOptimizeArgs(args) {
     plain: false,
     json: false,
     debug: false,
+    clipboard: false,
     clearSession: false,
     useContext: true,
     bypassOptimization: false,
@@ -2152,6 +2248,9 @@ function parseOptimizeArgs(args) {
       case "--json":
         parsed.json = true;
         break;
+      case "--clipboard":
+        parsed.clipboard = true;
+        break;
       case "--debug":
         parsed.debug = true;
         break;
@@ -2218,6 +2317,7 @@ function getHelpText() {
     "  --sqlite-path <path>",
     "  --plain",
     "  --json",
+    "  --clipboard       Copy optimized prompt to clipboard",
     "  --debug",
     "  --save-context",
     "  --no-context",
@@ -2287,6 +2387,23 @@ function readPackageVersion() {
     return "dev";
   }
 }
+function copyToClipboard(text) {
+  const commands = [
+    { cmd: "xclip", args: ["-selection", "clipboard"], platform: "linux" },
+    { cmd: "xsel", args: ["-b"], platform: "linux" },
+    { cmd: "wl-copy", args: [], platform: "linux" },
+    { cmd: "pbcopy", args: [], platform: "darwin" }
+  ];
+  for (const { cmd } of commands) {
+    try {
+      execSync(`which ${cmd} > /dev/null 2>&1`);
+      execSync(cmd, { input: text, stdio: ["pipe", "ignore", "ignore"] });
+      return true;
+    } catch {
+    }
+  }
+  return false;
+}
 if (isMainModule()) {
   runCli(process.argv.slice(2)).then(
     (code) => {